Fix less-then-frame off by one slice, add db write toggle and disable

It turns out (i guess not so shockingly?) that `marketstore` doesn't
always teardown "gracefully" under SIGINT (seems to hang if there are
open client connections which are also in the midst of teardown?) so
this instead first tries the SIGINT and then fails over to a SIGKILL
(destroy loop) which seems to be much more reliable to ensure shutdown
without any downside - in terms of a "hard kill".

Originally i was thinking the issue was root perms related (which get
relegated solely to the `marketstored` daemon actor after spawn) but
actually it was indeed the signalling / application layer causing the
hold-up/latency on teardown. There's a bunch of lingering (now
commented) code which tried to solve this non-problem as well as a bunch
logging/prints to help decipher the root of the issue - this will all
get cleaned out shortly.

piker/_daemon.py

@@ -19,7 +19,7 @@ Structured, daemon tree service management.
 
 """
 from typing import Optional, Union, Callable, Any
-from contextlib import asynccontextmanager
+from contextlib import asynccontextmanager as acm
 from collections import defaultdict
 
 from pydantic import BaseModel
@@ -130,7 +130,7 @@ class Services(BaseModel):
 _services: Optional[Services] = None
 
 
-@asynccontextmanager
+@acm
 async def open_pikerd(
     start_method: str = 'trio',
     loglevel: Optional[str] = None,
@@ -185,7 +185,7 @@ async def open_pikerd(
             yield _services
 
 
-@asynccontextmanager
+@acm
 async def open_piker_runtime(
     name: str,
     enable_modules: list[str] = [],
@@ -226,7 +226,7 @@ async def open_piker_runtime(
         yield tractor.current_actor()
 
 
-@asynccontextmanager
+@acm
 async def maybe_open_runtime(
     loglevel: Optional[str] = None,
     **kwargs,
@@ -249,7 +249,7 @@ async def maybe_open_runtime(
         yield
 
 
-@asynccontextmanager
+@acm
 async def maybe_open_pikerd(
     loglevel: Optional[str] = None,
     **kwargs,
@@ -300,7 +300,36 @@ class Brokerd:
     locks = defaultdict(trio.Lock)
 
 
-@asynccontextmanager
+@acm
+async def find_service(
+    service_name: str,
+) -> Optional[tractor.Portal]:
+
+    log.info(f'Scanning for service `{service_name}`')
+    # attach to existing daemon by name if possible
+    async with tractor.find_actor(
+        service_name,
+        arbiter_sockaddr=_registry_addr,
+    ) as maybe_portal:
+        yield maybe_portal
+
+
+async def check_for_service(
+    service_name: str,
+
+) -> bool:
+    '''
+    Service daemon "liveness" predicate.
+
+    '''
+    async with tractor.query_actor(
+        service_name,
+        arbiter_sockaddr=_registry_addr,
+    ) as sockaddr:
+        return sockaddr
+
+
+@acm
 async def maybe_spawn_daemon(
 
     service_name: str,
@@ -310,7 +339,7 @@ async def maybe_spawn_daemon(
     **kwargs,
 
 ) -> tractor.Portal:
-    """
+    '''
     If no ``service_name`` daemon-actor can be found,
     spawn one in a local subactor and return a portal to it.
 
@@ -321,7 +350,7 @@ async def maybe_spawn_daemon(
     This can be seen as a service starting api for remote-actor
     clients.
 
-    """
+    '''
     if loglevel:
         get_console_log(loglevel)
 
@@ -330,19 +359,13 @@ async def maybe_spawn_daemon(
     lock = Brokerd.locks[service_name]
     await lock.acquire()
 
-    log.info(f'Scanning for existing {service_name}')
-    # attach to existing daemon by name if possible
-    async with tractor.find_actor(
-        service_name,
-        arbiter_sockaddr=_registry_addr,
-
-    ) as portal:
+    async with find_service(service_name) as portal:
         if portal is not None:
             lock.release()
             yield portal
             return
 
-        log.warning(f"Couldn't find any existing {service_name}")
+    log.warning(f"Couldn't find any existing {service_name}")
 
     # ask root ``pikerd`` daemon to spawn the daemon we need if
     # pikerd is not live we now become the root of the
@@ -423,7 +446,7 @@ async def spawn_brokerd(
     return True
 
 
-@asynccontextmanager
+@acm
 async def maybe_spawn_brokerd(
 
     brokername: str,
@@ -431,7 +454,9 @@ async def maybe_spawn_brokerd(
     **kwargs,
 
 ) -> tractor.Portal:
-    '''Helper to spawn a brokerd service.
+    '''
+    Helper to spawn a brokerd service *from* a client
+    who wishes to use the sub-actor-daemon.
 
     '''
     async with maybe_spawn_daemon(
@@ -483,7 +508,7 @@ async def spawn_emsd(
     return True
 
 
-@asynccontextmanager
+@acm
 async def maybe_open_emsd(
 
     brokername: str,

piker/brokers/_util.py

@@ -33,7 +33,41 @@ class SymbolNotFound(BrokerError):
 
 
 class NoData(BrokerError):
-    "Symbol data not permitted"
+    '''
+    Symbol data not permitted or no data
+    for time range found.
+
+    '''
+    def __init__(
+        self,
+        *args,
+        frame_size: int = 1000,
+
+    ) -> None:
+        super().__init__(*args)
+
+        # when raised, machinery can check if the backend
+        # set a "frame size" for doing datetime calcs.
+        self.frame_size: int = 1000
+
+
+class DataUnavailable(BrokerError):
+    '''
+    Signal storage requests to terminate.
+
+    '''
+    # TODO: add in a reason that can be displayed in the
+    # UI (for eg. `kraken` is bs and you should complain
+    # to them that you can't pull more OHLC data..)
+
+
+class DataThrottle(BrokerError):
+    '''
+    Broker throttled request rate for data.
+
+    '''
+    # TODO: add in throttle metrics/feedback
+
 
 
 def resproc(
@@ -50,12 +84,12 @@ def resproc(
     if not resp.status_code == 200:
         raise BrokerError(resp.body)
     try:
-        json = resp.json()
+        msg = resp.json()
     except json.decoder.JSONDecodeError:
         log.exception(f"Failed to process {resp}:\n{resp.text}")
         raise BrokerError(resp.text)
 
     if log_resp:
-        log.debug(f"Received json contents:\n{colorize_json(json)}")
+        log.debug(f"Received json contents:\n{colorize_json(msg)}")
 
-    return json if return_json else resp
+    return msg if return_json else resp

piker/brokers/binance.py

@@ -19,6 +19,7 @@ Binance backend
 
 """
 from contextlib import asynccontextmanager as acm
+from datetime import datetime
 from typing import (
     Any, Union, Optional,
     AsyncGenerator, Callable,
@@ -221,20 +222,22 @@ class Client:
     async def bars(
         self,
         symbol: str,
-        start_time: int = None,
-        end_time: int = None,
+        start_dt: Optional[datetime] = None,
+        end_dt: Optional[datetime] = None,
         limit: int = 1000,  # <- max allowed per query
         as_np: bool = True,
 
     ) -> dict:
 
-        if start_time is None:
-            start_time = binance_timestamp(
-                pendulum.now('UTC').start_of('minute').subtract(minutes=limit)
-            )
+        if end_dt is None:
+            end_dt = pendulum.now('UTC')
 
-        if end_time is None:
-            end_time = binance_timestamp(pendulum.now('UTC'))
+        if start_dt is None:
+            start_dt = end_dt.start_of(
+                'minute').subtract(minutes=limit)
+
+        start_time = binance_timestamp(start_dt)
+        end_time = binance_timestamp(end_dt)
 
         # https://binance-docs.github.io/apidocs/spot/en/#kline-candlestick-data
         bars = await self._api(
@@ -379,7 +382,27 @@ async def open_history_client(
 
     # TODO implement history getter for the new storage layer.
     async with open_cached_client('binance') as client:
-        yield client
+
+        async def get_ohlc(
+            end_dt: Optional[datetime] = None,
+            start_dt: Optional[datetime] = None,
+
+        ) -> tuple[
+            np.ndarray,
+            datetime,  # start
+            datetime,  # end
+        ]:
+
+            array = await client.bars(
+                symbol,
+                start_dt=start_dt,
+                end_dt=end_dt,
+            )
+            start_dt = pendulum.from_timestamp(array[0]['time'])
+            end_dt = pendulum.from_timestamp(array[-1]['time'])
+            return array, start_dt, end_dt
+
+        yield get_ohlc, {'erlangs': 4, 'rate': 4}
 
 
 async def backfill_bars(

piker/brokers/ib.py

@@ -57,6 +57,8 @@ from ib_insync.wrapper import Wrapper
 from ib_insync.client import Client as ib_Client
 from fuzzywuzzy import process as fuzzy
 import numpy as np
+import pendulum
+
 
 from .. import config
 from ..log import get_logger, get_console_log
@@ -295,6 +297,10 @@ class Client:
         global _enters
         # log.info(f'REQUESTING BARS {_enters} @ end={end_dt}')
         print(f'REQUESTING BARS {_enters} @ end={end_dt}')
+
+        if not end_dt:
+            end_dt = ''
+
         _enters += 1
 
         contract = await self.find_contract(fqsn)
@@ -1438,8 +1444,6 @@ async def get_bars(
     a ``MethoProxy``.
 
     '''
-    import pendulum
-
     fails = 0
     bars: Optional[list] = None
     first_dt: datetime = None
@@ -1467,7 +1471,9 @@ async def get_bars(
             time = bars_array['time']
             assert time[-1] == last_dt.timestamp()
             assert time[0] == first_dt.timestamp()
-            log.info(f'bars retreived for dts {first_dt}:{last_dt}')
+            log.info(
+                f'{len(bars)} bars retreived for {first_dt} -> {last_dt}'
+            )
 
             return (bars, bars_array, first_dt, last_dt), fails
 
@@ -1478,21 +1484,30 @@ async def get_bars(
 
             if 'No market data permissions for' in msg:
                 # TODO: signalling for no permissions searches
-                raise NoData(f'Symbol: {fqsn}')
-                break
+                raise NoData(
+                    f'Symbol: {fqsn}',
+                )
 
             elif (
                 err.code == 162
                 and 'HMDS query returned no data' in err.message
             ):
-                # try to decrement start point and look further back
-                end_dt = last_dt = last_dt.subtract(seconds=2000)
+                # XXX: this is now done in the storage mgmt layer
+                # and we shouldn't implicitly decrement the frame dt
+                # index since the upper layer may be doing so
+                # concurrently and we don't want to be delivering frames
+                # that weren't asked for.
                 log.warning(
-                    f'No data found ending @ {end_dt}\n'
-                    f'Starting another request for {end_dt}'
+                    f'NO DATA found ending @ {end_dt}\n'
                 )
 
-                continue
+                # try to decrement start point and look further back
+                # end_dt = last_dt = last_dt.subtract(seconds=2000)
+
+                raise NoData(
+                    f'Symbol: {fqsn}',
+                    frame_size=2000,
+                )
 
             elif _pacing in msg:
 
@@ -1546,8 +1561,8 @@ async def open_history_client(
     async with open_client_proxy() as proxy:
 
         async def get_hist(
-            end_dt: str,
-            start_dt: str = '',
+            end_dt: Optional[datetime] = None,
+            start_dt: Optional[datetime] = None,
 
         ) -> tuple[np.ndarray, str]:
 
@@ -1555,10 +1570,13 @@ async def open_history_client(
 
             # TODO: add logic here to handle tradable hours and only grab
             # valid bars in the range
-            if out == (None, None):
+            if out is None:
                 # could be trying to retreive bars over weekend
                 log.error(f"Can't grab bars starting at {end_dt}!?!?")
-                raise NoData(f'{end_dt}')
+                raise NoData(
+                    f'{end_dt}',
+                    frame_size=2000,
+                )
 
             bars, bars_array, first_dt, last_dt = out
 
@@ -1569,7 +1587,12 @@ async def open_history_client(
 
             return bars_array, first_dt, last_dt
 
-        yield get_hist
+        # TODO: it seems like we can do async queries for ohlc
+        # but getting the order right still isn't working and I'm not
+        # quite sure why.. needs some tinkering and probably
+        # a lookthrough of the ``ib_insync`` machinery, for eg. maybe
+        # we have to do the batch queries on the `asyncio` side?
+        yield get_hist, {'erlangs': 1, 'rate': 6}
 
 
 async def backfill_bars(
@@ -1831,6 +1854,7 @@ async def stream_quotes(
         symbol=sym,
     )
     first_quote = normalize(first_ticker)
+    # print(f'first quote: {first_quote}')
 
     def mk_init_msgs() -> dict[str, dict]:
         # pass back some symbol info like min_tick, trading_hours, etc.
@@ -2435,8 +2459,8 @@ async def data_reset_hack(
     try:
         import i3ipc
     except ImportError:
-        return False
         log.warning('IB data hack no-supported on ur platformz')
+        return False
 
     i3 = i3ipc.Connection()
     t = i3.get_tree()

piker/brokers/kraken.py

@@ -20,7 +20,8 @@ Kraken backend.
 '''
 from contextlib import asynccontextmanager as acm
 from dataclasses import asdict, field
-from typing import Any, Optional, AsyncIterator, Callable
+from datetime import datetime
+from typing import Any, Optional, AsyncIterator, Callable, Union
 import time
 
 from trio_typing import TaskStatus
@@ -40,7 +41,13 @@ import base64
 
 from .. import config
 from .._cacheables import open_cached_client
-from ._util import resproc, SymbolNotFound, BrokerError
+from ._util import (
+    resproc,
+    SymbolNotFound,
+    BrokerError,
+    DataThrottle,
+    DataUnavailable,
+)
 from ..log import get_logger, get_console_log
 from ..data import ShmArray
 from ..data._web_bs import open_autorecon_ws, NoBsWs
@@ -305,7 +312,7 @@ class Client:
         action: str,
         size: float,
         reqid: str = None,
-        validate: bool = False # set True test call without a real submission
+        validate: bool = False  # set True test call without a real submission
     ) -> dict:
         '''
         Place an order and return integer request id provided by client.
@@ -391,17 +398,26 @@ class Client:
     async def bars(
         self,
         symbol: str = 'XBTUSD',
+
         # UTC 2017-07-02 12:53:20
-        since: int = None,
+        since: Optional[Union[int, datetime]] = None,
         count: int = 720,  # <- max allowed per query
         as_np: bool = True,
+
     ) -> dict:
+
         if since is None:
             since = pendulum.now('UTC').start_of('minute').subtract(
                 minutes=count).timestamp()
 
+        elif isinstance(since, int):
+            since = pendulum.from_timestamp(since).timestamp()
+
+        else:  # presumably a pendulum datetime
+            since = since.timestamp()
+
         # UTC 2017-07-02 12:53:20 is oldest seconds value
-        since = str(max(1499000000, since))
+        since = str(max(1499000000, int(since)))
         json = await self._public(
             'OHLC',
             data={
@@ -445,7 +461,16 @@ class Client:
             array = np.array(new_bars, dtype=_ohlc_dtype) if as_np else bars
             return array
         except KeyError:
-            raise SymbolNotFound(json['error'][0] + f': {symbol}')
+            errmsg = json['error'][0]
+
+            if 'not found' in errmsg:
+                raise SymbolNotFound(errmsg + f': {symbol}')
+
+            elif 'Too many requests' in errmsg:
+                raise DataThrottle(f'{symbol}')
+
+            else:
+                raise BrokerError(errmsg)
 
 
 @acm
@@ -668,8 +693,8 @@ async def handle_order_requests(
                                 oid=msg.oid,
                                 reqid=msg.reqid,
                                 symbol=msg.symbol,
-                                # TODO: maybe figure out if pending cancels will
-                                # eventually get cancelled
+                                # TODO: maybe figure out if pending
+                                # cancels will eventually get cancelled
                                 reason="Order cancel is still pending?",
                                 broker_details=resp
                             ).dict()
@@ -1003,7 +1028,45 @@ async def open_history_client(
 
     # TODO implement history getter for the new storage layer.
     async with open_cached_client('kraken') as client:
-        yield client
+
+        # lol, kraken won't send any more then the "last"
+        # 720 1m bars.. so we have to just ignore further
+        # requests of this type..
+        queries: int = 0
+
+        async def get_ohlc(
+            end_dt: Optional[datetime] = None,
+            start_dt: Optional[datetime] = None,
+
+        ) -> tuple[
+            np.ndarray,
+            datetime,  # start
+            datetime,  # end
+        ]:
+
+            nonlocal queries
+            if queries > 0:
+                raise DataUnavailable
+
+            count = 0
+            while count <= 3:
+                try:
+                    array = await client.bars(
+                        symbol,
+                        since=end_dt,
+                    )
+                    count += 1
+                    queries += 1
+                    break
+                except DataThrottle:
+                    log.warning(f'kraken OHLC throttle for {symbol}')
+                    await trio.sleep(1)
+
+            start_dt = pendulum.from_timestamp(array[0]['time'])
+            end_dt = pendulum.from_timestamp(array[-1]['time'])
+            return array, start_dt, end_dt
+
+        yield get_ohlc, {'erlangs': 1, 'rate': 1}
 
 
 async def backfill_bars(

piker/cli/__init__.py

@@ -16,29 +16,22 @@ from .. import config
 log = get_logger('cli')
 DEFAULT_BROKER = 'questrade'
 
-_config_dir = click.get_app_dir('piker')
-_watchlists_data_path = os.path.join(_config_dir, 'watchlists.json')
-_context_defaults = dict(
-    default_map={
-        # Questrade specific quote poll rates
-        'monitor': {
-            'rate': 3,
-        },
-        'optschain': {
-            'rate': 1,
-        },
-    }
-)
-
 
 @click.command()
 @click.option('--loglevel', '-l', default='warning', help='Logging level')
 @click.option('--tl', is_flag=True, help='Enable tractor logging')
 @click.option('--pdb', is_flag=True, help='Enable tractor debug mode')
 @click.option('--host', '-h', default='127.0.0.1', help='Host address to bind')
-def pikerd(loglevel, host, tl, pdb):
-    """Spawn the piker broker-daemon.
-    """
+@click.option(
+    '--tsdb',
+    is_flag=True,
+    help='Enable local ``marketstore`` instance'
+)
+def pikerd(loglevel, host, tl, pdb, tsdb):
+    '''
+    Spawn the piker broker-daemon.
+
+    '''
     from .._daemon import open_pikerd
     log = get_console_log(loglevel)
 
@@ -52,13 +45,33 @@ def pikerd(loglevel, host, tl, pdb):
         ))
 
     async def main():
-        async with open_pikerd(loglevel=loglevel, debug_mode=pdb):
+
+        async with (
+            open_pikerd(
+                loglevel=loglevel,
+                debug_mode=pdb,
+            ),  # normally delivers a ``Services`` handle
+            trio.open_nursery() as n,
+        ):
+            if tsdb:
+                # TODO:
+                # async with maybe_open_marketstored():
+
+                from piker.data._ahab import start_ahab
+                log.info('Spawning `marketstore` supervisor')
+                ctn_ready = await n.start(
+                    start_ahab,
+                    'marketstored',
+                )
+                await ctn_ready.wait()
+                log.info('`marketstore` container:{uid} up')
+
             await trio.sleep_forever()
 
     trio.run(main)
 
 
-@click.group(context_settings=_context_defaults)
+@click.group(context_settings=config._context_defaults)
 @click.option(
     '--brokers', '-b',
     default=[DEFAULT_BROKER],
@@ -87,8 +100,8 @@ def cli(ctx, brokers, loglevel, tl, configdir):
         'loglevel': loglevel,
         'tractorloglevel': None,
         'log': get_console_log(loglevel),
-        'confdir': _config_dir,
-        'wl_path': _watchlists_data_path,
+        'confdir': config._config_dir,
+        'wl_path': config._watchlists_data_path,
     })
 
     # allow enabling same loglevel in ``tractor`` machinery

piker/config.py

@@ -17,6 +17,8 @@
 """
 Broker configuration mgmt.
 """
+import platform
+import sys
 import os
 from os.path import dirname
 import shutil
@@ -24,14 +26,100 @@ from typing import Optional
 
 from bidict import bidict
 import toml
-import click
 
 from .log import get_logger
 
 log = get_logger('broker-config')
 
-_config_dir = click.get_app_dir('piker')
+
+# taken from ``click`` since apparently they have some
+# super weirdness with sigint and sudo..no clue
+def get_app_dir(app_name, roaming=True, force_posix=False):
+    r"""Returns the config folder for the application.  The default behavior
+    is to return whatever is most appropriate for the operating system.
+
+    To give you an idea, for an app called ``"Foo Bar"``, something like
+    the following folders could be returned:
+
+    Mac OS X:
+      ``~/Library/Application Support/Foo Bar``
+    Mac OS X (POSIX):
+      ``~/.foo-bar``
+    Unix:
+      ``~/.config/foo-bar``
+    Unix (POSIX):
+      ``~/.foo-bar``
+    Win XP (roaming):
+      ``C:\Documents and Settings\<user>\Local Settings\Application Data\Foo Bar``
+    Win XP (not roaming):
+      ``C:\Documents and Settings\<user>\Application Data\Foo Bar``
+    Win 7 (roaming):
+      ``C:\Users\<user>\AppData\Roaming\Foo Bar``
+    Win 7 (not roaming):
+      ``C:\Users\<user>\AppData\Local\Foo Bar``
+
+    .. versionadded:: 2.0
+
+    :param app_name: the application name.  This should be properly capitalized
+                     and can contain whitespace.
+    :param roaming: controls if the folder should be roaming or not on Windows.
+                    Has no affect otherwise.
+    :param force_posix: if this is set to `True` then on any POSIX system the
+                        folder will be stored in the home folder with a leading
+                        dot instead of the XDG config home or darwin's
+                        application support folder.
+    """
+
+    def _posixify(name):
+        return "-".join(name.split()).lower()
+
+    # if WIN:
+    if platform.system() == 'Windows':
+        key = "APPDATA" if roaming else "LOCALAPPDATA"
+        folder = os.environ.get(key)
+        if folder is None:
+            folder = os.path.expanduser("~")
+        return os.path.join(folder, app_name)
+    if force_posix:
+        return os.path.join(os.path.expanduser("~/.{}".format(_posixify(app_name))))
+    if sys.platform == "darwin":
+        return os.path.join(
+            os.path.expanduser("~/Library/Application Support"), app_name
+        )
+    return os.path.join(
+        os.environ.get("XDG_CONFIG_HOME", os.path.expanduser("~/.config")),
+        _posixify(app_name),
+    )
+
+
+_config_dir = _click_config_dir = get_app_dir('piker')
+_parent_user = os.environ.get('SUDO_USER')
+
+if _parent_user:
+    non_root_user_dir = os.path.expanduser(
+        f'~{_parent_user}'
+    )
+    root = 'root'
+    _config_dir = (
+        non_root_user_dir +
+        _click_config_dir[
+            _click_config_dir.rfind(root) + len(root):
+        ]
+    )
+
 _file_name = 'brokers.toml'
+_watchlists_data_path = os.path.join(_config_dir, 'watchlists.json')
+_context_defaults = dict(
+    default_map={
+        # Questrade specific quote poll rates
+        'monitor': {
+            'rate': 3,
+        },
+        'optschain': {
+            'rate': 1,
+        },
+    }
+)
 
 
 def _override_config_dir(

piker/data/_ahab.py

@@ -0,0 +1,469 @@
+# piker: trading gear for hackers
+# Copyright (C) 2018-present  Tyler Goodlet (in stewardship of piker0)
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+'''
+Supervisor for docker with included specific-image service helpers.
+
+'''
+import os
+from typing import (
+    Optional,
+    # Any,
+)
+from contextlib import asynccontextmanager as acm
+
+import trio
+from trio_typing import TaskStatus
+import tractor
+import docker
+import json
+from docker.models.containers import Container as DockerContainer
+from docker.errors import DockerException, APIError
+from requests.exceptions import ConnectionError, ReadTimeout
+
+from ..log import get_logger, get_console_log
+from .. import config
+
+log = get_logger(__name__)
+
+
+_config = '''
+# piker's ``marketstore`` config.
+
+# mount this config using:
+# sudo docker run --mount \
+# type=bind,source="$HOME/.config/piker/",target="/etc" -i -p \
+# 5993:5993 alpacamarkets/marketstore:latest
+
+root_directory: data
+listen_port: 5993
+grpc_listen_port: 5995
+log_level: debug
+queryable: true
+stop_grace_period: 0
+wal_rotate_interval: 5
+stale_threshold: 5
+enable_add: true
+enable_remove: false
+
+triggers:
+  - module: ondiskagg.so
+    on: "*/1Sec/OHLCV"
+    config:
+        # filter: "nasdaq"
+        destinations:
+            - 1Min
+            - 5Min
+            - 15Min
+            - 1H
+            - 1D
+
+  - module: stream.so
+    on: '*/*/*'
+    # config:
+    #     filter: "nasdaq"
+
+'''
+
+
+class DockerNotStarted(Exception):
+    'Prolly you dint start da daemon bruh'
+
+
+@acm
+async def open_docker(
+    url: Optional[str] = None,
+    **kwargs,
+
+) -> docker.DockerClient:
+
+    client: Optional[docker.DockerClient] = None
+    try:
+        client = docker.DockerClient(
+            base_url=url,
+            **kwargs
+        ) if url else docker.from_env(**kwargs)
+
+        yield client
+
+    except (
+        DockerException,
+        APIError,
+    ) as err:
+
+        def unpack_msg(err: Exception) -> str:
+            args = getattr(err, 'args', None)
+            if args:
+                return args
+            else:
+                return str(err)
+
+        # could be more specific so let's check if it's just perms.
+        if err.args:
+            errs = err.args
+            for err in errs:
+                msg = unpack_msg(err)
+                if 'PermissionError' in msg:
+                    raise DockerException('You dint run as root yo!')
+
+                elif 'FileNotFoundError' in msg:
+                    raise DockerNotStarted('Did you start da service sister?')
+
+        # not perms?
+        raise
+
+    finally:
+        if client:
+            client.close()
+            # client.api._custom_adapter.close()
+            for c in client.containers.list():
+                c.kill()
+
+
+class Container:
+    '''
+    Wrapper around a ``docker.models.containers.Container`` to include
+    log capture and relay through our native logging system and helper
+    method(s) for cancellation/teardown.
+
+    '''
+    def __init__(
+        self,
+        cntr: DockerContainer,
+    ) -> None:
+
+        self.cntr = cntr
+        # log msg de-duplication
+        self.seen_so_far = set()
+
+    async def process_logs_until(
+        self,
+        patt: str,
+        bp_on_msg: bool = False,
+    ) -> bool:
+        '''
+        Attempt to capture container log messages and relay through our
+        native logging system.
+
+        '''
+        seen_so_far = self.seen_so_far
+
+        while True:
+            logs = self.cntr.logs()
+            entries = logs.decode().split('\n')
+            for entry in entries:
+
+                # ignore null lines
+                if not entry:
+                    continue
+
+                try:
+                    record = json.loads(entry.strip())
+                except json.JSONDecodeError:
+                    if 'Error' in entry:
+                        raise RuntimeError(entry)
+                    raise
+
+                msg = record['msg']
+                level = record['level']
+                if msg and entry not in seen_so_far:
+                    seen_so_far.add(entry)
+                    if bp_on_msg:
+                        await tractor.breakpoint()
+
+                    getattr(log, level, log.error)(f'{msg}')
+
+                if patt in msg:
+                    return True
+
+                # do a checkpoint so we don't block if cancelled B)
+                await trio.sleep(0.01)
+
+        return False
+
+    def try_signal(
+        self,
+        signal: str = 'SIGINT',
+
+    ) -> bool:
+        try:
+            # XXX: market store doesn't seem to shutdown nicely all the
+            # time with this (maybe because there are still open grpc
+            # connections?) noticably after client connections have been
+            # made or are in use/teardown. It works just fine if you
+            # just start and stop the container tho?..
+            log.cancel(f'SENDING {signal} to {self.cntr.id}')
+            self.cntr.kill(signal)
+            return True
+
+        except docker.errors.APIError as err:
+            # _err = err
+            if 'is not running' in err.explanation:
+                return False
+
+    async def cancel(
+        self,
+    ) -> None:
+
+        cid = self.cntr.id
+        self.try_signal('SIGINT')
+
+        with trio.move_on_after(0.5) as cs:
+            cs.shield = True
+            # print('PROCESSINGN LOGS')
+            await self.process_logs_until('initiating graceful shutdown')
+            # print('SHUTDOWN REPORTED BY CONTAINER')
+            await self.process_logs_until('exiting...',)
+
+        for _ in range(10):
+            with trio.move_on_after(0.5) as cs:
+                cs.shield = True
+                # print('waiting on EXITING')
+                await self.process_logs_until('exiting...',)
+                # print('got EXITING')
+                break
+
+            if cs.cancelled_caught:
+                # get out the big guns, bc apparently marketstore
+                # doesn't actually know how to terminate gracefully
+                # :eyeroll:...
+                self.try_signal('SIGKILL')
+
+                try:
+                    log.info('Waiting on container shutdown: {cid}')
+                    self.cntr.wait(
+                        timeout=0.1,
+                        condition='not-running',
+                    )
+                    break
+
+                except (
+                    ReadTimeout,
+                    ConnectionError,
+                ):
+                    log.error(f'failed to wait on container {cid}')
+                    raise
+
+        else:
+            raise RuntimeError('Failed to cancel container {cid}')
+
+        log.cancel(f'Container stopped: {cid}')
+
+
+@tractor.context
+async def open_marketstored(
+    ctx: tractor.Context,
+    **kwargs,
+
+) -> None:
+    '''
+    Start and supervise a marketstore instance with its config bind-mounted
+    in from the piker config directory on the system.
+
+    The equivalent cli cmd to this code is:
+
+        sudo docker run --mount \
+        type=bind,source="$HOME/.config/piker/",target="/etc" -i -p \
+        5993:5993 alpacamarkets/marketstore:latest
+
+    '''
+    log = get_console_log('info', name=__name__)
+
+    async with open_docker() as client:
+
+        # create a mount from user's local piker config dir into container
+        config_dir_mnt = docker.types.Mount(
+            target='/etc',
+            source=config._config_dir,
+            type='bind',
+        )
+
+        # create a user config subdir where the marketstore
+        # backing filesystem database can be persisted.
+        persistent_data_dir = os.path.join(
+            config._config_dir, 'data',
+        )
+        if not os.path.isdir(persistent_data_dir):
+            os.mkdir(persistent_data_dir)
+
+        data_dir_mnt = docker.types.Mount(
+            target='/data',
+            source=persistent_data_dir,
+            type='bind',
+        )
+
+        dcntr: DockerContainer = client.containers.run(
+            'alpacamarkets/marketstore:latest',
+            # do we need this for cmds?
+            # '-i',
+
+            # '-p 5993:5993',
+            ports={
+                '5993/tcp': 5993,  # jsonrpc
+                '5995/tcp': 5995,  # grpc
+            },
+            mounts=[config_dir_mnt, data_dir_mnt],
+            detach=True,
+            # stop_signal='SIGINT',
+            init=True,
+            # remove=True,
+        )
+        cntr = Container(dcntr)
+
+        with trio.move_on_after(1):
+            found = await cntr.process_logs_until(
+                "launching tcp listener for all services...",
+            )
+
+            if not found and cntr not in client.containers.list():
+                raise RuntimeError(
+                    'Failed to start `marketstore` check logs deats'
+                )
+
+        await ctx.started((cntr.cntr.id, os.getpid()))
+
+        # async with ctx.open_stream() as stream:
+
+        try:
+
+            # TODO: we might eventually want a proxy-style msg-prot here
+            # to allow remote control of containers without needing
+            # callers to have root perms?
+            await trio.sleep_forever()
+
+            # await cntr.cancel()
+            # with trio.CancelScope(shield=True):
+            #     # block for the expected "teardown log msg"..
+            #     # await cntr.process_logs_until('exiting...',)
+
+            #     # only msg should be to signal killing the
+            #     # container and this super daemon.
+            #     msg = await stream.receive()
+            #     # print("GOT CANCEL MSG")
+
+            #     cid = msg['cancel']
+            #     log.cancel(f'Cancelling container {cid}')
+
+            #     # print("CANCELLING CONTAINER")
+            #     await cntr.cancel()
+
+            #     # print("SENDING ACK")
+            #     await stream.send('ack')
+
+        except (
+            BaseException,
+            # trio.Cancelled,
+            # KeyboardInterrupt,
+        ):
+
+            with trio.CancelScope(shield=True):
+                await cntr.cancel()
+                # await stream.send('ack')
+
+            raise
+
+
+async def start_ahab(
+    service_name: str,
+    task_status: TaskStatus[trio.Event] = trio.TASK_STATUS_IGNORED,
+
+) -> None:
+    '''
+    Start a ``docker`` container supervisor with given service name.
+
+    Currently the actor calling this task should normally be started
+    with root permissions (until we decide to use something that doesn't
+    require this, like docker's rootless mode or some wrapper project) but
+    te root perms are de-escalated after the docker supervisor sub-actor
+    is started.
+
+    '''
+    cn_ready = trio.Event()
+    try:
+        async with tractor.open_nursery(
+            loglevel='runtime',
+        ) as tn:
+
+            portal = await tn.start_actor(
+                service_name,
+                enable_modules=[__name__]
+            )
+
+            # TODO: we have issues with this on teardown
+            # where ``tractor`` tries to issue ``os.kill()``
+            # and hits perms errors since the root process
+            # doesn't any longer have root perms..
+
+            # de-escalate root perms to the original user
+            # after the docker supervisor actor is spawned.
+            if config._parent_user:
+                import pwd
+                os.setuid(
+                    pwd.getpwnam(
+                        config._parent_user
+                    )[2]  # named user's uid
+                )
+
+            task_status.started(cn_ready)
+
+            async with portal.open_context(
+                open_marketstored,
+            ) as (ctx, first):
+
+                cid, pid = first
+
+                await trio.sleep_forever()
+                # async with ctx.open_stream() as stream:
+                #     try:
+                #         # run till cancelled
+                #         await trio.sleep_forever()
+                #     finally:
+                #         with trio.CancelScope(shield=True):
+                #             # print('SENDING CANCEL TO MARKETSTORED')
+                #             await stream.send({'cancel': (cid, pid)})
+                #         assert await stream.receive() == 'ack'
+
+    # since we demoted root perms in this parent
+    # we'll get a perms error on proc cleanup in
+    # ``tractor`` nursery exit. just make sure
+    # the child is terminated and don't raise the
+    # error if so.
+
+    # TODO: we could also consider adding
+    # a ``tractor.ZombieDetected`` or something that we could raise
+    # if we find the child didn't terminate.
+    # await tractor.breakpoint()
+    except PermissionError:
+        log.warning('Failed to cancel root permsed container')
+
+    except (
+        trio.MultiError,
+    ) as err:
+        for subexc in err.exceptions:
+            if isinstance(subexc, PermissionError):
+                log.warning('Failed to cancel root perms-ed container')
+                return
+        else:
+            raise
+
+
+async def main():
+    await start_ahab()
+    await trio.sleep_forever()
+
+
+if __name__ == '__main__':
+    trio.run(main)

piker/data/_sampling.py

@@ -22,14 +22,16 @@ financial data flows.
 from __future__ import annotations
 from collections import Counter
 import time
+from typing import TYPE_CHECKING, Optional
 
 import tractor
 import trio
 from trio_typing import TaskStatus
 
-from ._sharedmem import ShmArray
 from ..log import get_logger
 
+if TYPE_CHECKING:
+    from ._sharedmem import ShmArray
 
 log = get_logger(__name__)
 
@@ -88,6 +90,7 @@ async def increment_ohlc_buffer(
 
     total_s = 0  # total seconds counted
     lowest = min(sampler.ohlcv_shms.keys())
+    lowest_shm = sampler.ohlcv_shms[lowest][0]
     ad = lowest - 0.001
 
     with trio.CancelScope() as cs:
@@ -131,21 +134,46 @@ async def increment_ohlc_buffer(
                     # write to the buffer
                     shm.push(last)
 
-            # broadcast the buffer index step to any subscribers for
-            # a given sample period.
-            subs = sampler.subscribers.get(delay_s, ())
+            await broadcast(delay_s, shm=lowest_shm)
 
-            for stream in subs:
-                try:
-                    await stream.send({'index': shm._last.value})
-                except (
-                    trio.BrokenResourceError,
-                    trio.ClosedResourceError
-                ):
-                    log.error(
-                        f'{stream._ctx.chan.uid} dropped connection'
-                    )
-                    subs.remove(stream)
+
+async def broadcast(
+    delay_s: int,
+    shm: Optional[ShmArray] = None,
+
+) -> None:
+    # broadcast the buffer index step to any subscribers for
+    # a given sample period.
+    subs = sampler.subscribers.get(delay_s, ())
+
+    last = -1
+
+    if shm is None:
+        periods = sampler.ohlcv_shms.keys()
+        # if this is an update triggered by a history update there
+        # might not actually be any sampling bus setup since there's
+        # no "live feed" active yet.
+        if periods:
+            lowest = min(periods)
+            shm = sampler.ohlcv_shms[lowest][0]
+            last = shm._last.value
+
+    for stream in subs:
+        try:
+            await stream.send({'index': last})
+        except (
+            trio.BrokenResourceError,
+            trio.ClosedResourceError
+        ):
+            log.error(
+                f'{stream._ctx.chan.uid} dropped connection'
+            )
+            try:
+                subs.remove(stream)
+            except ValueError:
+                log.warning(
+                    f'{stream._ctx.chan.uid} sub already removed!?'
+                )
 
 
 @tractor.context
@@ -365,7 +393,12 @@ async def uniform_rate_send(
 
         if left_to_sleep > 0:
             with trio.move_on_after(left_to_sleep) as cs:
-                sym, last_quote = await quote_stream.receive()
+                try:
+                    sym, last_quote = await quote_stream.receive()
+                except trio.EndOfChannel:
+                    log.exception(f"feed for {stream} ended?")
+                    break
+
                 diff = time.time() - last_send
 
                 if not first_quote:

piker/data/_sharedmem.py

@@ -22,7 +22,6 @@ from __future__ import annotations
 from sys import byteorder
 from typing import Optional
 from multiprocessing.shared_memory import SharedMemory, _USE_POSIX
-from multiprocessing import resource_tracker as mantracker
 
 if _USE_POSIX:
     from _posixshmem import shm_unlink
@@ -30,6 +29,7 @@ if _USE_POSIX:
 import tractor
 import numpy as np
 from pydantic import BaseModel
+from numpy.lib import recfunctions as rfn
 
 from ..log import get_logger
 from ._source import base_iohlc_dtype
@@ -40,32 +40,39 @@ log = get_logger(__name__)
 
 # how  much is probably dependent on lifestyle
 _secs_in_day = int(60 * 60 * 24)
-# we try for 3 times but only on a run-every-other-day kinda week.
-_default_size = 10 * _secs_in_day
+# we try for a buncha times, but only on a run-every-other-day kinda week.
+_days_worth = 16
+_default_size = _days_worth * _secs_in_day
 # where to start the new data append index
-_rt_buffer_start = int(9*_secs_in_day)
+_rt_buffer_start = int((_days_worth - 1) * _secs_in_day)
 
 
-# Tell the "resource tracker" thing to fuck off.
-class ManTracker(mantracker.ResourceTracker):
-    def register(self, name, rtype):
-        pass
+def cuckoff_mantracker():
 
-    def unregister(self, name, rtype):
-        pass
+    from multiprocessing import resource_tracker as mantracker
 
-    def ensure_running(self):
-        pass
+    # Tell the "resource tracker" thing to fuck off.
+    class ManTracker(mantracker.ResourceTracker):
+        def register(self, name, rtype):
+            pass
+
+        def unregister(self, name, rtype):
+            pass
+
+        def ensure_running(self):
+            pass
+
+    # "know your land and know your prey"
+    # https://www.dailymotion.com/video/x6ozzco
+    mantracker._resource_tracker = ManTracker()
+    mantracker.register = mantracker._resource_tracker.register
+    mantracker.ensure_running = mantracker._resource_tracker.ensure_running
+    # ensure_running = mantracker._resource_tracker.ensure_running
+    mantracker.unregister = mantracker._resource_tracker.unregister
+    mantracker.getfd = mantracker._resource_tracker.getfd
 
 
-# "know your land and know your prey"
-# https://www.dailymotion.com/video/x6ozzco
-mantracker._resource_tracker = ManTracker()
-mantracker.register = mantracker._resource_tracker.register
-mantracker.ensure_running = mantracker._resource_tracker.ensure_running
-ensure_running = mantracker._resource_tracker.ensure_running
-mantracker.unregister = mantracker._resource_tracker.unregister
-mantracker.getfd = mantracker._resource_tracker.getfd
+cuckoff_mantracker()
 
 
 class SharedInt:
@@ -191,7 +198,11 @@ class ShmArray:
         self._post_init: bool = False
 
         # pushing data does not write the index (aka primary key)
-        self._write_fields = list(shmarr.dtype.fields.keys())[1:]
+        dtype = shmarr.dtype
+        if dtype.fields:
+            self._write_fields = list(shmarr.dtype.fields.keys())[1:]
+        else:
+            self._write_fields = None
 
     # TODO: ringbuf api?
 
@@ -237,6 +248,48 @@ class ShmArray:
 
         return a
 
+    def ustruct(
+        self,
+        fields: Optional[list[str]] = None,
+
+        # type that all field values will be cast to
+        # in the returned view.
+        common_dtype: np.dtype = np.float,
+
+    ) -> np.ndarray:
+
+        array = self._array
+
+        if fields:
+            selection = array[fields]
+            # fcount = len(fields)
+        else:
+            selection = array
+            # fcount = len(array.dtype.fields)
+
+        # XXX: manual ``.view()`` attempt that also doesn't work.
+        # uview = selection.view(
+        #     dtype='<f16',
+        # ).reshape(-1, 4, order='A')
+
+        # assert len(selection) == len(uview)
+
+        u = rfn.structured_to_unstructured(
+            selection,
+            # dtype=float,
+            copy=True,
+        )
+
+        # unstruct = np.ndarray(u.shape, dtype=a.dtype, buffer=shm.buf)
+        # array[:] = a[:]
+        return u
+        # return ShmArray(
+        #     shmarr=u,
+        #     first=self._first,
+        #     last=self._last,
+        #     shm=self._shm
+        # )
+
     def last(
         self,
         length: int = 1,
@@ -255,6 +308,7 @@ class ShmArray:
 
         field_map: Optional[dict[str, str]] = None,
         prepend: bool = False,
+        update_first: bool = True,
         start: Optional[int] = None,
 
     ) -> int:
@@ -267,10 +321,9 @@ class ShmArray:
 
         '''
         length = len(data)
-        index = start if start is not None else self._last.value
 
         if prepend:
-            index = self._first.value - length
+            index = (start or self._first.value) - length
 
             if index < 0:
                 raise ValueError(
@@ -278,6 +331,9 @@ class ShmArray:
                     f'You have passed {abs(index)} too many datums.'
                 )
 
+        else:
+            index = start if start is not None else self._last.value
+
         end = index + length
 
         if field_map:
@@ -295,12 +351,17 @@ class ShmArray:
             # tries to access ``.array`` (which due to the index
             # overlap will be empty). Pretty sure we've fixed it now
             # but leaving this here as a reminder.
-            if prepend:
+            if prepend and update_first and length:
                 assert index < self._first.value
 
-            if index < self._first.value:
+            if (
+                index < self._first.value
+                and update_first
+            ):
+                assert prepend, 'prepend=True not passed but index decreased?'
                 self._first.value = index
-            else:
+
+            elif not prepend:
                 self._last.value = end
 
             self._post_init = True
@@ -336,6 +397,7 @@ class ShmArray:
                 f"Input array has unknown field(s): {only_in_theirs}"
             )
 
+    # TODO: support "silent" prepends that don't update ._first.value?
     def prepend(
         self,
         data: np.ndarray,
@@ -386,7 +448,11 @@ def open_shm_array(
         create=True,
         size=a.nbytes
     )
-    array = np.ndarray(a.shape, dtype=a.dtype, buffer=shm.buf)
+    array = np.ndarray(
+        a.shape,
+        dtype=a.dtype,
+        buffer=shm.buf
+    )
     array[:] = a[:]
     array.setflags(write=int(not readonly))
 

piker/data/_source.py

@@ -21,9 +21,9 @@ from __future__ import annotations
 from typing import Any
 import decimal
 
+from bidict import bidict
 import numpy as np
-import pandas as pd
-from pydantic import BaseModel, validate_arguments
+from pydantic import BaseModel
 # from numba import from_dtype
 
 
@@ -48,16 +48,16 @@ base_ohlc_dtype = np.dtype(ohlc_fields)
 # https://github.com/numba/numba/issues/4511
 # numba_ohlc_dtype = from_dtype(base_ohlc_dtype)
 
-# map time frame "keys" to minutes values
-tf_in_1m = {
-    '1m': 1,
-    '5m':  5,
-    '15m': 15,
-    '30m':  30,
-    '1h': 60,
-    '4h': 240,
-    '1d': 1440,
-}
+# map time frame "keys" to seconds values
+tf_in_1s = bidict({
+    1: '1s',
+    60: '1m',
+    60*5: '5m',
+    60*15: '15m',
+    60*30: '30m',
+    60*60: '1h',
+    60*60*24: '1d',
+})
 
 
 def mk_fqsn(
@@ -127,11 +127,11 @@ def unpack_fqsn(fqsn: str) -> tuple[str, str, str]:
 
 
 class Symbol(BaseModel):
-    """I guess this is some kinda container thing for dealing with
+    '''
+    I guess this is some kinda container thing for dealing with
     all the different meta-data formats from brokers?
 
-    Yah, i guess dats what it izz.
-    """
+    '''
     key: str
     tick_size: float = 0.01
     lot_tick_size: float = 0.0  # "volume" precision as min step value
@@ -254,61 +254,6 @@ class Symbol(BaseModel):
         return keys
 
 
-def from_df(
-
-    df: pd.DataFrame,
-    source=None,
-    default_tf=None
-
-) -> np.recarray:
-    """Convert OHLC formatted ``pandas.DataFrame`` to ``numpy.recarray``.
-
-    """
-    df.reset_index(inplace=True)
-
-    # hackery to convert field names
-    date = 'Date'
-    if 'date' in df.columns:
-        date = 'date'
-
-    # convert to POSIX time
-    df[date] = [d.timestamp() for d in df[date]]
-
-    # try to rename from some camel case
-    columns = {
-        'Date': 'time',
-        'date': 'time',
-        'Open': 'open',
-        'High': 'high',
-        'Low': 'low',
-        'Close': 'close',
-        'Volume': 'volume',
-
-        # most feeds are providing this over sesssion anchored
-        'vwap': 'bar_wap',
-
-        # XXX: ib_insync calls this the "wap of the bar"
-        # but no clue what is actually is...
-        # https://github.com/pikers/piker/issues/119#issuecomment-729120988
-        'average': 'bar_wap',
-    }
-
-    df = df.rename(columns=columns)
-
-    for name in df.columns:
-        # if name not in base_ohlc_dtype.names[1:]:
-        if name not in base_ohlc_dtype.names:
-            del df[name]
-
-    # TODO: it turns out column access on recarrays is actually slower:
-    # https://jakevdp.github.io/PythonDataScienceHandbook/02.09-structured-data-numpy.html#RecordArrays:-Structured-Arrays-with-a-Twist
-    # it might make sense to make these structured arrays?
-    array = df.to_records(index=False)
-    _nan_to_closest_num(array)
-
-    return array
-
-
 def _nan_to_closest_num(array: np.ndarray):
     """Return interpolated values instead of NaN.
 

piker/data/cli.py

@@ -16,26 +16,34 @@
 
 """
 marketstore cli.
+
 """
-from typing import List
 from functools import partial
 from pprint import pformat
 
+from anyio_marketstore import open_marketstore_client
 import trio
 import tractor
 import click
+import numpy as np
 
 from .marketstore import (
     get_client,
-    stream_quotes,
+    # stream_quotes,
     ingest_quote_stream,
-    _url,
+    # _url,
     _tick_tbk_ids,
     mk_tbk,
 )
 from ..cli import cli
 from .. import watchlists as wl
 from ..log import get_logger
+from ._sharedmem import (
+    maybe_open_shm_array,
+)
+from ._source import (
+    base_iohlc_dtype,
+)
 
 
 log = get_logger(__name__)
@@ -49,51 +57,58 @@ log = get_logger(__name__)
 )
 @click.argument('names', nargs=-1)
 @click.pass_obj
-def ms_stream(config: dict, names: List[str], url: str):
-    """Connect to a marketstore time bucket stream for (a set of) symbols(s)
+def ms_stream(
+    config: dict,
+    names: list[str],
+    url: str,
+) -> None:
+    '''
+    Connect to a marketstore time bucket stream for (a set of) symbols(s)
     and print to console.
-    """
+
+    '''
     async def main():
-        async for quote in stream_quotes(symbols=names):
-            log.info(f"Received quote:\n{quote}")
+        # async for quote in stream_quotes(symbols=names):
+        #    log.info(f"Received quote:\n{quote}")
+        ...
 
     trio.run(main)
 
 
-@cli.command()
-@click.option(
-    '--url',
-    default=_url,
-    help='HTTP URL of marketstore instance'
-)
-@click.argument('names', nargs=-1)
-@click.pass_obj
-def ms_destroy(config: dict, names: List[str], url: str) -> None:
-    """Destroy symbol entries in the local marketstore instance.
-    """
-    async def main():
-        nonlocal names
-        async with get_client(url) as client:
-
-            if not names:
-                names = await client.list_symbols()
-
-            # default is to wipe db entirely.
-            answer = input(
-                "This will entirely wipe you local marketstore db @ "
-                f"{url} of the following symbols:\n {pformat(names)}"
-                "\n\nDelete [N/y]?\n")
-
-            if answer == 'y':
-                for sym in names:
-                    # tbk = _tick_tbk.format(sym)
-                    tbk = tuple(sym, *_tick_tbk_ids)
-                    print(f"Destroying {tbk}..")
-                    await client.destroy(mk_tbk(tbk))
-            else:
-                print("Nothing deleted.")
-
-    tractor.run(main)
+# @cli.command()
+# @click.option(
+#     '--url',
+#     default=_url,
+#     help='HTTP URL of marketstore instance'
+# )
+# @click.argument('names', nargs=-1)
+# @click.pass_obj
+# def ms_destroy(config: dict, names: list[str], url: str) -> None:
+#     """Destroy symbol entries in the local marketstore instance.
+#     """
+#     async def main():
+#         nonlocal names
+#         async with get_client(url) as client:
+# 
+#             if not names:
+#                 names = await client.list_symbols()
+# 
+#             # default is to wipe db entirely.
+#             answer = input(
+#                 "This will entirely wipe you local marketstore db @ "
+#                 f"{url} of the following symbols:\n {pformat(names)}"
+#                 "\n\nDelete [N/y]?\n")
+# 
+#             if answer == 'y':
+#                 for sym in names:
+#                     # tbk = _tick_tbk.format(sym)
+#                     tbk = tuple(sym, *_tick_tbk_ids)
+#                     print(f"Destroying {tbk}..")
+#                     await client.destroy(mk_tbk(tbk))
+#             else:
+#                 print("Nothing deleted.")
+# 
+#     tractor.run(main)
 
 
 @cli.command()
@@ -102,41 +117,53 @@ def ms_destroy(config: dict, names: List[str], url: str) -> None:
     is_flag=True,
     help='Enable tractor logging')
 @click.option(
-    '--url',
-    default=_url,
-    help='HTTP URL of marketstore instance'
+    '--host',
+    default='localhost'
 )
-@click.argument('name', nargs=1, required=True)
+@click.option(
+    '--port',
+    default=5993
+)
+@click.argument('symbols', nargs=-1)
 @click.pass_obj
-def ms_shell(config, name, tl, url):
-    """Start an IPython shell ready to query the local marketstore db.
-    """
-    async def main():
-        async with get_client(url) as client:
-            query = client.query  # noqa
-            # TODO: write magics to query marketstore
-            from IPython import embed
-            embed()
+def storesh(
+    config,
+    tl,
+    host,
+    port,
+    symbols: list[str],
+):
+    '''
+    Start an IPython shell ready to query the local marketstore db.
 
-    tractor.run(main)
+    '''
+    from piker.data.marketstore import tsdb_history_update
+    from piker._daemon import open_piker_runtime
+
+    async def main():
+        nonlocal symbols
+
+        async with open_piker_runtime(
+            'storesh',
+            enable_modules=['piker.data._ahab'],
+        ):
+            symbol = symbols[0]
+            await tsdb_history_update(symbol)
+
+    trio.run(main)
 
 
 @cli.command()
 @click.option('--test-file', '-t', help='Test quote stream file')
 @click.option('--tl', is_flag=True, help='Enable tractor logging')
-@click.option('--tl', is_flag=True, help='Enable tractor logging')
-@click.option(
-    '--url',
-    default=_url,
-    help='HTTP URL of marketstore instance'
-)
 @click.argument('name', nargs=1, required=True)
 @click.pass_obj
-def ingest(config, name, test_file, tl, url):
-    """Ingest real-time broker quotes and ticks to a marketstore instance.
-    """
+def ingest(config, name, test_file, tl):
+    '''
+    Ingest real-time broker quotes and ticks to a marketstore instance.
+
+    '''
     # global opts
-    brokermod = config['brokermod']
     loglevel = config['loglevel']
     tractorloglevel = config['tractorloglevel']
     # log = config['log']
@@ -145,15 +172,25 @@ def ingest(config, name, test_file, tl, url):
     watchlists = wl.merge_watchlist(watchlist_from_file, wl._builtins)
     symbols = watchlists[name]
 
-    tractor.run(
-        partial(
-            ingest_quote_stream,
-            symbols,
-            brokermod.name,
-            tries=1,
-            loglevel=loglevel,
-        ),
-        name='ingest_marketstore',
-        loglevel=tractorloglevel,
-        debug_mode=True,
-    )
+    grouped_syms = {}
+    for sym in symbols:
+        symbol, _, provider = sym.rpartition('.')
+        if provider not in grouped_syms:
+            grouped_syms[provider] = []
+
+        grouped_syms[provider].append(symbol)
+
+    async def entry_point():
+        async with tractor.open_nursery() as n:
+            for provider, symbols in grouped_syms.items(): 
+                await n.run_in_actor(
+                    ingest_quote_stream,
+                    name='ingest_marketstore',
+                    symbols=symbols,
+                    brokername=provider,
+                    tries=1,
+                    actorloglevel=loglevel,
+                    loglevel=tractorloglevel
+                )
+
+    tractor.run(entry_point)

piker/data/feed.py

@@ -20,27 +20,35 @@ Data feed apis and infra.
 This module is enabled for ``brokerd`` daemons.
 
 """
+from __future__ import annotations
 from dataclasses import dataclass, field
+from datetime import datetime
 from contextlib import asynccontextmanager
 from functools import partial
 from types import ModuleType
 from typing import (
     Any,
     AsyncIterator, Optional,
+    Generator,
     Awaitable,
+    TYPE_CHECKING,
 )
 
 import trio
 from trio.abc import ReceiveChannel
 from trio_typing import TaskStatus
+import trimeter
 import tractor
 from pydantic import BaseModel
+import pendulum
+import numpy as np
 
 from ..brokers import get_brokermod
 from .._cacheables import maybe_open_context
 from ..log import get_logger, get_console_log
 from .._daemon import (
     maybe_spawn_brokerd,
+    check_for_service,
 )
 from ._sharedmem import (
     maybe_open_shm_array,
@@ -56,12 +64,19 @@ from ._source import (
 from ..ui import _search
 from ._sampling import (
     sampler,
+    broadcast,
     increment_ohlc_buffer,
     iter_ohlc_periods,
     sample_and_broadcast,
     uniform_rate_send,
 )
+from ..brokers._util import (
+    NoData,
+    DataUnavailable,
+)
 
+if TYPE_CHECKING:
+    from .marketstore import Storage
 
 log = get_logger(__name__)
 
@@ -124,7 +139,7 @@ class _FeedsBus(BaseModel):
 
     # def cancel_task(
     #     self,
-    #     task: trio.lowlevel.Task
+    #     task: trio.lowlevel.Task,
     # ) -> bool:
     #     ...
 
@@ -188,6 +203,388 @@ async def _setup_persistent_brokerd(
         await trio.sleep_forever()
 
 
+def diff_history(
+    array,
+    start_dt,
+    end_dt,
+    last_tsdb_dt: Optional[datetime] = None
+
+) -> np.ndarray:
+
+    to_push = array
+
+    if last_tsdb_dt:
+        s_diff = (start_dt - last_tsdb_dt).seconds
+
+        # if we detect a partial frame's worth of data
+        # that is new, slice out only that history and
+        # write to shm.
+        if (
+            s_diff < 0
+            and abs(s_diff) < len(array)
+        ):
+            # the + 1 is because ``last_tsdb_dt`` is pulled from
+            # the last row entry for the ``'time'`` field retreived
+            # from the tsdb.
+            to_push = array[abs(s_diff)+1:]
+            log.info(
+                f'Pushing partial frame {to_push.size} to shm'
+            )
+
+    return to_push
+
+
+async def start_backfill(
+    mod: ModuleType,
+    bfqsn: str,
+    shm: ShmArray,
+
+    last_tsdb_dt: Optional[datetime] = None,
+    storage: Optional[Storage] = None,
+    write_tsdb: bool = False,
+
+    task_status: TaskStatus[trio.CancelScope] = trio.TASK_STATUS_IGNORED,
+
+) -> int:
+
+    async with mod.open_history_client(bfqsn) as (hist, config):
+
+        # get latest query's worth of history all the way
+        # back to what is recorded in the tsdb
+        array, start_dt, end_dt = await hist(end_dt=None)
+
+        times = array['time']
+
+        # sample period step size in seconds
+        step_size_s = (
+            pendulum.from_timestamp(times[-1]) -
+            pendulum.from_timestamp(times[-2])
+        ).seconds
+
+        # "frame"'s worth of sample period steps in seconds
+        frame_size_s = len(array) * step_size_s
+
+        to_push = diff_history(
+            array,
+            start_dt,
+            end_dt,
+            last_tsdb_dt=last_tsdb_dt,
+        )
+
+        log.info(f'Pushing {to_push.size} to shm!')
+        shm.push(to_push)
+
+        for delay_s in sampler.subscribers:
+            await broadcast(delay_s)
+
+        # signal that backfilling to tsdb's end datum is complete
+        bf_done = trio.Event()
+
+        # let caller unblock and deliver latest history frame
+        task_status.started((shm, start_dt, end_dt, bf_done))
+
+        if last_tsdb_dt is None:
+            # maybe a better default (they don't seem to define epoch?!)
+
+            # based on the sample step size load a certain amount
+            # history
+            if step_size_s == 1:
+                last_tsdb_dt = pendulum.now().subtract(days=2)
+
+            elif step_size_s == 60:
+                last_tsdb_dt = pendulum.now().subtract(years=2)
+
+            else:
+                raise ValueError(
+                    '`piker` only needs to support 1m and 1s sampling '
+                    'but ur api is trying to deliver a longer '
+                    f'timeframe of {step_size_s} ' 'seconds.. so ye, dun '
+                    'do dat bruh.'
+                )
+
+        # configure async query throttling
+        erlangs = config.get('erlangs', 1)
+        rate = config.get('rate', 1)
+        frames = {}
+
+        def iter_dts(start: datetime):
+            while True:
+
+                hist_period = pendulum.period(
+                    start.subtract(seconds=step_size_s),
+                    last_tsdb_dt,
+                )
+                dtrange = hist_period.range('seconds', frame_size_s)
+
+                for end_dt in dtrange:
+                    log.warning(f'Yielding next frame start {end_dt}')
+                    start = yield end_dt
+
+                    # if caller sends a new start date, reset to that
+                    if start is not None:
+                        log.warning(f'Resetting date range: {start}')
+                        # import pdbpp
+                        # pdbpp.set_trace()
+                        break
+                else:
+                    # from while
+                    return
+
+        # pull new history frames until we hit latest
+        # already in the tsdb or a max count.
+        count = 0
+
+        # NOTE: when gaps are detected in the retreived history (by
+        # comparisor of the end - start versus the expected "frame size"
+        # in seconds) we need a way to alert the async request code not
+        # to continue to query for data "within the gap". This var is
+        # set in such cases such that further requests in that period
+        # are discarded and further we reset the "datetimem query frame
+        # index" in such cases to avoid needless noop requests.
+        earliest_end_dt: Optional[datetime] = start_dt
+
+        async def get_ohlc_frame(
+            input_end_dt: datetime,
+            iter_dts_gen: Generator[datetime],
+
+        ) -> np.ndarray:
+
+            nonlocal count, frames, earliest_end_dt, frame_size_s
+            count += 1
+
+            if input_end_dt > earliest_end_dt:
+                # if a request comes in for an inter-gap frame we
+                # discard it since likely this request is still
+                # lingering from before the reset of ``iter_dts()`` via
+                # ``.send()`` below.
+                log.info(f'Discarding request history ending @ {input_end_dt}')
+
+                # signals to ``trimeter`` loop to discard and
+                # ``continue`` in it's schedule loop.
+                return None
+
+            try:
+                log.info(
+                    f'Requesting {step_size_s}s frame ending in {input_end_dt}'
+                )
+                array, start_dt, end_dt = await hist(end_dt=input_end_dt)
+                assert array['time'][0] == start_dt.timestamp()
+
+            except NoData:
+                log.warning(
+                    f'NO DATA for {frame_size_s}s frame @ {input_end_dt} ?!?'
+                )
+                return None  # discard signal
+
+            except DataUnavailable as duerr:
+                # broker is being a bish and we can't pull
+                # any more..
+                log.warning('backend halted on data deliver !?!?')
+
+                # ugh, what's a better way?
+                # TODO: fwiw, we probably want a way to signal a throttle
+                # condition (eg. with ib) so that we can halt the
+                # request loop until the condition is resolved?
+                return duerr
+
+            diff = end_dt - start_dt
+            frame_time_diff_s = diff.seconds
+            expected_frame_size_s = frame_size_s + step_size_s
+
+            if frame_time_diff_s > expected_frame_size_s:
+
+                # XXX: query result includes a start point prior to our
+                # expected "frame size" and thus is likely some kind of
+                # history gap (eg. market closed period, outage, etc.)
+                # so indicate to the request loop that this gap is
+                # expected by both,
+                # - resetting the ``iter_dts()`` generator to start at
+                #   the new start point delivered in this result
+                # - setting the non-locally scoped ``earliest_end_dt``
+                #   to this new value so that the request loop doesn't
+                #   get tripped up thinking there's an out of order
+                #   request-result condition.
+
+                log.warning(
+                    f'History frame ending @ {end_dt} appears to have a gap:\n'
+                    f'{diff} ~= {frame_time_diff_s} seconds'
+                )
+
+                # reset dtrange gen to new start point
+                try:
+                    next_end = iter_dts_gen.send(start_dt)
+                    log.info(
+                        f'Reset frame index to start at {start_dt}\n'
+                        f'Was at {next_end}'
+                    )
+
+                    # NOTE: manually set "earliest end datetime" index-value
+                    # to avoid the request loop getting confused about
+                    # new frames that are earlier in history - i.e. this
+                    # **is not** the case of out-of-order frames from
+                    # an async batch request.
+                    earliest_end_dt = start_dt
+
+                except StopIteration:
+                    # gen already terminated meaning we probably already
+                    # exhausted it via frame requests.
+                    log.info(
+                        "Datetime index already exhausted, can't reset.."
+                    )
+
+            to_push = diff_history(
+                array,
+                start_dt,
+                end_dt,
+                last_tsdb_dt=last_tsdb_dt,
+            )
+            ln = len(to_push)
+            if ln:
+                log.info(f'{ln} bars for {start_dt} -> {end_dt}')
+                frames[input_end_dt.timestamp()] = (to_push, start_dt, end_dt)
+                return to_push, start_dt, end_dt
+
+            else:
+                log.warning(
+                    f'{ln} BARS TO PUSH after diff?!: {start_dt} -> {end_dt}'
+                )
+                return None
+
+        # initial dt index starts at the start of the first query result
+        idts = iter_dts(start_dt)
+
+        async with trimeter.amap(
+            partial(
+                get_ohlc_frame,
+                # we close in the ``iter_dt()`` gen in so we can send
+                # reset signals as needed for gap dection in the
+                # history.
+                iter_dts_gen=idts,
+            ),
+            idts,
+
+            capture_outcome=True,
+            include_value=True,
+
+            # better technical names bruv...
+            max_at_once=erlangs,
+            max_per_second=rate,
+
+        ) as outcomes:
+
+            # Then iterate over the return values, as they become available
+            # (i.e., not necessarily in the original order)
+            async for input_end_dt, outcome in outcomes:
+
+                try:
+                    out = outcome.unwrap()
+
+                    if out is None:
+                        # skip signal
+                        continue
+
+                    elif isinstance(out, DataUnavailable):
+                        # no data available case signal.. so just kill
+                        # further requests and basically just stop
+                        # trying...
+                        break
+
+                except Exception:
+                    log.exception('uhh trimeter bail')
+                    raise
+                else:
+                    to_push, start_dt, end_dt = out
+
+                if not len(to_push):
+                    # diff returned no new data (i.e. we probablyl hit
+                    # the ``last_tsdb_dt`` point).
+                    # TODO: raise instead?
+                    log.warning(f'No history for range {start_dt} -> {end_dt}')
+                    continue
+
+                # pipeline-style pull frames until we need to wait for
+                # the next in order to arrive.
+                # i = end_dts.index(input_end_dt)
+                # print(f'latest end_dt {end_dt} found at index {i}')
+
+                epochs = list(reversed(sorted(frames)))
+                for epoch in epochs:
+                    start = shm.array['time'][0]
+
+                    diff = epoch - start
+                    if abs(diff) > step_size_s:
+
+                        if earliest_end_dt < end_dt:
+                            # XXX: an expected gap was encountered (see
+                            # logic in ``get_ohlc_frame()``, so allow
+                            # this frame through to the storage layer.
+                            log.warning(
+                                f'there is an expected history gap of {diff}s:'
+                            )
+
+                        elif (
+                            erlangs > 1
+                            and len(epochs) < erlangs
+                        ):
+                            # we don't yet have the next frame to push
+                            # so break back to the async request loop
+                            # while we wait for more async frame-results
+                            # to arrive.
+                            expect_end = pendulum.from_timestamp(start)
+                            expect_start = expect_end.subtract(
+                                seconds=frame_size_s)
+                            log.warning(
+                                'waiting on out-of-order history frame:\n'
+                                f'{expect_end - expect_start}'
+                            )
+                            break
+
+                    to_push, start_dt, end_dt = frames.pop(epoch)
+                    ln = len(to_push)
+
+                    # bail gracefully on shm allocation overrun/full condition
+                    try:
+                        shm.push(to_push, prepend=True)
+                    except ValueError:
+                        log.info(
+                            f'Shm buffer overrun on: {start_dt} -> {end_dt}?'
+                        )
+                        break
+
+                    log.info(
+                        f'Shm pushed {ln} frame:\n'
+                        f'{start_dt} -> {end_dt}'
+                    )
+                    # keep track of most recent "prepended" ``start_dt``
+                    # both for detecting gaps and ensuring async
+                    # frame-result order.
+                    earliest_end_dt = start_dt
+
+                    if (
+                        storage is not None
+                        and write_tsdb
+                    ):
+                        log.info(
+                            f'Writing {ln} frame to storage:\n'
+                            f'{start_dt} -> {end_dt}'
+                        )
+                        await storage.write_ohlcv(
+                            f'{bfqsn}.{mod.name}',  # lul..
+                            to_push,
+                        )
+
+                # TODO: can we only trigger this if the respective
+                # history in "in view"?!?
+                # XXX: extremely important, there can be no checkpoints
+                # in the block above to avoid entering new ``frames``
+                # values while we're pipelining the current ones to
+                # memory...
+                for delay_s in sampler.subscribers:
+                    await broadcast(delay_s)
+
+        bf_done.set()
+
+
 async def manage_history(
     mod: ModuleType,
     bus: _FeedsBus,
@@ -216,50 +613,180 @@ async def manage_history(
         # we expect the sub-actor to write
         readonly=False,
     )
+    # TODO: history validation
+    if not opened:
+        raise RuntimeError(
+            "Persistent shm for sym was already open?!"
+        )
 
-    if opened:
+    log.info('Scanning for existing `marketstored`')
+
+    is_up = await check_for_service('marketstored')
+
+    # for now only do backfilling if no tsdb can be found
+    do_legacy_backfill = not is_up and opened
+
+    bfqsn = fqsn.replace('.' + mod.name, '')
+    open_history_client = getattr(mod, 'open_history_client', None)
+
+    if is_up and opened and open_history_client:
+
+        log.info('Found existing `marketstored`')
+        from . import marketstore
+        async with marketstore.open_storage_client(
+            fqsn,
+        ) as storage:
+
+            # TODO: this should be used verbatim for the pure
+            # shm backfiller approach below.
+
+            # start history anal and load missing new data via backend.
+            series, _, last_tsdb_dt = await storage.load(fqsn)
+
+            broker, symbol, expiry = unpack_fqsn(fqsn)
+            (
+                shm,
+                latest_start_dt,
+                latest_end_dt,
+                bf_done,
+            ) = await bus.nursery.start(
+                partial(
+                    start_backfill,
+                    mod,
+                    bfqsn,
+                    shm,
+                    last_tsdb_dt=last_tsdb_dt,
+                    storage=storage,
+                )
+            )
+
+            # if len(shm.array) < 2:
+            # TODO: there's an edge case here to solve where if the last
+            # frame before market close (at least on ib) was pushed and
+            # there was only "1 new" row pushed from the first backfill
+            # query-iteration, then the sample step sizing calcs will
+            # break upstream from here since you can't diff on at least
+            # 2 steps... probably should also add logic to compute from
+            # the tsdb series and stash that somewhere as meta data on
+            # the shm buffer?.. no se.
+
+            task_status.started(shm)
+            some_data_ready.set()
+
+            await bf_done.wait()
+            # do diff against last start frame of history and only fill
+            # in from the tsdb an allotment that allows for most recent
+            # to be loaded into mem *before* tsdb data.
+            if last_tsdb_dt:
+                dt_diff_s = (
+                    latest_start_dt - last_tsdb_dt
+                ).seconds
+            else:
+                dt_diff_s = 0
+
+            # await trio.sleep_forever()
+            # TODO: see if there's faster multi-field reads:
+            # https://numpy.org/doc/stable/user/basics.rec.html#accessing-multiple-fields
+            # re-index  with a `time` and index field
+            prepend_start = shm._first.value
+
+            # sanity check on most-recent-data loading
+            assert prepend_start > dt_diff_s
+
+            history = list(series.values())
+            if history:
+                fastest = history[0]
+                to_push = fastest[:prepend_start]
+
+                shm.push(
+                    to_push,
+
+                    # insert the history pre a "days worth" of samples
+                    # to leave some real-time buffer space at the end.
+                    prepend=True,
+                    # update_first=False,
+                    # start=prepend_start,
+                    field_map=marketstore.ohlc_key_map,
+                )
+
+                # load as much from storage into shm as space will
+                # allow according to user's shm size settings.
+                count = 0
+                end = fastest['Epoch'][0]
+
+                while shm._first.value > 0:
+                    count += 1
+                    series = await storage.read_ohlcv(
+                        fqsn,
+                        end=end,
+                    )
+                    history = list(series.values())
+                    fastest = history[0]
+                    end = fastest['Epoch'][0]
+                    prepend_start -= len(to_push)
+                    to_push = fastest[:prepend_start]
+
+                    shm.push(
+                        to_push,
+
+                        # insert the history pre a "days worth" of samples
+                        # to leave some real-time buffer space at the end.
+                        prepend=True,
+                        # update_first=False,
+                        # start=prepend_start,
+                        field_map=marketstore.ohlc_key_map,
+                    )
+
+                    # manually trigger step update to update charts/fsps
+                    # which need an incremental update.
+                    for delay_s in sampler.subscribers:
+                        await broadcast(delay_s)
+
+                    if count > 6:
+                        break
+
+                log.info(f'Loaded {to_push.shape} datums from storage')
+
+                # TODO: write new data to tsdb to be ready to for next read.
+
+    if do_legacy_backfill:
+        # do a legacy incremental backfill from the provider.
         log.info('No existing `marketstored` found..')
 
         # start history backfill task ``backfill_bars()`` is
         # a required backend func this must block until shm is
         # filled with first set of ohlc bars
-        _ = await bus.nursery.start(mod.backfill_bars, fqsn, shm)
-
-    # yield back after client connect with filled shm
-    task_status.started(shm)
-
-    # indicate to caller that feed can be delivered to
-    # remote requesting client since we've loaded history
-    # data that can be used.
-    some_data_ready.set()
-
-    # detect sample step size for sampled historical data
-    times = shm.array['time']
-    delay_s = times[-1] - times[times != times[-1]][-1]
-
-    # begin real-time updates of shm and tsb once the feed
-    # goes live.
-    await feed_is_live.wait()
-
-    if opened:
-        sampler.ohlcv_shms.setdefault(delay_s, []).append(shm)
-
-        # start shm incrementing for OHLC sampling at the current
-        # detected sampling period if one dne.
-        if sampler.incrementers.get(delay_s) is None:
-            await bus.start_task(
-                increment_ohlc_buffer,
-                delay_s,
+        await bus.nursery.start(
+            partial(
+                start_backfill,
+                mod,
+                bfqsn,
+                shm,
             )
+        )
 
+        # yield back after client connect with filled shm
+        task_status.started(shm)
+
+        # indicate to caller that feed can be delivered to
+        # remote requesting client since we've loaded history
+        # data that can be used.
+        some_data_ready.set()
+
+    # history retreival loop depending on user interaction and thus
+    # a small RPC-prot for remotely controllinlg what data is loaded
+    # for viewing.
     await trio.sleep_forever()
 
 
 async def allocate_persistent_feed(
     bus: _FeedsBus,
+
     brokername: str,
     symbol: str,
+
     loglevel: str,
+    start_stream: bool = True,
 
     task_status: TaskStatus[trio.CancelScope] = trio.TASK_STATUS_IGNORED,
 
@@ -277,6 +804,7 @@ async def allocate_persistent_feed(
     - a real-time streaming task which connec
 
     '''
+    # load backend module
     try:
         mod = get_brokermod(brokername)
     except ImportError:
@@ -319,7 +847,7 @@ async def allocate_persistent_feed(
         manage_history,
         mod,
         bus,
-        bfqsn,
+        '.'.join((bfqsn, brokername)),
         some_data_ready,
         feed_is_live,
     )
@@ -333,7 +861,10 @@ async def allocate_persistent_feed(
     # true fqsn
     fqsn = '.'.join((bfqsn, brokername))
     # add a fqsn entry that includes the ``.<broker>`` suffix
+    # and an entry that includes the broker-specific fqsn (including
+    # any new suffixes or elements as injected by the backend).
     init_msg[fqsn] = msg
+    init_msg[bfqsn] = msg
 
     # TODO: pretty sure we don't need this? why not just leave 1s as
     # the fastest "sample period" since we'll probably always want that
@@ -347,13 +878,14 @@ async def allocate_persistent_feed(
     await some_data_ready.wait()
 
     # append ``.<broker>`` suffix to each quote symbol
-    bsym = symbol + f'.{brokername}'
+    acceptable_not_fqsn_with_broker_suffix = symbol + f'.{brokername}'
+
     generic_first_quotes = {
-        bsym: first_quote,
+        acceptable_not_fqsn_with_broker_suffix: first_quote,
         fqsn: first_quote,
     }
 
-    bus.feeds[symbol] = bus.feeds[fqsn] = (
+    bus.feeds[symbol] = bus.feeds[bfqsn] = (
         init_msg,
         generic_first_quotes,
     )
@@ -363,9 +895,25 @@ async def allocate_persistent_feed(
     # task_status.started((init_msg,  generic_first_quotes))
     task_status.started()
 
-    # backend will indicate when real-time quotes have begun.
+    if not start_stream:
+        await trio.sleep_forever()
+
+    # begin real-time updates of shm and tsb once the feed goes live and
+    # the backend will indicate when real-time quotes have begun.
     await feed_is_live.wait()
 
+    # start shm incrementer task for OHLC style sampling
+    # at the current detected step period.
+    times = shm.array['time']
+    delay_s = times[-1] - times[times != times[-1]][-1]
+
+    sampler.ohlcv_shms.setdefault(delay_s, []).append(shm)
+    if sampler.incrementers.get(delay_s) is None:
+        await bus.start_task(
+            increment_ohlc_buffer,
+            delay_s,
+        )
+
     sum_tick_vlm: bool = init_msg.get(
         'shm_write_opts', {}
     ).get('sum_tick_vlm', True)
@@ -388,7 +936,7 @@ async def open_feed_bus(
 
     ctx: tractor.Context,
     brokername: str,
-    symbol: str,
+    symbol: str,  # normally expected to the broker-specific fqsn
     loglevel: str,
     tick_throttle:  Optional[float] = None,
     start_stream: bool = True,
@@ -410,7 +958,9 @@ async def open_feed_bus(
     # TODO: check for any stale shm entries for this symbol
     # (after we also group them in a nice `/dev/shm/piker/` subdir).
     # ensure we are who we think we are
-    assert 'brokerd' in tractor.current_actor().name
+    servicename = tractor.current_actor().name
+    assert 'brokerd' in servicename
+    assert brokername in servicename
 
     bus = get_feed_bus(brokername)
 
@@ -420,7 +970,7 @@ async def open_feed_bus(
     entry = bus.feeds.get(symbol)
     if entry is None:
         # allocate a new actor-local stream bus which
-        # will persist for this `brokerd`.
+        # will persist for this `brokerd`'s service lifetime.
         async with bus.task_lock:
             await bus.nursery.start(
                 partial(
@@ -428,13 +978,12 @@ async def open_feed_bus(
 
                     bus=bus,
                     brokername=brokername,
-
                     # here we pass through the selected symbol in native
                     # "format" (i.e. upper vs. lowercase depending on
                     # provider).
                     symbol=symbol,
-
                     loglevel=loglevel,
+                    start_stream=start_stream,
                 )
             )
             # TODO: we can remove this?
@@ -450,7 +999,7 @@ async def open_feed_bus(
     # true fqsn
     fqsn = '.'.join([bfqsn, brokername])
     assert fqsn in first_quotes
-    assert bus.feeds[fqsn]
+    assert bus.feeds[bfqsn]
 
     # broker-ambiguous symbol (provided on cli - eg. mnq.globex.ib)
     bsym = symbol + f'.{brokername}'

piker/data/marketstore.py

@@ -14,36 +14,60 @@
 # You should have received a copy of the GNU Affero General Public License
 # along with this program.  If not, see <https://www.gnu.org/licenses/>.
 
-"""
+'''
 ``marketstore`` integration.
 
 - client management routines
 - ticK data ingest routines
 - websocket client for subscribing to write triggers
 - todo: tick sequence stream-cloning for testing
-- todo: docker container management automation
-"""
-from contextlib import asynccontextmanager
-from typing import Dict, Any, List, Callable, Tuple
+
+'''
+from contextlib import asynccontextmanager as acm
+from datetime import datetime
+from pprint import pformat
+from typing import (
+    Any,
+    Optional,
+    Union,
+)
 import time
 from math import isnan
 
+from bidict import bidict
 import msgpack
+import pyqtgraph as pg
 import numpy as np
 import pandas as pd
-import pymarketstore as pymkts
 import tractor
 from trio_websocket import open_websocket_url
+from anyio_marketstore import (
+    open_marketstore_client,
+    MarketstoreClient,
+    Params,
+)
+import pendulum
+import purerpc
 
+from .feed import maybe_open_feed
 from ..log import get_logger, get_console_log
-from ..data import open_feed
 
 
 log = get_logger(__name__)
 
-_tick_tbk_ids: Tuple[str, str] = ('1Sec', 'TICK')
+_tick_tbk_ids: tuple[str, str] = ('1Sec', 'TICK')
 _tick_tbk: str = '{}/' + '/'.join(_tick_tbk_ids)
-_url: str = 'http://localhost:5993/rpc'
+
+_tick_dt = [
+    # these two are required for as a "primary key"
+    ('Epoch', 'i8'),
+    ('Nanoseconds', 'i4'),
+    ('IsTrade', 'i1'),
+    ('IsBid', 'i1'),
+    ('Price', 'f4'),
+    ('Size', 'f4')
+]
+
 _quote_dt = [
     # these two are required for as a "primary key"
     ('Epoch', 'i8'),
@@ -61,6 +85,7 @@ _quote_dt = [
     # ('brokerd_ts', 'i64'),
     # ('VWAP', 'f4')
 ]
+
 _quote_tmp = {}.fromkeys(dict(_quote_dt).keys(), np.nan)
 _tick_map = {
     'Up': 1,
@@ -69,28 +94,49 @@ _tick_map = {
     None: np.nan,
 }
 
+_ohlcv_dt = [
+    # these two are required for as a "primary key"
+    ('Epoch', 'i8'),
+    # ('Nanoseconds', 'i4'),
 
-class MarketStoreError(Exception):
-    "Generic marketstore client error"
+    # ohlcv sampling
+    ('Open', 'f4'),
+    ('High', 'f4'),
+    ('Low', 'i8'),
+    ('Close', 'i8'),
+    ('Volume', 'f4'),
+]
 
 
-def err_on_resp(response: dict) -> None:
-    """Raise any errors found in responses from client request.
-    """
-    responses = response['responses']
-    if responses is not None:
-        for r in responses:
-            err = r['error']
-            if err:
-                raise MarketStoreError(err)
+ohlc_key_map = bidict({
+    'Epoch': 'time',
+    'Open': 'open',
+    'High': 'high',
+    'Low': 'low',
+    'Close': 'close',
+    'Volume': 'volume',
+})
+
+
+def mk_tbk(keys: tuple[str, str, str]) -> str:
+    '''
+    Generate a marketstore table key from a tuple.
+    Converts,
+        ``('SPY', '1Sec', 'TICK')`` -> ``"SPY/1Sec/TICK"```
+
+    '''
+    return '/'.join(keys)
 
 
 def quote_to_marketstore_structarray(
-    quote: Dict[str, Any],
-    last_fill: str,
+    quote: dict[str, Any],
+    last_fill: Optional[float]
+
 ) -> np.array:
-    """Return marketstore writeable structarray from quote ``dict``.
-    """
+    '''
+    Return marketstore writeable structarray from quote ``dict``.
+
+    '''
     if last_fill:
         # new fill bby
         now = timestamp(last_fill)
@@ -101,7 +147,7 @@ def quote_to_marketstore_structarray(
 
     secs, ns = now / 10**9, now % 10**9
 
-    # pack into List[Tuple[str, Any]]
+    # pack into list[tuple[str, Any]]
     array_input = []
 
     # insert 'Epoch' entry first and then 'Nanoseconds'.
@@ -123,146 +169,456 @@ def quote_to_marketstore_structarray(
     return np.array([tuple(array_input)], dtype=_quote_dt)
 
 
-def timestamp(datestr: str) -> int:
-    """Return marketstore compatible 'Epoch' integer in nanoseconds
+def timestamp(date, **kwargs) -> int:
+    '''
+    Return marketstore compatible 'Epoch' integer in nanoseconds
     from a date formatted str.
-    """
-    return int(pd.Timestamp(datestr).value)
+
+    '''
+    return int(pd.Timestamp(date, **kwargs).value)
 
 
-def mk_tbk(keys: Tuple[str, str, str]) -> str:
-    """Generate a marketstore table key from a tuple.
-
-    Converts,
-        ``('SPY', '1Sec', 'TICK')`` -> ``"SPY/1Sec/TICK"```
-    """
-    return '{}/' + '/'.join(keys)
-
-
-class Client:
-    """Async wrapper around the alpaca ``pymarketstore`` sync client.
-
-    This will server as the shell for building out a proper async client
-    that isn't horribly documented and un-tested..
-    """
-    def __init__(self, url: str):
-        self._client = pymkts.Client(url)
-
-    async def _invoke(
-        self,
-        meth: Callable,
-        *args,
-        **kwargs,
-    ) -> Any:
-        return err_on_resp(meth(*args, **kwargs))
-
-    async def destroy(
-        self,
-        tbk: Tuple[str, str, str],
-    ) -> None:
-        return await self._invoke(self._client.destroy, mk_tbk(tbk))
-
-    async def list_symbols(
-        self,
-        tbk: str,
-    ) -> List[str]:
-        return await self._invoke(self._client.list_symbols, mk_tbk(tbk))
-
-    async def write(
-        self,
-        symbol: str,
-        array: np.ndarray,
-    ) -> None:
-        start = time.time()
-        await self._invoke(
-            self._client.write,
-            array,
-            _tick_tbk.format(symbol),
-            isvariablelength=True
-        )
-        log.debug(f"{symbol} write time (s): {time.time() - start}")
-
-    def query(
-        self,
-        symbol,
-        tbk: Tuple[str, str] = _tick_tbk_ids,
-    ) -> pd.DataFrame:
-        # XXX: causes crash
-        # client.query(pymkts.Params(symbol, '*', 'OHCLV'
-        result = self._client.query(
-            pymkts.Params(symbol, *tbk),
-        )
-        return result.first().df()
-
-
-@asynccontextmanager
+@acm
 async def get_client(
-    url: str = _url,
-) -> Client:
-    yield Client(url)
+    host: str = 'localhost',
+    port: int = 5995
+
+) -> MarketstoreClient:
+    '''
+    Load a ``anyio_marketstore`` grpc client connected
+    to an existing ``marketstore`` server.
+
+    '''
+    async with open_marketstore_client(
+        host,
+        port
+    ) as client:
+        yield client
+
+
+class MarketStoreError(Exception):
+    "Generic marketstore client error"
+
+
+# def err_on_resp(response: dict) -> None:
+#     """Raise any errors found in responses from client request.
+#     """
+#     responses = response['responses']
+#     if responses is not None:
+#         for r in responses:
+#             err = r['error']
+#             if err:
+#                 raise MarketStoreError(err)
+
+
+# map of seconds ints to "time frame" accepted keys
+tf_in_1s = bidict({
+    1: '1Sec',
+    60: '1Min',
+    60*5: '5Min',
+    60*15: '15Min',
+    60*30: '30Min',
+    60*60: '1H',
+    60*60*24: '1D',
+})
+
+
+class Storage:
+    '''
+    High level storage api for both real-time and historical ingest.
+
+    '''
+    def __init__(
+        self,
+        client: MarketstoreClient,
+
+    ) -> None:
+        # TODO: eventually this should be an api/interface type that
+        # ensures we can support multiple tsdb backends.
+        self.client = client
+
+        # series' cache from tsdb reads
+        self._arrays: dict[str, np.ndarray] = {}
+
+    async def list_keys(self) -> list[str]:
+        return await self.client.list_symbols()
+
+    async def search_keys(self, pattern: str) -> list[str]:
+        '''
+        Search for time series key in the storage backend.
+
+        '''
+        ...
+
+    async def write_ticks(self, ticks: list) -> None:
+        ...
+
+    async def load(
+        self,
+        fqsn: str,
+
+    ) -> tuple[
+        dict[int, np.ndarray],  # timeframe (in secs) to series
+        Optional[datetime],  # first dt
+        Optional[datetime],  # last dt
+    ]:
+
+        first_tsdb_dt, last_tsdb_dt = None, None
+        tsdb_arrays = await self.read_ohlcv(fqsn)
+        log.info(f'Loaded tsdb history {tsdb_arrays}')
+
+        if tsdb_arrays:
+            fastest = list(tsdb_arrays.values())[0]
+            times = fastest['Epoch']
+            first, last = times[0], times[-1]
+            first_tsdb_dt, last_tsdb_dt = map(
+                pendulum.from_timestamp, [first, last]
+            )
+
+        return tsdb_arrays, first_tsdb_dt, last_tsdb_dt
+
+    async def read_ohlcv(
+        self,
+        fqsn: str,
+        timeframe: Optional[Union[int, str]] = None,
+        end: Optional[int] = None,
+
+    ) -> tuple[
+        MarketstoreClient,
+        Union[dict, np.ndarray]
+    ]:
+        client = self.client
+        syms = await client.list_symbols()
+
+        if fqsn not in syms:
+            return {}
+
+        tfstr = tf_in_1s[1]
+
+        params = Params(
+            symbols=fqsn,
+            timeframe=tfstr,
+            attrgroup='OHLCV',
+            end=end,
+            # limit_from_start=True,
+
+            # TODO: figure the max limit here given the
+            # ``purepc`` msg size limit of purerpc: 33554432
+            limit=int(800e3),
+        )
+
+        if timeframe is None:
+            log.info(f'starting {fqsn} tsdb granularity scan..')
+            # loop through and try to find highest granularity
+            for tfstr in tf_in_1s.values():
+                try:
+                    log.info(f'querying for {tfstr}@{fqsn}')
+                    params.set('timeframe', tfstr)
+                    result = await client.query(params)
+                    break
+
+                except purerpc.grpclib.exceptions.UnknownError:
+                    # XXX: this is already logged by the container and
+                    # thus shows up through `marketstored` logs relay.
+                    # log.warning(f'{tfstr}@{fqsn} not found')
+                    continue
+            else:
+                return {}
+
+        else:
+            result = await client.query(params)
+
+        # TODO: it turns out column access on recarrays is actually slower:
+        # https://jakevdp.github.io/PythonDataScienceHandbook/02.09-structured-data-numpy.html#RecordArrays:-Structured-Arrays-with-a-Twist
+        # it might make sense to make these structured arrays?
+        # Fill out a `numpy` array-results map
+        arrays = {}
+        for fqsn, data_set in result.by_symbols().items():
+            arrays.setdefault(fqsn, {})[
+                tf_in_1s.inverse[data_set.timeframe]
+            ] = data_set.array
+
+        return arrays[fqsn][timeframe] if timeframe else arrays[fqsn]
+
+    async def delete_ts(
+        self,
+        key: str,
+        timeframe: Optional[Union[int, str]] = None,
+
+    ) -> bool:
+
+        client = self.client
+        syms = await client.list_symbols()
+        print(syms)
+        # if key not in syms:
+        #     raise KeyError(f'`{fqsn}` table key not found?')
+
+        return await client.destroy(tbk=key)
+
+    async def write_ohlcv(
+        self,
+        fqsn: str,
+        ohlcv: np.ndarray,
+        append_and_duplicate: bool = True,
+        limit: int = int(800e3),
+
+    ) -> None:
+        # build mkts schema compat array for writing
+        mkts_dt = np.dtype(_ohlcv_dt)
+        mkts_array = np.zeros(
+            len(ohlcv),
+            dtype=mkts_dt,
+        )
+        # copy from shm array (yes it's this easy):
+        # https://numpy.org/doc/stable/user/basics.rec.html#assignment-from-other-structured-arrays
+        mkts_array[:] = ohlcv[[
+            'time',
+            'open',
+            'high',
+            'low',
+            'close',
+            'volume',
+        ]]
+
+        m, r = divmod(len(mkts_array), limit)
+
+        for i in range(m, 1):
+            to_push = mkts_array[i-1:i*limit]
+
+            # write to db
+            resp = await self.client.write(
+                to_push,
+                tbk=f'{fqsn}/1Sec/OHLCV',
+
+                # NOTE: will will append duplicates
+                # for the same timestamp-index.
+                # TODO: pre deduplicate?
+                isvariablelength=append_and_duplicate,
+            )
+
+            log.info(
+                f'Wrote {mkts_array.size} datums to tsdb\n'
+            )
+
+            for resp in resp.responses:
+                err = resp.error
+                if err:
+                    raise MarketStoreError(err)
+
+        if r:
+            to_push = mkts_array[m*limit:]
+
+            # write to db
+            resp = await self.client.write(
+                to_push,
+                tbk=f'{fqsn}/1Sec/OHLCV',
+
+                # NOTE: will will append duplicates
+                # for the same timestamp-index.
+                # TODO: pre deduplicate?
+                isvariablelength=append_and_duplicate,
+            )
+
+            log.info(
+                f'Wrote {mkts_array.size} datums to tsdb\n'
+            )
+
+            for resp in resp.responses:
+                err = resp.error
+                if err:
+                    raise MarketStoreError(err)
+
+
+@acm
+async def open_storage_client(
+    fqsn: str,
+    period: Optional[Union[int, str]] = None,  # in seconds
+
+) -> tuple[Storage, dict[str, np.ndarray]]:
+    '''
+    Load a series by key and deliver in ``numpy`` struct array format.
+
+    '''
+    async with (
+        # eventually a storage backend endpoint
+        get_client() as client,
+    ):
+        # slap on our wrapper api
+        yield Storage(client)
+
+
+async def tsdb_history_update(
+    fqsn: Optional[str] = None,
+
+) -> list[str]:
+
+    # TODO: real-time dedicated task for ensuring
+    # history consistency between the tsdb, shm and real-time feed..
+
+    # update sequence design notes:
+
+    # - load existing highest frequency data from mkts
+    #   * how do we want to offer this to the UI?
+    #    - lazy loading?
+    #    - try to load it all and expect graphics caching/diffing
+    #      to  hide extra bits that aren't in view?
+
+    # - compute the diff between latest data from broker and shm
+    #   * use sql api in mkts to determine where the backend should
+    #     start querying for data?
+    #   * append any diff with new shm length
+    #   * determine missing (gapped) history by scanning
+    #   * how far back do we look?
+
+    # - begin rt update ingest and aggregation
+    #   * could start by always writing ticks to mkts instead of
+    #     worrying about a shm queue for now.
+    #   * we have a short list of shm queues worth groking:
+    #     - https://github.com/pikers/piker/issues/107
+    #   * the original data feed arch blurb:
+    #     - https://github.com/pikers/piker/issues/98
+    #
+    profiler = pg.debug.Profiler(
+        disabled=False,  # not pg_profile_enabled(),
+        delayed=False,
+    )
+
+    async with (
+        open_storage_client(fqsn) as storage,
+
+        maybe_open_feed(
+            [fqsn],
+            start_stream=False,
+
+        ) as (feed, stream),
+    ):
+        profiler(f'opened feed for {fqsn}')
+
+
+        to_append = feed.shm.array
+        to_prepend = None
+
+        if fqsn:
+            symbol = feed.symbols.get(fqsn)
+            if symbol:
+                fqsn = symbol.front_fqsn()
+
+            # diff db history with shm and only write the missing portions
+            ohlcv = feed.shm.array
+
+            # TODO: use pg profiler
+            tsdb_arrays = await storage.read_ohlcv(fqsn)
+            # hist diffing
+            if tsdb_arrays:
+                onesec = tsdb_arrays[1]
+                to_append = ohlcv[ohlcv['time'] > onesec['Epoch'][-1]]
+                to_prepend = ohlcv[ohlcv['time'] < onesec['Epoch'][0]]
+
+            profiler('Finished db arrays diffs')
+
+        syms = await storage.client.list_symbols()
+        log.info(f'Existing tsdb symbol set:\n{pformat(syms)}')
+        profiler(f'listed symbols {syms}')
+
+        # TODO: ask if user wants to write history for detected
+        # available shm buffers?
+        from tractor.trionics import ipython_embed
+        await ipython_embed()
+
+        # for array in [to_append, to_prepend]:
+        #     if array is None:
+        #         continue
+
+        #     log.info(
+        #         f'Writing datums {array.size} -> to tsdb from shm\n'
+        #     )
+        #     await storage.write_ohlcv(fqsn, array)
+
+        # profiler('Finished db writes')
 
 
 async def ingest_quote_stream(
-    symbols: List[str],
+    symbols: list[str],
     brokername: str,
     tries: int = 1,
     loglevel: str = None,
+
 ) -> None:
-    """Ingest a broker quote stream into marketstore in (sampled) tick format.
-    """
-    async with open_feed(
-        brokername,
-        symbols,
-        loglevel=loglevel,
-    ) as (first_quotes, qstream):
+    '''
+    Ingest a broker quote stream into a ``marketstore`` tsdb.
 
-        quote_cache = first_quotes.copy()
+    '''
+    async with (
+        maybe_open_feed(brokername, symbols, loglevel=loglevel) as feed,
+        get_client() as ms_client,
+    ):
+        async for quotes in feed.stream:
+            log.info(quotes)
+            for symbol, quote in quotes.items():
+                for tick in quote.get('ticks', ()):
+                    ticktype = tick.get('type', 'n/a')
 
-        async with get_client() as ms_client:
+            # techtonic tick write
+            array = quote_to_marketstore_structarray({
+                'IsTrade': 1 if ticktype == 'trade' else 0,
+                'IsBid': 1 if ticktype in ('bid', 'bsize') else 0,
+                'Price': tick.get('price'),
+                'Size': tick.get('size')
+            }, last_fill=quote.get('broker_ts', None))
 
-            # start ingest to marketstore
-            async for quotes in qstream:
-                log.info(quotes)
-                for symbol, quote in quotes.items():
+            await ms_client.write(array, _tick_tbk)
 
-                    # remap tick strs to ints
-                    quote['tick'] = _tick_map[quote.get('tick', 'Equal')]
+            # LEGACY WRITE LOOP (using old tick dt)
+            # quote_cache = {
+            #     'size': 0,
+            #     'tick': 0
+            # }
 
-                    # check for volume update (i.e. did trades happen
-                    # since last quote)
-                    new_vol = quote.get('volume', None)
-                    if new_vol is None:
-                        log.debug(f"No fills for {symbol}")
-                        if new_vol == quote_cache.get('volume'):
-                            # should never happen due to field diffing
-                            # on sender side
-                            log.error(
-                                f"{symbol}: got same volume as last quote?")
+            # async for quotes in qstream:
+            #     log.info(quotes)
+            #     for symbol, quote in quotes.items():
 
-                    quote_cache.update(quote)
+            #         # remap tick strs to ints
+            #         quote['tick'] = _tick_map[quote.get('tick', 'Equal')]
 
-                    a = quote_to_marketstore_structarray(
-                        quote,
-                        # TODO: check this closer to the broker query api
-                        last_fill=quote.get('fill_time', '')
-                    )
-                    await ms_client.write(symbol, a)
+            #         # check for volume update (i.e. did trades happen
+            #         # since last quote)
+            #         new_vol = quote.get('volume', None)
+            #         if new_vol is None:
+            #             log.debug(f"No fills for {symbol}")
+            #             if new_vol == quote_cache.get('volume'):
+            #                 # should never happen due to field diffing
+            #                 # on sender side
+            #                 log.error(
+            #                     f"{symbol}: got same volume as last quote?")
+
+            #         quote_cache.update(quote)
+
+            #         a = quote_to_marketstore_structarray(
+            #             quote,
+            #             # TODO: check this closer to the broker query api
+            #             last_fill=quote.get('fill_time', '')
+            #         )
+            #         await ms_client.write(symbol, a)
 
 
 async def stream_quotes(
-    symbols: List[str],
+    symbols: list[str],
     host: str = 'localhost',
     port: int = 5993,
     diff_cached: bool = True,
     loglevel: str = None,
+
 ) -> None:
-    """Open a symbol stream from a running instance of marketstore and
+    '''
+    Open a symbol stream from a running instance of marketstore and
     log to console.
-    """
+
+    '''
     # XXX: required to propagate ``tractor`` loglevel to piker logging
     get_console_log(loglevel or tractor.current_actor().loglevel)
 
-    tbks: Dict[str, str] = {sym: f"{sym}/*/*" for sym in symbols}
+    tbks: dict[str, str] = {sym: f"{sym}/*/*" for sym in symbols}
 
     async with open_websocket_url(f'ws://{host}:{port}/ws') as ws:
         # send subs topics to server
@@ -271,7 +627,7 @@ async def stream_quotes(
         )
         log.info(resp)
 
-        async def recv() -> Dict[str, Any]:
+        async def recv() -> dict[str, Any]:
             return msgpack.loads((await ws.get_message()), encoding='utf-8')
 
         streams = (await recv())['streams']

piker/fsp/_momo.py

@@ -167,6 +167,7 @@ def _wma(
 
     assert length == len(weights)
 
+    # lol, for long sequences this is nutso slow and expensive..
     return np.convolve(signal, weights, 'valid')
 
 

piker/fsp/_volume.py

@@ -309,7 +309,7 @@ async def flow_rates(
 
         if period > 1:
             trade_rate_wma = _wma(
-                dvlm_shm.array['trade_count'],
+                dvlm_shm.array['trade_count'][-period:],
                 period,
                 weights=weights,
             )
@@ -332,7 +332,7 @@ async def flow_rates(
 
         if period > 1:
             dark_trade_rate_wma = _wma(
-                dvlm_shm.array['dark_trade_count'],
+                dvlm_shm.array['dark_trade_count'][-period:],
                 period,
                 weights=weights,
             )

piker/ui/_axes.py

@@ -223,8 +223,9 @@ class DynamicDateAxis(Axis):
     ) -> List[str]:
 
         chart = self.linkedsplits.chart
-        bars = chart._arrays[chart.name]
-        shm = self.linkedsplits.chart._shm
+        flow = chart._flows[chart.name]
+        shm = flow.shm
+        bars = shm.array
         first = shm._first.value
 
         bars_len = len(bars)

piker/ui/_chart.py

@@ -34,9 +34,7 @@ from PyQt5.QtWidgets import (
     QVBoxLayout,
     QSplitter,
 )
-import msgspec
 import numpy as np
-# from pydantic import BaseModel
 import pyqtgraph as pg
 import trio
 
@@ -49,6 +47,7 @@ from ._cursor import (
     Cursor,
     ContentsLabel,
 )
+from ..data._sharedmem import ShmArray
 from ._l1 import L1Labels
 from ._ohlc import BarItems
 from ._curve import FastAppendCurve
@@ -60,15 +59,12 @@ from ._style import (
 )
 from ..data.feed import Feed
 from ..data._source import Symbol
-from ..data._sharedmem import (
-    ShmArray,
-    # _Token,
-)
 from ..log import get_logger
 from ._interaction import ChartView
 from ._forms import FieldsForm
 from .._profile import pg_profile_enabled, ms_slower_then
 from ._overlay import PlotItemOverlay
+from ._flows import Flow
 
 if TYPE_CHECKING:
     from ._display import DisplayState
@@ -419,7 +415,7 @@ class LinkedSplits(QWidget):
         self,
 
         symbol: Symbol,
-        array: np.ndarray,
+        shm: ShmArray,
         sidepane: FieldsForm,
 
         style: str = 'bar',
@@ -444,7 +440,7 @@ class LinkedSplits(QWidget):
         self.chart = self.add_plot(
 
             name=symbol.key,
-            array=array,
+            shm=shm,
             style=style,
             _is_main=True,
 
@@ -472,7 +468,7 @@ class LinkedSplits(QWidget):
         self,
 
         name: str,
-        array: np.ndarray,
+        shm: ShmArray,
 
         array_key: Optional[str] = None,
         style: str = 'line',
@@ -516,7 +512,6 @@ class LinkedSplits(QWidget):
             name=name,
             data_key=array_key or name,
 
-            array=array,
             parent=qframe,
             linkedsplits=self,
             axisItems=axes,
@@ -580,7 +575,7 @@ class LinkedSplits(QWidget):
 
             graphics, data_key = cpw.draw_ohlc(
                 name,
-                array,
+                shm,
                 array_key=array_key
             )
             self.cursor.contents_labels.add_label(
@@ -594,7 +589,7 @@ class LinkedSplits(QWidget):
             add_label = True
             graphics, data_key = cpw.draw_curve(
                 name,
-                array,
+                shm,
                 array_key=array_key,
                 color='default_light',
             )
@@ -603,7 +598,7 @@ class LinkedSplits(QWidget):
             add_label = True
             graphics, data_key = cpw.draw_curve(
                 name,
-                array,
+                shm,
                 array_key=array_key,
                 step_mode=True,
                 color='davies',
@@ -691,7 +686,6 @@ class ChartPlotWidget(pg.PlotWidget):
 
         # the "data view" we generate graphics from
         name: str,
-        array: np.ndarray,
         data_key: str,
         linkedsplits: LinkedSplits,
 
@@ -744,14 +738,6 @@ class ChartPlotWidget(pg.PlotWidget):
         self._max_l1_line_len: float = 0
 
         # self.setViewportMargins(0, 0, 0, 0)
-        # self._ohlc = array  # readonly view of ohlc data
-
-        # TODO: move to Aggr above XD
-        # readonly view of data arrays
-        self._arrays = {
-            self.data_key: array,
-        }
-        self._graphics = {}  # registry of underlying graphics
 
         # registry of overlay curve names
         self._flows: dict[str, Flow] = {}
@@ -767,7 +753,6 @@ class ChartPlotWidget(pg.PlotWidget):
         # show background grid
         self.showGrid(x=False, y=True, alpha=0.3)
 
-        self.default_view()
         self.cv.enable_auto_yrange()
 
         self.pi_overlay: PlotItemOverlay = PlotItemOverlay(self.plotItem)
@@ -816,14 +801,8 @@ class ChartPlotWidget(pg.PlotWidget):
         Return a range tuple for the bars present in view.
 
         '''
-        l, r = self.view_range()
-        array = self._arrays[self.name]
-        start, stop = self._xrange = (
-            array[0]['index'],
-            array[-1]['index'],
-        )
-        lbar = max(l, start)
-        rbar = min(r, stop)
+        main_flow = self._flows[self.name]
+        ifirst, l, lbar, rbar, r, ilast = main_flow.datums_range()
         return l, lbar, rbar, r
 
     def curve_width_pxs(
@@ -877,40 +856,51 @@ class ChartPlotWidget(pg.PlotWidget):
 
     def default_view(
         self,
-        steps_on_screen: Optional[int] = None
+        bars_from_y: int = 3000,
 
     ) -> None:
         '''
         Set the view box to the "default" startup view of the scene.
 
         '''
-        try:
-            index = self._arrays[self.name]['index']
-        except IndexError:
-            log.warning(f'array for {self.name} not loaded yet?')
+        flow = self._flows.get(self.name)
+        if not flow:
+            log.warning(f'`Flow` for {self.name} not loaded yet?')
             return
 
+        index = flow.shm.array['index']
         xfirst, xlast = index[0], index[-1]
         l, lbar, rbar, r = self.bars_range()
-
-        marker_pos, l1_len = self.pre_l1_xs()
-        end = xlast + l1_len + 1
+        view = self.view
 
         if (
             rbar < 0
             or l < xfirst
+            or l < 0
             or (rbar - lbar) < 6
         ):
-            # set fixed bars count on screen that approx includes as
+            # TODO: set fixed bars count on screen that approx includes as
             # many bars as possible before a downsample line is shown.
-            begin = xlast - round(6116 / 6)
+            begin = xlast - bars_from_y
+            view.setXRange(
+                min=begin,
+                max=xlast,
+                padding=0,
+            )
+            # re-get range
+            l, lbar, rbar, r = self.bars_range()
 
-        else:
-            begin = end - (r - l)
+        # we get the L1 spread label "length" in view coords
+        # terms now that we've scaled either by user control
+        # or to the default set of bars as per the immediate block
+        # above.
+        marker_pos, l1_len = self.pre_l1_xs()
+        end = xlast + l1_len + 1
+        begin = end - (r - l)
 
         # for debugging
         # print(
-        #     f'bars range: {brange}\n'
+        #     # f'bars range: {brange}\n'
         #     f'xlast: {xlast}\n'
         #     f'marker pos: {marker_pos}\n'
         #     f'l1 len: {l1_len}\n'
@@ -922,14 +912,13 @@ class ChartPlotWidget(pg.PlotWidget):
         if self._static_yrange == 'axis':
             self._static_yrange = None
 
-        view = self.view
         view.setXRange(
             min=begin,
             max=end,
             padding=0,
         )
-        view._set_yrange()
         self.view.maybe_downsample_graphics()
+        view._set_yrange()
         try:
             self.linked.graphics_cycle()
         except IndexError:
@@ -960,7 +949,7 @@ class ChartPlotWidget(pg.PlotWidget):
     def draw_ohlc(
         self,
         name: str,
-        data: np.ndarray,
+        shm: ShmArray,
 
         array_key: Optional[str] = None,
 
@@ -980,19 +969,21 @@ class ChartPlotWidget(pg.PlotWidget):
         # the np array buffer to be drawn on next render cycle
         self.plotItem.addItem(graphics)
 
-        # draw after to allow self.scene() to work...
-        graphics.draw_from_data(data)
-
         data_key = array_key or name
-        self._graphics[data_key] = graphics
 
         self._flows[data_key] = Flow(
             name=name,
             plot=self.plotItem,
+            _shm=shm,
             is_ohlc=True,
             graphics=graphics,
         )
 
+        # TODO: i think we can eventually remove this if
+        # we write the ``Flow.update_graphics()`` method right?
+        # draw after to allow self.scene() to work...
+        graphics.draw_from_data(shm.array)
+
         self._add_sticky(name, bg_color='davies')
 
         return graphics, data_key
@@ -1058,7 +1049,7 @@ class ChartPlotWidget(pg.PlotWidget):
         self,
 
         name: str,
-        data: np.ndarray,
+        shm: ShmArray,
 
         array_key: Optional[str] = None,
         overlay: bool = False,
@@ -1071,7 +1062,7 @@ class ChartPlotWidget(pg.PlotWidget):
     ) -> (pg.PlotDataItem, str):
         '''
         Draw a "curve" (line plot graphics) for the provided data in
-        the input array ``data``.
+        the input shm array ``shm``.
 
         '''
         color = color or self.pen_color or 'default_light'
@@ -1082,6 +1073,7 @@ class ChartPlotWidget(pg.PlotWidget):
         data_key = array_key or name
 
         # yah, we wrote our own B)
+        data = shm.array
         curve = FastAppendCurve(
             y=data[data_key],
             x=data['index'],
@@ -1105,16 +1097,14 @@ class ChartPlotWidget(pg.PlotWidget):
         # and is disastrous for performance.
         # curve.setCacheMode(QtWidgets.QGraphicsItem.ItemCoordinateCache)
 
-        # register curve graphics and backing array for name
-        self._graphics[name] = curve
-        self._arrays[data_key] = data
-
         pi = pi or self.plotItem
 
         self._flows[data_key] = Flow(
             name=name,
             plot=pi,
+            _shm=shm,
             is_ohlc=False,
+            # register curve graphics with this flow
             graphics=curve,
         )
 
@@ -1175,16 +1165,11 @@ class ChartPlotWidget(pg.PlotWidget):
         )
         return last
 
-    def update_graphics_from_array(
+    def update_graphics_from_flow(
         self,
         graphics_name: str,
-
-        array: Optional[np.ndarray] = None,
         array_key: Optional[str] = None,
 
-        use_vr: bool = True,
-        render: bool = True,
-
         **kwargs,
 
     ) -> pg.GraphicsObject:
@@ -1192,63 +1177,11 @@ class ChartPlotWidget(pg.PlotWidget):
         Update the named internal graphics from ``array``.
 
         '''
-        if array is not None:
-            assert len(array)
-
-        data_key = array_key or graphics_name
-        if graphics_name not in self._flows:
-            data_key = self.name
-
-        if array is not None:
-            # write array to internal graphics table
-            self._arrays[data_key] = array
-        else:
-            array = self._arrays[data_key]
-
-        # array key and graphics "name" might be different..
-        graphics = self._graphics[graphics_name]
-
-        # compute "in-view" indices
-        l, lbar, rbar, r = self.bars_range()
-        indexes = array['index']
-        ifirst = indexes[0]
-        ilast = indexes[-1]
-
-        lbar_i = max(l, ifirst) - ifirst
-        rbar_i = min(r, ilast) - ifirst
-
-        # TODO: we could do it this way as well no?
-        # to_draw = array[lbar - ifirst:(rbar - ifirst) + 1]
-        in_view = array[lbar_i: rbar_i + 1]
-
-        if (
-            not in_view.size
-            or not render
-        ):
-            return graphics
-
-        if isinstance(graphics, BarItems):
-            graphics.update_from_array(
-                array,
-                in_view,
-                view_range=(lbar_i, rbar_i) if use_vr else None,
-
-                **kwargs,
-            )
-
-        else:
-            graphics.update_from_array(
-                x=array['index'],
-                y=array[data_key],
-
-                x_iv=in_view['index'],
-                y_iv=in_view[data_key],
-                view_range=(lbar_i, rbar_i) if use_vr else None,
-
-                **kwargs
-            )
-
-        return graphics
+        flow = self._flows[array_key or graphics_name]
+        return flow.update_graphics(
+            array_key=array_key,
+            **kwargs,
+        )
 
     # def _label_h(self, yhigh: float, ylow: float) -> float:
     #     # compute contents label "height" in view terms
@@ -1295,7 +1228,7 @@ class ChartPlotWidget(pg.PlotWidget):
 
         # TODO: this should go onto some sort of
         # data-view thinger..right?
-        ohlc = self._shm.array
+        ohlc = self._flows[self.name].shm.array
 
         # XXX: not sure why the time is so off here
         # looks like we're gonna have to do some fixing..
@@ -1341,9 +1274,6 @@ class ChartPlotWidget(pg.PlotWidget):
             delayed=True,
         )
 
-        l, lbar, rbar, r = bars_range or self.bars_range()
-        profiler(f'{self.name} got bars range')
-
         # TODO: here we should instead look up the ``Flow.shm.array``
         # and read directly from shm to avoid copying to memory first
         # and then reading it again here.
@@ -1353,112 +1283,19 @@ class ChartPlotWidget(pg.PlotWidget):
             flow is None
         ):
             log.error(f"flow {flow_key} doesn't exist in chart {self.name} !?")
-            res = 0, 0
+            key = res = 0, 0
 
         else:
+            first, l, lbar, rbar, r, last = bars_range or flow.datums_range()
+            profiler(f'{self.name} got bars range')
+
             key = round(lbar), round(rbar)
             res = flow.maxmin(*key)
-            profiler(f'yrange mxmn: {key} -> {res}')
             if res == (None, None):
                 log.error(
                     f"{flow_key} no mxmn for bars_range => {key} !?"
                 )
                 res = 0, 0
 
+        profiler(f'yrange mxmn: {key} -> {res}')
         return res
-
-
-# class FlowsTable(pydantic.BaseModel):
-#     '''
-#     Data-AGGRegate: high level API onto multiple (categorized)
-#     ``Flow``s with high level processing routines for
-#     multi-graphics computations and display.
-
-#     '''
-#     flows: dict[str, np.ndarray] = {}
-
-
-class Flow(msgspec.Struct):  # , frozen=True):
-    '''
-    (FinancialSignal-)Flow compound type which wraps a real-time
-    graphics (curve) and its backing data stream together for high level
-    access and control.
-
-    The intention is for this type to eventually be capable of shm-passing
-    of incrementally updated graphics stream data between actors.
-
-    '''
-    name: str
-    plot: pg.PlotItem
-    is_ohlc: bool = False
-    graphics: pg.GraphicsObject
-
-    # TODO: hackery to be able to set a shm later
-    # but whilst also allowing this type to hashable,
-    # likely will require serializable token that is used to attach
-    # to the underlying shm ref after startup?
-    _shm: Optional[ShmArray] = None  # currently, may be filled in "later"
-
-    # cache of y-range values per x-range input.
-    _mxmns: dict[tuple[int, int], tuple[float, float]] = {}
-
-    @property
-    def shm(self) -> ShmArray:
-        return self._shm
-
-    @shm.setter
-    def shm(self, shm: ShmArray) -> ShmArray:
-        self._shm = shm
-
-    def maxmin(
-        self,
-        lbar,
-        rbar,
-
-    ) -> tuple[float, float]:
-        '''
-        Compute the cached max and min y-range values for a given
-        x-range determined by ``lbar`` and ``rbar``.
-
-        '''
-        rkey = (lbar, rbar)
-        cached_result = self._mxmns.get(rkey)
-        if cached_result:
-            return cached_result
-
-        shm = self.shm
-        if shm is None:
-            mxmn = None
-
-        else:  # new block for profiling?..
-            arr = shm.array
-
-            # build relative indexes into shm array
-            # TODO: should we just add/use a method
-            # on the shm to do this?
-            ifirst = arr[0]['index']
-            slice_view = arr[
-                lbar - ifirst:
-                (rbar - ifirst) + 1
-            ]
-
-            if not slice_view.size:
-                mxmn = None
-
-            else:
-                if self.is_ohlc:
-                    ylow = np.min(slice_view['low'])
-                    yhigh = np.max(slice_view['high'])
-
-                else:
-                    view = slice_view[self.name]
-                    ylow = np.min(view)
-                    yhigh = np.max(view)
-
-                mxmn = ylow, yhigh
-
-            if mxmn is not None:
-                # cache new mxmn result
-                self._mxmns[rkey] = mxmn
-
-            return mxmn

piker/ui/_compression.py

@@ -162,7 +162,7 @@ def ohlc_to_m4_line(
             flat,
             px_width=px_width,
             uppx=uppx,
-            log_scale=bool(uppx)
+            # log_scale=bool(uppx)
         )
         x = np.broadcast_to(x[:, None], y.shape)
         x = (x + np.array([-0.43, 0, 0, 0.43])).flatten()
@@ -181,7 +181,8 @@ def ds_m4(
     # in display-device-local pixel units.
     px_width: int,
     uppx: Optional[float] = None,
-    log_scale: bool = True,
+    xrange: Optional[float] = None,
+    # log_scale: bool = True,
 
 ) -> tuple[int, np.ndarray, np.ndarray]:
     '''
@@ -210,50 +211,77 @@ def ds_m4(
 
     # optionally log-scale down the "supposed pxs on screen"
     # as the units-per-px (uppx) get's large.
-    if log_scale:
-        assert uppx, 'You must provide a `uppx` value to use log scaling!'
+    # if log_scale:
+    #     assert uppx, 'You must provide a `uppx` value to use log scaling!'
+    #     # uppx = uppx * math.log(uppx, 2)
 
-        # scaler = 2**7 / (1 + math.log(uppx, 2))
-        scaler = round(
-            max(
-                # NOTE: found that a 16x px width brought greater
-                # detail, likely due to dpi scaling?
-                # px_width=px_width * 16,
-                2**7 / (1 + math.log(uppx, 2)),
-                1
-            )
-        )
-        px_width *= scaler
+    #     # scaler = 2**7 / (1 + math.log(uppx, 2))
+    #     scaler = round(
+    #         max(
+    #             # NOTE: found that a 16x px width brought greater
+    #             # detail, likely due to dpi scaling?
+    #             # px_width=px_width * 16,
+    #             2**7 / (1 + math.log(uppx, 2)),
+    #             1
+    #         )
+    #     )
+    #     px_width *= scaler
 
-    assert px_width > 1  # width of screen in pxs?
+    # else:
+    #     px_width *= 16
+
+    # should never get called unless actually needed
+    assert px_width > 1 and uppx > 0
 
     # NOTE: if we didn't pre-slice the data to downsample
     # you could in theory pass these as the slicing params,
     # do we care though since we can always just pre-slice the
     # input?
     x_start = x[0]  # x value start/lowest in domain
-    x_end = x[-1]  # x end value/highest in domain
+
+    if xrange is None:
+        x_end = x[-1]  # x end value/highest in domain
+        xrange = (x_end - x_start)
 
     # XXX: always round up on the input pixels
-    px_width = math.ceil(px_width)
+    # lnx = len(x)
+    # uppx *= max(4 / (1 + math.log(uppx, 2)), 1)
 
-    x_range = x_end - x_start
+    pxw = math.ceil(xrange / uppx)
+    # px_width = math.ceil(px_width)
 
     # ratio of indexed x-value to width of raster in pixels.
     # this is more or less, uppx: units-per-pixel.
-    w = x_range / float(px_width)
+    # w = xrange / float(px_width)
+    # uppx = uppx * math.log(uppx, 2)
+    # w2 = px_width / uppx
+
+    # scale up the width as the uppx get's large
+    w = uppx  # * math.log(uppx, 666)
 
     # ensure we make more then enough
     # frames (windows) for the output pixel
-    frames = px_width
+    frames = pxw
 
     # if we have more and then exact integer's
     # (uniform quotient output) worth of datum-domain-points
     # per windows-frame, add one more window to ensure
     # we have room for all output down-samples.
-    pts_per_pixel, r = divmod(len(x), frames)
+    pts_per_pixel, r = divmod(xrange, frames)
     if r:
+        # while r:
         frames += 1
+        pts_per_pixel, r = divmod(xrange, frames)
+
+    # print(
+    #     f'uppx: {uppx}\n'
+    #     f'xrange: {xrange}\n'
+    #     f'px_width: {px_width}\n'
+    #     f'pxw: {pxw}\n'
+    #     f'WTF w:{w}, w2:{w2}\n'
+    #     f'frames: {frames}\n'
+    # )
+    assert frames >= (xrange / uppx)
 
     # call into ``numba``
     nb, i_win, y_out = _m4(

piker/ui/_cursor.py

@@ -43,8 +43,8 @@ log = get_logger(__name__)
 # latency (in terms of perceived lag in cross hair) so really be sure
 # there's an improvement if you want to change it!
 
-_mouse_rate_limit = 120  # TODO; should we calc current screen refresh rate?
-_debounce_delay = 1 / 40
+_mouse_rate_limit = 60  # TODO; should we calc current screen refresh rate?
+_debounce_delay = 0
 _ch_label_opac = 1
 
 
@@ -191,6 +191,9 @@ class ContentsLabel(pg.LabelItem):
 
         self.setText(
             "<b>i</b>:{index}<br/>"
+            # NB: these fields must be indexed in the correct order via
+            # the slice syntax below.
+            "<b>epoch</b>:{}<br/>"
             "<b>O</b>:{}<br/>"
             "<b>H</b>:{}<br/>"
             "<b>L</b>:{}<br/>"
@@ -198,7 +201,15 @@ class ContentsLabel(pg.LabelItem):
             "<b>V</b>:{}<br/>"
             "<b>wap</b>:{}".format(
                 *array[index - first][
-                    ['open', 'high', 'low', 'close', 'volume', 'bar_wap']
+                    [
+                        'time',
+                        'open',
+                        'high',
+                        'low',
+                        'close',
+                        'volume',
+                        'bar_wap',
+                    ]
                 ],
                 name=name,
                 index=index,
@@ -243,13 +254,13 @@ class ContentsLabels:
     def update_labels(
         self,
         index: int,
-        # array_name: str,
 
     ) -> None:
-        # for name, (label, update) in self._labels.items():
         for chart, name, label, update in self._labels:
 
-            array = chart._arrays[name]
+            flow = chart._flows[name]
+            array = flow.shm.array
+
             if not (
                 index >= 0
                 and index < array[-1]['index']
@@ -258,8 +269,6 @@ class ContentsLabels:
                 print('WTF out of range?')
                 continue
 
-            # array = chart._arrays[name]
-
             # call provided update func with data point
             try:
                 label.show()
@@ -295,7 +304,8 @@ class ContentsLabels:
 
 
 class Cursor(pg.GraphicsObject):
-    '''Multi-plot cursor for use on a ``LinkedSplits`` chart (set).
+    '''
+    Multi-plot cursor for use on a ``LinkedSplits`` chart (set).
 
     '''
     def __init__(
@@ -310,7 +320,7 @@ class Cursor(pg.GraphicsObject):
 
         self.linked = linkedsplits
         self.graphics: dict[str, pg.GraphicsObject] = {}
-        self.plots: List['PlotChartWidget'] = []  # type: ignore # noqa
+        self.plots: list['PlotChartWidget'] = []  # type: ignore # noqa
         self.active_plot = None
         self.digits: int = digits
         self._datum_xy: tuple[int, float] = (0, 0)
@@ -439,7 +449,10 @@ class Cursor(pg.GraphicsObject):
         if plot.linked.xaxis_chart is plot:
             xlabel = self.xaxis_label = XAxisLabel(
                 parent=self.plots[plot_index].getAxis('bottom'),
-                # parent=self.plots[plot_index].pi_overlay.get_axis(plot.plotItem, 'bottom'),
+                # parent=self.plots[plot_index].pi_overlay.get_axis(
+                #     plot.plotItem, 'bottom'
+                # ),
+
                 opacity=_ch_label_opac,
                 bg_color=self.label_color,
             )
@@ -457,9 +470,12 @@ class Cursor(pg.GraphicsObject):
     ) -> LineDot:
         # if this plot contains curves add line dot "cursors" to denote
         # the current sample under the mouse
+        main_flow = plot._flows[plot.name]
+        # read out last index
+        i = main_flow.shm.array[-1]['index']
         cursor = LineDot(
             curve,
-            index=plot._arrays[plot.name][-1]['index'],
+            index=i,
             plot=plot
         )
         plot.addItem(cursor)

piker/ui/_curve.py

@@ -44,55 +44,78 @@ from ..log import get_logger
 log = get_logger(__name__)
 
 
-def step_path_arrays_from_1d(
-    x: np.ndarray,
-    y: np.ndarray,
-    include_endpoints: bool = False,
+# TODO: numba this instead..
+# def step_path_arrays_from_1d(
+#     x: np.ndarray,
+#     y: np.ndarray,
+#     include_endpoints: bool = True,
 
-) -> (np.ndarray, np.ndarray):
-    '''
-    Generate a "step mode" curve aligned with OHLC style bars
-    such that each segment spans each bar (aka "centered" style).
+# ) -> (np.ndarray, np.ndarray):
+#     '''
+#     Generate a "step mode" curve aligned with OHLC style bars
+#     such that each segment spans each bar (aka "centered" style).
 
-    '''
-    y_out = y.copy()
-    x_out = x.copy()
-    x2 = np.empty(
-        # the data + 2 endpoints on either end for
-        # "termination of the path".
-        (len(x) + 1, 2),
-        # we want to align with OHLC or other sampling style
-        # bars likely so we need fractinal values
-        dtype=float,
-    )
-    x2[0] = x[0] - 0.5
-    x2[1] = x[0] + 0.5
-    x2[1:] = x[:, np.newaxis] + 0.5
+#     '''
+#     # y_out = y.copy()
+#     # x_out = x.copy()
 
-    # flatten to 1-d
-    x_out = x2.reshape(x2.size)
+#     # x2 = np.empty(
+#     #     # the data + 2 endpoints on either end for
+#     #     # "termination of the path".
+#     #     (len(x) + 1, 2),
+#     #     # we want to align with OHLC or other sampling style
+#     #     # bars likely so we need fractinal values
+#     #     dtype=float,
+#     # )
 
-    # we create a 1d with 2 extra indexes to
-    # hold the start and (current) end value for the steps
-    # on either end
-    y2 = np.empty((len(y), 2), dtype=y.dtype)
-    y2[:] = y[:, np.newaxis]
+#     x2 = np.broadcast_to(
+#         x[:, None],
+#         (
+#             x.size + 1,
+#             # 4,  # only ohlc
+#             2,
+#         ),
+#     ) + np.array([-0.5, 0.5])
 
-    y_out = np.empty(
-        2*len(y) + 2,
-        dtype=y.dtype
-    )
+#     # x2[0] = x[0] - 0.5
+#     # x2[1] = x[0] + 0.5
+#     # x2[0, 0] = x[0] - 0.5
+#     # x2[0, 1] = x[0] + 0.5
+#     # x2[1:] = x[:, np.newaxis] + 0.5
+#     # import pdbpp
+#     # pdbpp.set_trace()
 
-    # flatten and set 0 endpoints
-    y_out[1:-1] = y2.reshape(y2.size)
-    y_out[0] = 0
-    y_out[-1] = 0
+#     # flatten to 1-d
+#     # x_out = x2.reshape(x2.size)
+#     # x_out = x2
 
-    if not include_endpoints:
-        return x_out[:-1], y_out[:-1]
+#     # we create a 1d with 2 extra indexes to
+#     # hold the start and (current) end value for the steps
+#     # on either end
+#     y2 = np.empty(
+#         (len(y) + 1, 2),
+#         dtype=y.dtype,
+#     )
+#     y2[:] = y[:, np.newaxis]
+#     # y2[-1] = 0
 
-    else:
-        return x_out, y_out
+#     # y_out = y2
+
+# #     y_out = np.empty(
+# #         2*len(y) + 2,
+# #         dtype=y.dtype
+# #     )
+
+#     # flatten and set 0 endpoints
+#     # y_out[1:-1] = y2.reshape(y2.size)
+#     # y_out[0] = 0
+#     # y_out[-1] = 0
+
+#     if not include_endpoints:
+#         return x2[:-1], y2[:-1]
+
+#     else:
+#         return x2, y2
 
 
 _line_styles: dict[str, int] = {
@@ -119,8 +142,8 @@ class FastAppendCurve(pg.GraphicsObject):
     def __init__(
         self,
 
-        x: np.ndarray,
-        y: np.ndarray,
+        x: np.ndarray = None,
+        y: np.ndarray = None,
         *args,
 
         step_mode: bool = False,
@@ -137,6 +160,9 @@ class FastAppendCurve(pg.GraphicsObject):
         # brutaaalll, see comments within..
         self._y = self.yData = y
         self._x = self.xData = x
+        self._vr: Optional[tuple] = None
+        self._avr: Optional[tuple] = None
+        self._br = None
 
         self._name = name
         self.path: Optional[QtGui.QPainterPath] = None
@@ -150,6 +176,7 @@ class FastAppendCurve(pg.GraphicsObject):
 
         # self._xrange: tuple[int, int] = self.dataBounds(ax=0)
         self._xrange: Optional[tuple[int, int]] = None
+        # self._x_iv_range = None
 
         # self._last_draw = time.time()
         self._in_ds: bool = False
@@ -181,16 +208,12 @@ class FastAppendCurve(pg.GraphicsObject):
         # interactions slower (such as zooming) and if so maybe if/when
         # we implement a "history" mode for the view we disable this in
         # that mode?
-        if step_mode:
-            # don't enable caching by default for the case where the
-            # only thing drawn is the "last" line segment which can
-            # have a weird artifact where it won't be fully drawn to its
-            # endpoint (something we saw on trade rate curves)
-            self.setCacheMode(
-                QGraphicsItem.DeviceCoordinateCache
-            )
-
-        self.update()
+        # if step_mode:
+        # don't enable caching by default for the case where the
+        # only thing drawn is the "last" line segment which can
+        # have a weird artifact where it won't be fully drawn to its
+        # endpoint (something we saw on trade rate curves)
+        self.setCacheMode(QGraphicsItem.DeviceCoordinateCache)
 
     # TODO: probably stick this in a new parent
     # type which will contain our own version of
@@ -242,7 +265,7 @@ class FastAppendCurve(pg.GraphicsObject):
             y,
             px_width=px_width,
             uppx=uppx,
-            log_scale=bool(uppx)
+            # log_scale=bool(uppx)
         )
         x = np.broadcast_to(x[:, None], y.shape)
         # x = (x + np.array([-0.43, 0, 0, 0.43])).flatten()
@@ -266,6 +289,10 @@ class FastAppendCurve(pg.GraphicsObject):
 
         view_range: Optional[tuple[int, int]] = None,
         profiler: Optional[pg.debug.Profiler] = None,
+        draw_last: bool = True,
+        slice_to_head: int = -1,
+        do_append: bool = True,
+        should_redraw: bool = False,
 
     ) -> QtGui.QPainterPath:
         '''
@@ -280,78 +307,137 @@ class FastAppendCurve(pg.GraphicsObject):
             disabled=not pg_profile_enabled(),
             gt=ms_slower_then,
         )
-        # flip_cache = False
+        flip_cache = False
 
         if self._xrange:
             istart, istop = self._xrange
         else:
             self._xrange = istart, istop = x[0], x[-1]
+
+        # compute the length diffs between the first/last index entry in
+        # the input data and the last indexes we have on record from the
+        # last time we updated the curve index.
+        prepend_length = int(istart - x[0])
+        append_length = int(x[-1] - istop)
+
+        # this is the diff-mode, "data"-rendered index
+        # tracking var..
+        self._xrange = x[0], x[-1]
+
         # print(f"xrange: {self._xrange}")
 
         # XXX: lol brutal, the internals of `CurvePoint` (inherited by
         # our `LineDot`) required ``.getData()`` to work..
-        self.xData = x
-        self.yData = y
-        self._x, self._y = x, y
-
-        if view_range:
-            profiler(f'view range slice {view_range}')
+        # self.xData = x
+        # self.yData = y
+        # self._x, self._y = x, y
 
         # downsampling incremental state checking
         uppx = self.x_uppx()
         px_width = self.px_width()
         uppx_diff = (uppx - self._last_uppx)
 
-        should_ds = False
-        should_redraw = False
+        new_sample_rate = False
+        should_ds = self._in_ds
+        showing_src_data = self._in_ds
+        # should_redraw = False
 
         # if a view range is passed, plan to draw the
         # source ouput that's "in view" of the chart.
-        if view_range and not self._in_ds:
+        if (
+            view_range
+            # and not self._in_ds
+            # and not prepend_length > 0
+        ):
             # print(f'{self._name} vr: {view_range}')
 
             # by default we only pull data up to the last (current) index
-            x_out, y_out = x_iv[:-1], y_iv[:-1]
+            x_out, y_out = x_iv[:slice_to_head], y_iv[:slice_to_head]
+            profiler(f'view range slice {view_range}')
 
-            # step mode: draw flat top discrete "step"
-            # over the index space for each datum.
-            if self._step_mode:
-                # TODO: numba this bish
-                x_out, y_out = step_path_arrays_from_1d(
-                    x_out,
-                    y_out
+            vl, vr = view_range
+
+            # last_ivr = self._x_iv_range
+            # ix_iv, iy_iv = self._x_iv_range = (x_iv[0], x_iv[-1])
+
+            zoom_or_append = False
+            last_vr = self._vr
+            last_ivr = self._avr
+
+            if last_vr:
+                # relative slice indices
+                lvl, lvr = last_vr
+                # abs slice indices
+                al, ar = last_ivr
+
+                # append_length = int(x[-1] - istop)
+                # append_length = int(x_iv[-1] - ar)
+
+                # left_change = abs(x_iv[0] - al) >= 1
+                # right_change = abs(x_iv[-1] - ar) >= 1
+
+                if (
+                    # likely a zoom view change
+                    (vr - lvr) > 2 or vl < lvl
+                    # append / prepend update
+                    # we had an append update where the view range
+                    # didn't change but the data-viewed (shifted)
+                    # underneath, so we need to redraw.
+                    # or left_change and right_change and last_vr == view_range
+
+                        # not (left_change and right_change) and ivr
+                    # (
+                    # or abs(x_iv[ivr] - livr) > 1
+                ):
+                    zoom_or_append = True
+
+            # if last_ivr:
+            #     liivl, liivr = last_ivr
+
+            if (
+                view_range != last_vr
+                and (
+                    append_length > 1
+                    or zoom_or_append
                 )
-                profiler('generated step arrays')
+            ):
+                should_redraw = True
+                # print("REDRAWING BRUH")
 
-            should_redraw = True
-            profiler('sliced in-view array history')
+            self._vr = view_range
+            self._avr = x_iv[0], x_iv[slice_to_head]
 
             # x_last = x_iv[-1]
             # y_last = y_iv[-1]
-            self._last_vr = view_range
+            # self._last_vr = view_range
 
             # self.disable_cache()
             # flip_cache = True
 
         else:
-            self._xrange = x[0], x[-1]
+            # if (
+            #     not view_range
+            #     or self._in_ds
+            # ):
+            # by default we only pull data up to the last (current) index
+            x_out, y_out = x[:slice_to_head], y[:slice_to_head]
 
-        x_last = x[-1]
-        y_last = y[-1]
+            if prepend_length > 0:
+                should_redraw = True
 
         # check for downsampling conditions
         if (
             # std m4 downsample conditions
             px_width
-            and uppx_diff >= 4
-            or uppx_diff <= -3
-            or self._step_mode and abs(uppx_diff) >= 4
-
+            and abs(uppx_diff) >= 1
         ):
             log.info(
                 f'{self._name} sampler change: {self._last_uppx} -> {uppx}'
             )
             self._last_uppx = uppx
+            new_sample_rate = True
+            showing_src_data = False
+            should_redraw = True
             should_ds = True
 
         elif (
@@ -362,53 +448,54 @@ class FastAppendCurve(pg.GraphicsObject):
             # source data so we clear our path data in prep
             # to generate a new one from original source data.
             should_redraw = True
+            new_sample_rate = True
             should_ds = False
-
-        # compute the length diffs between the first/last index entry in
-        # the input data and the last indexes we have on record from the
-        # last time we updated the curve index.
-        prepend_length = int(istart - x[0])
-        append_length = int(x[-1] - istop)
+            showing_src_data = True
 
         # no_path_yet = self.path is None
         if (
             self.path is None
             or should_redraw
-            or should_ds
+            or new_sample_rate
             or prepend_length > 0
         ):
-            if (
-                not view_range
-                or self._in_ds
-            ):
-                # by default we only pull data up to the last (current) index
-                x_out, y_out = x[:-1], y[:-1]
+            # if (
+            #     not view_range
+            #     or self._in_ds
+            # ):
+            #     # by default we only pull data up to the last (current) index
+            #     x_out, y_out = x[:-1], y[:-1]
 
-                # step mode: draw flat top discrete "step"
-                # over the index space for each datum.
-                if self._step_mode:
-                    x_out, y_out = step_path_arrays_from_1d(
-                        x_out,
-                        y_out,
-                    )
-                    # TODO: numba this bish
-                    profiler('generated step arrays')
+            # step mode: draw flat top discrete "step"
+            # over the index space for each datum.
+            # if self._step_mode:
+            #     self.disable_cache()
+            #     flip_cache = True
+            #     x_out, y_out = step_path_arrays_from_1d(
+            #         x_out,
+            #         y_out,
+            #     )
+
+            #     # TODO: numba this bish
+            #     profiler('generated step arrays')
 
             if should_redraw:
-                profiler('path reversion to non-ds')
                 if self.path:
+                    # print(f'CLEARING PATH {self._name}')
                     self.path.clear()
 
                 if self.fast_path:
                     self.fast_path.clear()
 
-            if should_redraw and not should_ds:
-                if self._in_ds:
-                    log.info(f'DEDOWN -> {self._name}')
+                profiler('cleared paths due to `should_redraw` set')
+
+            if new_sample_rate and showing_src_data:
+                # if self._in_ds:
+                log.info(f'DEDOWN -> {self._name}')
 
                 self._in_ds = False
 
-            elif should_ds and px_width:
+            elif should_ds and uppx and px_width > 1:
                 x_out, y_out = self.downsample(
                     x_out,
                     y_out,
@@ -425,7 +512,10 @@ class FastAppendCurve(pg.GraphicsObject):
                 finiteCheck=False,
                 path=self.path,
             )
-            profiler('generated fresh path')
+            self.prepareGeometryChange()
+            profiler(
+                f'generated fresh path. (should_redraw: {should_redraw} should_ds: {should_ds} new_sample_rate: {new_sample_rate})'
+            )
             # profiler(f'DRAW PATH IN VIEW -> {self._name}')
 
             # reserve mem allocs see:
@@ -457,32 +547,44 @@ class FastAppendCurve(pg.GraphicsObject):
 
         elif (
             append_length > 0
-            and not view_range
+            and do_append
+            and not should_redraw
+            # and not view_range
         ):
-            new_x = x[-append_length - 2:-1]
-            new_y = y[-append_length - 2:-1]
+            print(f'{self._name} append len: {append_length}')
+            new_x = x[-append_length - 2:slice_to_head]
+            new_y = y[-append_length - 2:slice_to_head]
+            profiler('sliced append path')
 
-            if self._step_mode:
-                new_x, new_y = step_path_arrays_from_1d(
-                    new_x,
-                    new_y,
-                )
-                # [1:] since we don't need the vertical line normally at
-                # the beginning of the step curve taking the first (x,
-                # y) poing down to the x-axis **because** this is an
-                # appended path graphic.
-                new_x = new_x[1:]
-                new_y = new_y[1:]
+            # if self._step_mode:
+            # #     new_x, new_y = step_path_arrays_from_1d(
+            # #         new_x,
+            # #         new_y,
+            # #     )
+            # #     # [1:] since we don't need the vertical line normally at
+            # #     # the beginning of the step curve taking the first (x,
+            # #     # y) poing down to the x-axis **because** this is an
+            # #     # appended path graphic.
+            # #     new_x = new_x[1:]
+            # #     new_y = new_y[1:]
 
-            profiler('diffed append arrays')
+            #     self.disable_cache()
+            #     flip_cache = True
 
-            if should_ds:
-                new_x, new_y = self.downsample(
-                    new_x,
-                    new_y,
-                    **should_ds,
-                )
-                profiler(f'fast path downsample redraw={should_ds}')
+            #     profiler('generated step data')
+
+            profiler(
+                f'diffed array input, append_length={append_length}'
+            )
+
+            # if should_ds:
+            #     new_x, new_y = self.downsample(
+            #         new_x,
+            #         new_y,
+            #         px_width,
+            #         uppx,
+            #     )
+            #     profiler(f'fast path downsample redraw={should_ds}')
 
             append_path = pg.functions.arrayToQPath(
                 new_x,
@@ -491,12 +593,13 @@ class FastAppendCurve(pg.GraphicsObject):
                 finiteCheck=False,
                 path=self.fast_path,
             )
+            profiler('generated append qpath')
 
             if self.use_fpath:
                 # an attempt at trying to make append-updates faster..
                 if self.fast_path is None:
                     self.fast_path = append_path
-                    self.fast_path.reserve(int(6e3))
+                    # self.fast_path.reserve(int(6e3))
                 else:
                     self.fast_path.connectPath(append_path)
                     size = self.fast_path.capacity()
@@ -529,16 +632,43 @@ class FastAppendCurve(pg.GraphicsObject):
             # self.disable_cache()
             # flip_cache = True
 
+        if draw_last:
+            self.draw_last(x, y)
+            profiler('draw last segment')
+
+
+        # if flip_cache:
+        # #     # XXX: seems to be needed to avoid artifacts (see above).
+        #     self.setCacheMode(QGraphicsItem.DeviceCoordinateCache)
+
+        # trigger redraw of path
+        # do update before reverting to cache mode
+        self.update()
+        profiler('.update()')
+
+    def draw_last(
+        self,
+        x: np.ndarray,
+        y: np.ndarray,
+
+    ) -> None:
+        x_last = x[-1]
+        y_last = y[-1]
+
         # draw the "current" step graphic segment so it lines up with
         # the "middle" of the current (OHLC) sample.
         if self._step_mode:
             self._last_line = QLineF(
                 x_last - 0.5, 0,
                 x_last + 0.5, 0,
+                # x_last, 0,
+                # x_last, 0,
             )
             self._last_step_rect = QRectF(
                 x_last - 0.5, 0,
                 x_last + 0.5, y_last
+                # x_last, 0,
+                # x_last, y_last
             )
             # print(
             #     f"path br: {self.path.boundingRect()}",
@@ -548,20 +678,10 @@ class FastAppendCurve(pg.GraphicsObject):
         else:
             self._last_line = QLineF(
                 x[-2], y[-2],
-                x[-1], y_last
+                x_last, y_last
             )
 
-        profiler('draw last segment')
-
-        # trigger redraw of path
-        # do update before reverting to cache mode
-        # self.prepareGeometryChange()
         self.update()
-        profiler('.update()')
-
-        # if flip_cache:
-        #     # XXX: seems to be needed to avoid artifacts (see above).
-        #     self.setCacheMode(QGraphicsItem.DeviceCoordinateCache)
 
     # XXX: lol brutal, the internals of `CurvePoint` (inherited by
     # our `LineDot`) required ``.getData()`` to work..
@@ -596,6 +716,10 @@ class FastAppendCurve(pg.GraphicsObject):
         # self.disable_cache()
         # self.setCacheMode(QGraphicsItem.DeviceCoordinateCache)
 
+    def reset_cache(self) -> None:
+        self.disable_cache()
+        self.setCacheMode(QGraphicsItem.DeviceCoordinateCache)
+
     def disable_cache(self) -> None:
         '''
         Disable the use of the pixel coordinate cache and trigger a geo event.
@@ -604,7 +728,7 @@ class FastAppendCurve(pg.GraphicsObject):
         # XXX: pretty annoying but, without this there's little
         # artefacts on the append updates to the curve...
         self.setCacheMode(QtWidgets.QGraphicsItem.NoCache)
-        self.prepareGeometryChange()
+        # self.prepareGeometryChange()
 
     def boundingRect(self):
         '''
@@ -624,6 +748,7 @@ class FastAppendCurve(pg.GraphicsObject):
 
         '''
         hb = self.path.controlPointRect()
+        # hb = self.path.boundingRect()
         hb_size = hb.size()
 
         fp = self.fast_path
@@ -632,17 +757,47 @@ class FastAppendCurve(pg.GraphicsObject):
             hb_size = fhb.size() + hb_size
         # print(f'hb_size: {hb_size}')
 
+        # if self._last_step_rect:
+        #     hb_size += self._last_step_rect.size()
+
+        # if self._line:
+        #     br = self._last_step_rect.bottomRight()
+
+        # tl = QPointF(
+        #     # self._vr[0],
+        #     # hb.topLeft().y(),
+        #     # 0,
+        #     # hb_size.height() + 1
+        # )
+
+        # if self._last_step_rect:
+        #     br = self._last_step_rect.bottomRight()
+
+        # else:
+        # hb_size += QSizeF(1, 1)
         w = hb_size.width() + 1
         h = hb_size.height() + 1
 
+        # br = QPointF(
+        #     self._vr[-1],
+        #     # tl.x() + w,
+        #     tl.y() + h,
+        # )
+
         br = QRectF(
 
             # top left
+            # hb.topLeft()
+            # tl,
             QPointF(hb.topLeft()),
 
+            # br,
             # total size
+            # QSizeF(hb_size)
+            # hb_size,
             QSizeF(w, h)
         )
+        self._br = br
         # print(f'bounding rect: {br}')
         return br
 
@@ -659,6 +814,7 @@ class FastAppendCurve(pg.GraphicsObject):
             disabled=not pg_profile_enabled(),
             gt=ms_slower_then,
         )
+        self.prepareGeometryChange()
 
         if (
             self._step_mode
@@ -681,7 +837,7 @@ class FastAppendCurve(pg.GraphicsObject):
 
         if path:
             p.drawPath(path)
-            profiler('.drawPath(path)')
+            profiler(f'.drawPath(path): {path.capacity()}')
 
         fp = self.fast_path
         if fp:

piker/ui/_display.py

@@ -29,6 +29,7 @@ from typing import Optional, Any, Callable
 import numpy as np
 import tractor
 import trio
+import pendulum
 import pyqtgraph as pg
 
 from .. import brokers
@@ -47,6 +48,7 @@ from ._fsp import (
     open_vlm_displays,
 )
 from ..data._sharedmem import ShmArray
+from ..data._source import tf_in_1s
 from ._forms import (
     FieldsForm,
     mk_order_pane_layout,
@@ -61,7 +63,7 @@ from ..log import get_logger
 log = get_logger(__name__)
 
 # TODO: load this from a config.toml!
-_quote_throttle_rate: int = 12  # Hz
+_quote_throttle_rate: int = 22  # Hz
 
 
 # a working tick-type-classes template
@@ -94,28 +96,19 @@ def chart_maxmin(
     Compute max and min datums "in view" for range limits.
 
     '''
-    array = ohlcv_shm.array
-    ifirst = array[0]['index']
-
     last_bars_range = chart.bars_range()
-    l, lbar, rbar, r = last_bars_range
-    in_view = array[lbar - ifirst:rbar - ifirst + 1]
+    out = chart.maxmin()
 
-    if not in_view.size:
-        log.warning('Resetting chart to data')
-        chart.default_view()
+    if out is None:
         return (last_bars_range, 0, 0, 0)
 
-    mx, mn = (
-        np.nanmax(in_view['high']),
-        np.nanmin(in_view['low'],)
-    )
+    mn, mx = out
 
     mx_vlm_in_view = 0
     if vlm_chart:
-        mx_vlm_in_view = np.max(
-            in_view['volume']
-        )
+        out = vlm_chart.maxmin()
+        if out:
+            _, mx_vlm_in_view = out
 
     return (
         last_bars_range,
@@ -338,7 +331,7 @@ def graphics_update_cycle(
     vars = ds.vars
     tick_margin = vars['tick_margin']
 
-    update_uppx = 6
+    update_uppx = 16
 
     for sym, quote in ds.quotes.items():
 
@@ -373,7 +366,7 @@ def graphics_update_cycle(
         mx = mx_in_view + tick_margin
         mn = mn_in_view - tick_margin
         profiler('maxmin call')
-        liv = r > i_step  # the last datum is in view
+        liv = r >= i_step  # the last datum is in view
 
         # don't real-time "shift" the curve to the
         # left unless we get one of the following:
@@ -381,7 +374,6 @@ def graphics_update_cycle(
             (
                 i_diff > 0  # no new sample step
                 and xpx < 4  # chart is zoomed out very far
-                and r >= i_step  # the last datum isn't in view
                 and liv
             )
             or trigger_all
@@ -398,16 +390,17 @@ def graphics_update_cycle(
             )
 
             if (
-                (xpx < update_uppx or i_diff > 0)
+                (
+                    xpx < update_uppx
+                    or i_diff > 0
+                    and liv
+                )
                 or trigger_all
-                and liv
             ):
                 # TODO: make it so this doesn't have to be called
                 # once the $vlm is up?
-                vlm_chart.update_graphics_from_array(
+                vlm_chart.update_graphics_from_flow(
                     'volume',
-                    array,
-
                     # UGGGh, see ``maxmin()`` impl in `._fsp` for
                     # the overlayed plotitems... we need a better
                     # bay to invoke a maxmin per overlay..
@@ -432,11 +425,16 @@ def graphics_update_cycle(
                     vars['last_mx_vlm'] = mx_vlm_in_view
 
                 for curve_name, flow in vlm_chart._flows.items():
+
+                    if not flow.render:
+                        continue
+
                     update_fsp_chart(
                         vlm_chart,
                         flow,
                         curve_name,
                         array_key=curve_name,
+                        do_append=xpx < update_uppx,
                     )
                     # is this even doing anything?
                     # (pretty sure it's the real-time
@@ -494,10 +492,11 @@ def graphics_update_cycle(
         if (
             xpx < update_uppx
             or i_diff > 0
+            or trigger_all
         ):
-            chart.update_graphics_from_array(
+            chart.update_graphics_from_flow(
                 chart.name,
-                array,
+                do_append=xpx < update_uppx,
             )
 
         # iterate in FIFO order per tick-frame
@@ -510,8 +509,9 @@ def graphics_update_cycle(
             # tick frames to determine the y-range for chart
             # auto-scaling.
             # TODO: we need a streaming minmax algo here, see def above.
-            mx = max(price + tick_margin, mx)
-            mn = min(price - tick_margin, mn)
+            if liv:
+                mx = max(price + tick_margin, mx)
+                mn = min(price - tick_margin, mn)
 
             if typ in clear_types:
 
@@ -534,9 +534,8 @@ def graphics_update_cycle(
 
                 if wap_in_history:
                     # update vwap overlay line
-                    chart.update_graphics_from_array(
+                    chart.update_graphics_from_flow(
                         'bar_wap',
-                        array,
                     )
 
             # L1 book label-line updates
@@ -552,7 +551,7 @@ def graphics_update_cycle(
 
                 if (
                     label is not None
-                    # and liv
+                    and liv
                 ):
                     label.update_fields(
                         {'level': price, 'size': size}
@@ -566,7 +565,7 @@ def graphics_update_cycle(
                 typ in _tick_groups['asks']
                 # TODO: instead we could check if the price is in the
                 # y-view-range?
-                # and liv
+                and liv
             ):
                 l1.ask_label.update_fields({'level': price, 'size': size})
 
@@ -574,7 +573,7 @@ def graphics_update_cycle(
                 typ in _tick_groups['bids']
                 # TODO: instead we could check if the price is in the
                 # y-view-range?
-                # and liv
+                and liv
             ):
                 l1.bid_label.update_fields({'level': price, 'size': size})
 
@@ -589,6 +588,7 @@ def graphics_update_cycle(
                 main_vb._ic is None
                 or not main_vb._ic.is_set()
             ):
+                # print(f'updating range due to mxmn')
                 main_vb._set_yrange(
                     # TODO: we should probably scale
                     # the view margin based on the size
@@ -599,6 +599,7 @@ def graphics_update_cycle(
                     yrange=(mn, mx),
                 )
 
+        # XXX: update this every draw cycle to make L1-always-in-view work.
         vars['last_mx'], vars['last_mn'] = mx, mn
 
         # run synchronous update on all linked flows
@@ -661,11 +662,17 @@ async def display_symbol_data(
         symbol = feed.symbols[sym]
         fqsn = symbol.front_fqsn()
 
+        times = bars['time']
+        end = pendulum.from_timestamp(times[-1])
+        start = pendulum.from_timestamp(times[times != times[-1]][-1])
+        step_size_s = (end - start).seconds
+        tf_key = tf_in_1s[step_size_s]
+
         # load in symbol's ohlc data
         godwidget.window.setWindowTitle(
             f'{fqsn} '
             f'tick:{symbol.tick_size} '
-            f'step:1s '
+            f'step:{tf_key} '
         )
 
         linked = godwidget.linkedsplits
@@ -681,32 +688,31 @@ async def display_symbol_data(
         # create main OHLC chart
         chart = linked.plot_ohlc_main(
             symbol,
-            bars,
+            ohlcv,
             sidepane=pp_pane,
         )
+        chart.default_view()
         chart._feeds[symbol.key] = feed
         chart.setFocus()
 
         # plot historical vwap if available
         wap_in_history = False
 
-        if brokermod._show_wap_in_history:
+        # XXX: FOR SOME REASON THIS IS CAUSING HANGZ!?!
+        # if brokermod._show_wap_in_history:
 
-            if 'bar_wap' in bars.dtype.fields:
-                wap_in_history = True
-                chart.draw_curve(
-                    name='bar_wap',
-                    data=bars,
-                    add_label=False,
-                )
+        #     if 'bar_wap' in bars.dtype.fields:
+        #         wap_in_history = True
+        #         chart.draw_curve(
+        #             name='bar_wap',
+        #             shm=ohlcv,
+        #             color='default_light',
+        #             add_label=False,
+        #         )
 
         # size view to data once at outset
         chart.cv._set_yrange()
 
-        # TODO: a data view api that makes this less shit
-        chart._shm = ohlcv
-        chart._flows[chart.data_key].shm = ohlcv
-
         # NOTE: we must immediately tell Qt to show the OHLC chart
         # to avoid a race where the subplots get added/shown to
         # the linked set *before* the main price chart!
@@ -769,6 +775,5 @@ async def display_symbol_data(
                 sbar._status_groups[loading_sym_key][1]()
 
                 # let the app run.. bby
-                chart.default_view()
                 # linked.graphics_cycle()
                 await trio.sleep_forever()

piker/ui/_editors.py

@@ -343,7 +343,7 @@ class SelectRect(QtGui.QGraphicsRectItem):
         nbars = ixmx - ixmn + 1
 
         chart = self._chart
-        data = chart._arrays[chart.name][ixmn:ixmx]
+        data = chart._flows[chart.name].shm.array[ixmn:ixmx]
 
         if len(data):
             std = data['close'].std()

piker/ui/_flows.py

@@ -0,0 +1,924 @@
+# piker: trading gear for hackers
+# Copyright (C) Tyler Goodlet (in stewardship for pikers)
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+'''
+High level streaming graphics primitives.
+
+This is an intermediate layer which associates real-time low latency
+graphics primitives with underlying FSP related data structures for fast
+incremental update.
+
+'''
+from __future__ import annotations
+from functools import partial
+from typing import (
+    Optional,
+    Callable,
+)
+
+import msgspec
+import numpy as np
+from numpy.lib import recfunctions as rfn
+import pyqtgraph as pg
+from PyQt5.QtGui import QPainterPath
+from PyQt5.QtCore import (
+    # Qt,
+    QLineF,
+    # QSizeF,
+    QRectF,
+    # QPointF,
+)
+
+from ..data._sharedmem import (
+    ShmArray,
+    # open_shm_array,
+)
+from .._profile import pg_profile_enabled, ms_slower_then
+from ._ohlc import (
+    BarItems,
+    gen_qpath,
+)
+from ._curve import (
+    FastAppendCurve,
+    # step_path_arrays_from_1d,
+)
+from ._compression import (
+    # ohlc_flatten,
+    ds_m4,
+)
+from ..log import get_logger
+
+
+log = get_logger(__name__)
+
+# class FlowsTable(msgspec.Struct):
+#     '''
+#     Data-AGGRegate: high level API onto multiple (categorized)
+#     ``Flow``s with high level processing routines for
+#     multi-graphics computations and display.
+
+#     '''
+#     flows: dict[str, np.ndarray] = {}
+
+# @classmethod
+# def from_token(
+#     cls,
+#     shm_token: tuple[
+#         str,
+#         str,
+#         tuple[str, str],
+#     ],
+
+# ) -> Renderer:
+
+#     shm = attach_shm_array(token)
+#     return cls(shm)
+
+
+def rowarr_to_path(
+    rows_array: np.ndarray,
+    x_basis: np.ndarray,
+    flow: Flow,
+
+) -> QPainterPath:
+
+    # TODO: we could in theory use ``numba`` to flatten
+    # if needed?
+
+    # to 1d
+    y = rows_array.flatten()
+
+    return pg.functions.arrayToQPath(
+        # these get passed at render call time
+        x=x_basis[:y.size],
+        y=y,
+        connect='all',
+        finiteCheck=False,
+        path=flow.path,
+    )
+
+
+def mk_ohlc_flat_copy(
+    ohlc_shm: ShmArray,
+
+    # XXX: we bind this in currently..
+    # x_basis: np.ndarray,
+
+    # vr: Optional[slice] = None,
+
+) -> tuple[np.ndarray, np.ndarray]:
+    '''
+    Return flattened-non-copy view into an OHLC shm array.
+
+    '''
+    ohlc = ohlc_shm._array[['open', 'high', 'low', 'close']]
+    # if vr:
+    #     ohlc = ohlc[vr]
+    #     x = x_basis[vr]
+
+    unstructured = rfn.structured_to_unstructured(
+        ohlc,
+        copy=False,
+    )
+    # breakpoint()
+    y = unstructured.flatten()
+    # x = x_basis[:y.size]
+    return y
+
+
+class Flow(msgspec.Struct):  # , frozen=True):
+    '''
+    (Financial Signal-)Flow compound type which wraps a real-time
+    shm array stream with displayed graphics (curves, charts)
+    for high level access and control as well as efficient incremental
+    update.
+
+    The intention is for this type to eventually be capable of shm-passing
+    of incrementally updated graphics stream data between actors.
+
+    '''
+    name: str
+    plot: pg.PlotItem
+    graphics: pg.GraphicsObject
+    _shm: ShmArray
+
+    is_ohlc: bool = False
+    render: bool = True  # toggle for display loop
+    gy: Optional[ShmArray] = None
+    gx: Optional[np.ndarray] = None
+    _iflat_last: int = 0
+    _iflat_first: int = 0
+
+    _last_uppx: float = 0
+    _in_ds: bool = False
+
+    _graphics_tranform_fn: Optional[Callable[ShmArray, np.ndarray]] = None
+
+    # map from uppx -> (downsampled data, incremental graphics)
+    _src_r: Optional[Renderer] = None
+    _render_table: dict[
+        Optional[int],
+        tuple[Renderer, pg.GraphicsItem],
+    ] = {}
+
+    # TODO: hackery to be able to set a shm later
+    # but whilst also allowing this type to hashable,
+    # likely will require serializable token that is used to attach
+    # to the underlying shm ref after startup?
+    # _shm: Optional[ShmArray] = None  # currently, may be filled in "later"
+
+    # last read from shm (usually due to an update call)
+    _last_read: Optional[np.ndarray] = None
+
+    # cache of y-range values per x-range input.
+    _mxmns: dict[tuple[int, int], tuple[float, float]] = {}
+
+    @property
+    def shm(self) -> ShmArray:
+        return self._shm
+
+    # TODO: remove this and only allow setting through
+    # private ``._shm`` attr?
+    @shm.setter
+    def shm(self, shm: ShmArray) -> ShmArray:
+        print(f'{self.name} DO NOT SET SHM THIS WAY!?')
+        self._shm = shm
+
+    def maxmin(
+        self,
+        lbar,
+        rbar,
+
+    ) -> tuple[float, float]:
+        '''
+        Compute the cached max and min y-range values for a given
+        x-range determined by ``lbar`` and ``rbar``.
+
+        '''
+        rkey = (lbar, rbar)
+        cached_result = self._mxmns.get(rkey)
+        if cached_result:
+            return cached_result
+
+        shm = self.shm
+        if shm is None:
+            mxmn = None
+
+        else:  # new block for profiling?..
+            arr = shm.array
+
+            # build relative indexes into shm array
+            # TODO: should we just add/use a method
+            # on the shm to do this?
+            ifirst = arr[0]['index']
+            slice_view = arr[
+                lbar - ifirst:
+                (rbar - ifirst) + 1
+            ]
+
+            if not slice_view.size:
+                mxmn = None
+
+            else:
+                if self.is_ohlc:
+                    ylow = np.min(slice_view['low'])
+                    yhigh = np.max(slice_view['high'])
+
+                else:
+                    view = slice_view[self.name]
+                    ylow = np.min(view)
+                    yhigh = np.max(view)
+
+                mxmn = ylow, yhigh
+
+            if mxmn is not None:
+                # cache new mxmn result
+                self._mxmns[rkey] = mxmn
+
+            return mxmn
+
+    def view_range(self) -> tuple[int, int]:
+        '''
+        Return the indexes in view for the associated
+        plot displaying this flow's data.
+
+        '''
+        vr = self.plot.viewRect()
+        return int(vr.left()), int(vr.right())
+
+    def datums_range(self) -> tuple[
+        int, int, int, int, int, int
+    ]:
+        '''
+        Return a range tuple for the datums present in view.
+
+        '''
+        l, r = self.view_range()
+
+        # TODO: avoid this and have shm passed
+        # in earlier.
+        if self.shm is None:
+            # haven't initialized the flow yet
+            return (0, l, 0, 0, r, 0)
+
+        array = self.shm.array
+        index = array['index']
+        start = index[0]
+        end = index[-1]
+        lbar = max(l, start)
+        rbar = min(r, end)
+        return (
+            start, l, lbar, rbar, r, end,
+        )
+
+    def read(self) -> tuple[
+            int, int, np.ndarray,
+            int, int, np.ndarray,
+    ]:
+        # read call
+        array = self.shm.array
+
+        indexes = array['index']
+        ifirst = indexes[0]
+        ilast = indexes[-1]
+
+        ifirst, l, lbar, rbar, r, ilast = self.datums_range()
+
+        # get read-relative indices adjusting
+        # for master shm index.
+        lbar_i = max(l, ifirst) - ifirst
+        rbar_i = min(r, ilast) - ifirst
+
+        # TODO: we could do it this way as well no?
+        # to_draw = array[lbar - ifirst:(rbar - ifirst) + 1]
+        in_view = array[lbar_i: rbar_i + 1]
+
+        return (
+            # abs indices + full data set
+            ifirst, ilast, array,
+
+            # relative indices + in view datums
+            lbar_i, rbar_i, in_view,
+        )
+
+    def update_graphics(
+        self,
+        use_vr: bool = True,
+        render: bool = True,
+        array_key: Optional[str] = None,
+
+        profiler: Optional[pg.debug.Profiler] = None,
+
+        **kwargs,
+
+    ) -> pg.GraphicsObject:
+        '''
+        Read latest datums from shm and render to (incrementally)
+        render to graphics.
+
+        '''
+
+        profiler = profiler or pg.debug.Profiler(
+            msg=f'Flow.update_graphics() for {self.name}',
+            disabled=not pg_profile_enabled(),
+            gt=ms_slower_then,
+            delayed=True,
+        )
+        # shm read and slice to view
+        read = (
+            xfirst, xlast, array,
+            ivl, ivr, in_view,
+        ) = self.read()
+        profiler('read src shm data')
+
+        if (
+            not in_view.size
+            or not render
+        ):
+            return self.graphics
+
+        graphics = self.graphics
+        if isinstance(graphics, BarItems):
+
+            # if no source data renderer exists create one.
+            r = self._src_r
+            if not r:
+                # OHLC bars path renderer
+                r = self._src_r = Renderer(
+                    flow=self,
+                    # TODO: rename this to something with ohlc
+                    draw_path=gen_qpath,
+                    last_read=read,
+                )
+
+                ds_curve_r = Renderer(
+                    flow=self,
+
+                    # just swap in the flat view
+                    # data_t=lambda array: self.gy.array,
+                    last_read=read,
+                    draw_path=partial(
+                        rowarr_to_path,
+                        x_basis=None,
+                    ),
+
+                )
+                curve = FastAppendCurve(
+                    name='OHLC',
+                    color=graphics._color,
+                )
+                curve.hide()
+                self.plot.addItem(curve)
+
+                # baseline "line" downsampled OHLC curve that should
+                # kick on only when we reach a certain uppx threshold.
+                self._render_table[0] = (
+                    ds_curve_r,
+                    curve,
+                )
+
+            dsc_r, curve = self._render_table[0]
+
+            # do checks for whether or not we require downsampling:
+            # - if we're **not** downsampling then we simply want to
+            #   render the bars graphics curve and update..
+            # - if insteam we are in a downsamplig state then we to
+            x_gt = 6
+            uppx = curve.x_uppx()
+            in_line = should_line = curve.isVisible()
+            if (
+                should_line
+                and uppx < x_gt
+            ):
+                print('FLIPPING TO BARS')
+                should_line = False
+
+            elif (
+                not should_line
+                and uppx >= x_gt
+            ):
+                print('FLIPPING TO LINE')
+                should_line = True
+
+            profiler(f'ds logic complete line={should_line}')
+
+            # do graphics updates
+            if should_line:
+
+                fields = ['open', 'high', 'low', 'close']
+                if self.gy is None:
+                    # create a flattened view onto the OHLC array
+                    # which can be read as a line-style format
+                    shm = self.shm
+
+                    # flat = self.gy = self.shm.unstruct_view(fields)
+                    self.gy = self.shm.ustruct(fields)
+                    first = self._iflat_first = self.shm._first.value
+                    last = self._iflat_last = self.shm._last.value
+
+                    # write pushed data to flattened copy
+                    self.gy[first:last] = rfn.structured_to_unstructured(
+                        self.shm.array[fields]
+                    )
+
+                    # generate an flat-interpolated x-domain
+                    self.gx = (
+                        np.broadcast_to(
+                            shm._array['index'][:, None],
+                            (
+                                shm._array.size,
+                                # 4,  # only ohlc
+                                self.gy.shape[1],
+                            ),
+                        ) + np.array([-0.5, 0, 0, 0.5])
+                    )
+                    assert self.gy.any()
+
+                # print(f'unstruct diff: {time.time() - start}')
+                # profiler('read unstr view bars to line')
+                # start = self.gy._first.value
+                # update flatted ohlc copy
+                (
+                    iflat_first,
+                    iflat,
+                    ishm_last,
+                    ishm_first,
+                ) = (
+                    self._iflat_first,
+                    self._iflat_last,
+                    self.shm._last.value,
+                    self.shm._first.value
+                )
+
+                # check for shm prepend updates since last read.
+                if iflat_first != ishm_first:
+
+                    # write newly prepended data to flattened copy
+                    self.gy[
+                        ishm_first:iflat_first
+                    ] = rfn.structured_to_unstructured(
+                        self.shm._array[fields][ishm_first:iflat_first]
+                    )
+                    self._iflat_first = ishm_first
+
+                #     # flat = self.gy = self.shm.unstruct_view(fields)
+                #     self.gy = self.shm.ustruct(fields)
+                #     # self._iflat_last = self.shm._last.value
+
+                #     # self._iflat_first = self.shm._first.value
+                #     # do an update for the most recent prepend
+                #     # index
+                #     iflat = ishm_first
+
+                to_update = rfn.structured_to_unstructured(
+                    self.shm._array[iflat:ishm_last][fields]
+                )
+
+                self.gy[iflat:ishm_last][:] = to_update
+                profiler('updated ustruct OHLC data')
+
+                # slice out up-to-last step contents
+                y_flat = self.gy[ishm_first:ishm_last]
+                x_flat = self.gx[ishm_first:ishm_last]
+
+                # update local last-index tracking
+                self._iflat_last = ishm_last
+
+                # reshape to 1d for graphics rendering
+                y = y_flat.reshape(-1)
+                x = x_flat.reshape(-1)
+                profiler('flattened ustruct OHLC data')
+
+                # do all the same for only in-view data
+                y_iv_flat = y_flat[ivl:ivr]
+                x_iv_flat = x_flat[ivl:ivr]
+                y_iv = y_iv_flat.reshape(-1)
+                x_iv = x_iv_flat.reshape(-1)
+                profiler('flattened ustruct in-view OHLC data')
+
+                # legacy full-recompute-everytime method
+                # x, y = ohlc_flatten(array)
+                # x_iv, y_iv = ohlc_flatten(in_view)
+                # profiler('flattened OHLC data')
+
+                curve.update_from_array(
+                    x,
+                    y,
+                    x_iv=x_iv,
+                    y_iv=y_iv,
+                    view_range=(ivl, ivr),  # hack
+                    profiler=profiler,
+                    # should_redraw=False,
+
+                    # NOTE: already passed through by display loop?
+                    # do_append=uppx < 16,
+                    **kwargs,
+                )
+                curve.show()
+                profiler('updated ds curve')
+
+            else:
+                # render incremental or in-view update
+                # and apply ouput (path) to graphics.
+                path, last = r.render(
+                    read,
+                    only_in_view=True,
+                )
+
+                graphics.path = path
+                graphics.draw_last(last)
+
+                # NOTE: on appends we used to have to flip the coords
+                # cache thought it doesn't seem to be required any more?
+                # graphics.setCacheMode(QtWidgets.QGraphicsItem.NoCache)
+                # graphics.setCacheMode(QtWidgets.QGraphicsItem.DeviceCoordinateCache)
+
+                # graphics.prepareGeometryChange()
+                graphics.update()
+
+            if (
+                not in_line
+                and should_line
+            ):
+                # change to line graphic
+
+                log.info(
+                    f'downsampling to line graphic {self.name}'
+                )
+                graphics.hide()
+                # graphics.update()
+                curve.show()
+                curve.update()
+
+            elif in_line and not should_line:
+                log.info(f'showing bars graphic {self.name}')
+                curve.hide()
+                graphics.show()
+                graphics.update()
+
+            #   update our pre-downsample-ready data and then pass that
+            #   new data the downsampler algo for incremental update.
+
+                # graphics.update_from_array(
+                #     array,
+                #     in_view,
+                #     view_range=(ivl, ivr) if use_vr else None,
+
+                #     **kwargs,
+                # )
+
+                # generate and apply path to graphics obj
+                # graphics.path, last = r.render(
+                #     read,
+                #     only_in_view=True,
+                # )
+                # graphics.draw_last(last)
+
+        else:
+            # ``FastAppendCurve`` case:
+            array_key = array_key or self.name
+            uppx = graphics.x_uppx()
+
+            if graphics._step_mode and self.gy is None:
+                self._iflat_first = self.shm._first.value
+
+                # create a flattened view onto the OHLC array
+                # which can be read as a line-style format
+                shm = self.shm
+
+                # fields = ['index', array_key]
+                i = shm._array['index'].copy()
+                out = shm._array[array_key].copy()
+
+                self.gx = np.broadcast_to(
+                    i[:, None],
+                    (i.size, 2),
+                ) + np.array([-0.5, 0.5])
+
+                # self.gy = np.broadcast_to(
+                #     out[:, None], (out.size, 2),
+                # )
+                self.gy = np.empty((len(out), 2), dtype=out.dtype)
+                self.gy[:] = out[:, np.newaxis]
+
+                # start y at origin level
+                self.gy[0, 0] = 0
+
+            if graphics._step_mode:
+                (
+                    iflat_first,
+                    iflat,
+                    ishm_last,
+                    ishm_first,
+                ) = (
+                    self._iflat_first,
+                    self._iflat_last,
+                    self.shm._last.value,
+                    self.shm._first.value
+                )
+
+                il = max(iflat - 1, 0)
+
+                # check for shm prepend updates since last read.
+                if iflat_first != ishm_first:
+
+                    print(f'prepend {array_key}')
+
+                    # i_prepend = self.shm._array['index'][
+                    #   ishm_first:iflat_first]
+                    y_prepend = self.shm._array[array_key][
+                        ishm_first:iflat_first
+                    ]
+
+                    y2_prepend = np.broadcast_to(
+                        y_prepend[:, None], (y_prepend.size, 2),
+                    )
+
+                    # write newly prepended data to flattened copy
+                    self.gy[ishm_first:iflat_first] = y2_prepend
+                    self._iflat_first = ishm_first
+
+                append_diff = ishm_last - iflat
+                if append_diff:
+
+                    # slice up to the last datum since last index/append update
+                    # new_x = self.shm._array[il:ishm_last]['index']
+                    new_y = self.shm._array[il:ishm_last][array_key]
+
+                    new_y2 = np.broadcast_to(
+                        new_y[:, None], (new_y.size, 2),
+                    )
+                    self.gy[il:ishm_last] = new_y2
+                    profiler('updated step curve data')
+
+                    # print(
+                    #     f'append size: {append_diff}\n'
+                    #     f'new_x: {new_x}\n'
+                    #     f'new_y: {new_y}\n'
+                    #     f'new_y2: {new_y2}\n'
+                    #     f'new gy: {gy}\n'
+                    # )
+
+                    # update local last-index tracking
+                    self._iflat_last = ishm_last
+
+                # slice out up-to-last step contents
+                x_step = self.gx[ishm_first:ishm_last+2]
+                # shape to 1d
+                x = x_step.reshape(-1)
+
+                y_step = self.gy[ishm_first:ishm_last+2]
+                lasts = self.shm.array[['index', array_key]]
+                last = lasts[array_key][-1]
+                y_step[-1] = last
+                # shape to 1d
+                y = y_step.reshape(-1)
+
+                # s = 6
+                # print(f'lasts: {x[-2*s:]}, {y[-2*s:]}')
+
+                profiler('sliced step data')
+
+                # do all the same for only in-view data
+                ys_iv = y_step[ivl:ivr+1]
+                xs_iv = x_step[ivl:ivr+1]
+                y_iv = ys_iv.reshape(ys_iv.size)
+                x_iv = xs_iv.reshape(xs_iv.size)
+                # print(
+                #     f'ys_iv : {ys_iv[-s:]}\n'
+                #     f'y_iv: {y_iv[-s:]}\n'
+                #     f'xs_iv: {xs_iv[-s:]}\n'
+                #     f'x_iv: {x_iv[-s:]}\n'
+                # )
+                profiler('flattened ustruct in-view OHLC data')
+
+                # legacy full-recompute-everytime method
+                # x, y = ohlc_flatten(array)
+                # x_iv, y_iv = ohlc_flatten(in_view)
+                # profiler('flattened OHLC data')
+
+                x_last = array['index'][-1]
+                y_last = array[array_key][-1]
+                graphics._last_line = QLineF(
+                    x_last - 0.5, 0,
+                    x_last + 0.5, 0,
+                )
+                graphics._last_step_rect = QRectF(
+                    x_last - 0.5, 0,
+                    x_last + 0.5, y_last,
+                )
+                # graphics.update()
+
+                graphics.update_from_array(
+                    x=x,
+                    y=y,
+
+                    x_iv=x_iv,
+                    y_iv=y_iv,
+
+                    view_range=(ivl, ivr) if use_vr else None,
+
+                    draw_last=False,
+                    slice_to_head=-2,
+
+                    should_redraw=bool(append_diff),
+
+                    # NOTE: already passed through by display loop?
+                    # do_append=uppx < 16,
+
+                    **kwargs
+                )
+                # graphics.reset_cache()
+                # print(
+                #     f"path br: {graphics.path.boundingRect()}\n",
+                #     # f"fast path br: {graphics.fast_path.boundingRect()}",
+                #     f"last rect br: {graphics._last_step_rect}\n",
+                #     f"full br: {graphics._br}\n",
+                # )
+
+            else:
+                x = array['index']
+                y = array[array_key]
+                x_iv = in_view['index']
+                y_iv = in_view[array_key]
+
+                # graphics.draw_last(x, y)
+                profiler(f'draw last segment {array_key}')
+
+                graphics.update_from_array(
+                    x=x,
+                    y=y,
+
+                    x_iv=x_iv,
+                    y_iv=y_iv,
+
+                    view_range=(ivl, ivr) if use_vr else None,
+
+                    # NOTE: already passed through by display loop?
+                    # do_append=uppx < 16,
+                    **kwargs
+                )
+
+        return graphics
+
+
+def xy_downsample(
+    x,
+    y,
+    px_width,
+    uppx,
+
+    x_spacer: float = 0.5,
+
+) -> tuple[np.ndarray, np.ndarray]:
+
+    # downsample whenever more then 1 pixels per datum can be shown.
+    # always refresh data bounds until we get diffing
+    # working properly, see above..
+    bins, x, y = ds_m4(
+        x,
+        y,
+        px_width=px_width,
+        uppx=uppx,
+        log_scale=bool(uppx)
+    )
+
+    # flatten output to 1d arrays suitable for path-graphics generation.
+    x = np.broadcast_to(x[:, None], y.shape)
+    x = (x + np.array(
+        [-x_spacer, 0, 0, x_spacer]
+    )).flatten()
+    y = y.flatten()
+
+    return x, y
+
+
+class Renderer(msgspec.Struct):
+
+    flow: Flow
+
+    # called to render path graphics
+    draw_path: Callable[np.ndarray, QPainterPath]
+
+    # called on input data but before any graphics format
+    # conversions or processing.
+    data_t: Optional[Callable[ShmArray, np.ndarray]] = None
+    data_t_shm: Optional[ShmArray] = None
+
+    # called on the final data (transform) output to convert
+    # to "graphical data form" a format that can be passed to
+    # the ``.draw()`` implementation.
+    graphics_t: Optional[Callable[ShmArray, np.ndarray]] = None
+    graphics_t_shm: Optional[ShmArray] = None
+
+    # path graphics update implementation methods
+    prepend_fn: Optional[Callable[QPainterPath, QPainterPath]] = None
+    append_fn: Optional[Callable[QPainterPath, QPainterPath]] = None
+
+    # last array view read
+    last_read: Optional[np.ndarray] = None
+
+    # output graphics rendering, the main object
+    # processed in ``QGraphicsObject.paint()``
+    path: Optional[QPainterPath] = None
+
+    # def diff(
+    #     self,
+    #     latest_read: tuple[np.ndarray],
+
+    # ) -> tuple[np.ndarray]:
+    #     # blah blah blah
+    #     # do diffing for prepend, append and last entry
+    #     return (
+    #         to_prepend
+    #         to_append
+    #         last,
+    #     )
+
+    def render(
+        self,
+
+        new_read,
+
+        # only render datums "in view" of the ``ChartView``
+        only_in_view: bool = False,
+
+    ) -> list[QPainterPath]:
+        '''
+        Render the current graphics path(s)
+
+        There are (at least) 3 stages from source data to graphics data:
+        - a data transform (which can be stored in additional shm)
+        - a graphics transform which converts discrete basis data to
+          a `float`-basis view-coords graphics basis. (eg. ``ohlc_flatten()``,
+          ``step_path_arrays_from_1d()``, etc.)
+
+        - blah blah blah (from notes)
+
+        '''
+        # do full source data render to path
+        (
+            xfirst, xlast, array,
+            ivl, ivr, in_view,
+        ) = self.last_read
+
+        if only_in_view:
+            array = in_view
+            # # get latest data from flow shm
+            # self.last_read = (
+            #     xfirst, xlast, array, ivl, ivr, in_view
+            # ) = new_read
+
+        if self.path is None or only_in_view:
+            # redraw the entire source data if we have either of:
+            # - no prior path graphic rendered or,
+            # - we always intend to re-render the data only in view
+
+            # data transform: convert source data to a format
+            # expected to be incrementally updates and later rendered
+            # to a more graphics native format.
+            if self.data_t:
+                array = self.data_t(array)
+
+                # maybe allocate shm for data transform output
+                # if self.data_t_shm is None:
+                #     fshm = self.flow.shm
+
+                #     shm, opened = maybe_open_shm_array(
+                #         f'{self.flow.name}_data_t',
+                #         # TODO: create entry for each time frame
+                #         dtype=array.dtype,
+                #         readonly=False,
+                #     )
+                #     assert opened
+                #     shm.push(array)
+                #     self.data_t_shm = shm
+
+        elif self.path:
+            print(f'inremental update not supported yet {self.flow.name}')
+            # TODO: do incremental update
+            # prepend, append, last = self.diff(self.flow.read())
+
+            # do path generation for each segment
+            # and then push into graphics object.
+
+        hist, last = array[:-1], array[-1]
+
+        # call path render func on history
+        self.path = self.draw_path(hist)
+
+        self.last_read = new_read
+        return self.path, last

piker/ui/_fsp.py

@@ -75,6 +75,7 @@ def update_fsp_chart(
     flow,
     graphics_name: str,
     array_key: Optional[str],
+    **kwargs,
 
 ) -> None:
 
@@ -93,10 +94,10 @@ def update_fsp_chart(
     # update graphics
     # NOTE: this does a length check internally which allows it
     # staying above the last row check below..
-    chart.update_graphics_from_array(
+    chart.update_graphics_from_flow(
         graphics_name,
-        array,
         array_key=array_key or graphics_name,
+        **kwargs,
     )
 
     # XXX: re: ``array_key``: fsp func names must be unique meaning we
@@ -106,9 +107,6 @@ def update_fsp_chart(
     # read from last calculated value and update any label
     last_val_sticky = chart._ysticks.get(graphics_name)
     if last_val_sticky:
-        # array = shm.array[array_key]
-        # if len(array):
-        #     value = array[-1]
         last = last_row[array_key]
         last_val_sticky.update_from_data(-1, last)
 
@@ -246,20 +244,18 @@ async def run_fsp_ui(
 
             chart.draw_curve(
                 name=name,
-                data=shm.array,
+                shm=shm,
                 overlay=True,
                 color='default_light',
                 array_key=name,
                 **conf.get('chart_kwargs', {})
             )
-            # specially store ref to shm for lookup in display loop
-            chart._flows[name].shm = shm
 
         else:
             # create a new sub-chart widget for this fsp
             chart = linkedsplits.add_plot(
                 name=name,
-                array=shm.array,
+                shm=shm,
 
                 array_key=name,
                 sidepane=sidepane,
@@ -271,12 +267,6 @@ async def run_fsp_ui(
                 **conf.get('chart_kwargs', {})
             )
 
-            # XXX: ONLY for sub-chart fsps, overlays have their
-            # data looked up from the chart's internal array set.
-            # TODO: we must get a data view api going STAT!!
-            chart._shm = shm
-            chart._flows[chart.data_key].shm = shm
-
             # should **not** be the same sub-chart widget
             assert chart.name != linkedsplits.chart.name
 
@@ -626,7 +616,7 @@ async def open_vlm_displays(
         shm = ohlcv
         chart = linked.add_plot(
             name='volume',
-            array=shm.array,
+            shm=shm,
 
             array_key='volume',
             sidepane=sidepane,
@@ -639,7 +629,6 @@ async def open_vlm_displays(
             # the curve item internals are pretty convoluted.
             style='step',
         )
-        chart._flows['volume'].shm = ohlcv
 
         # force 0 to always be in view
         def maxmin(
@@ -666,11 +655,6 @@ async def open_vlm_displays(
         # chart.hideAxis('right')
         # chart.showAxis('left')
 
-        # XXX: ONLY for sub-chart fsps, overlays have their
-        # data looked up from the chart's internal array set.
-        # TODO: we must get a data view api going STAT!!
-        chart._shm = shm
-
         # send back new chart to caller
         task_status.started(chart)
 
@@ -685,9 +669,9 @@ async def open_vlm_displays(
 
         last_val_sticky.update_from_data(-1, value)
 
-        vlm_curve = chart.update_graphics_from_array(
+        vlm_curve = chart.update_graphics_from_flow(
             'volume',
-            shm.array,
+            # shm.array,
         )
 
         # size view to data once at outset
@@ -795,9 +779,8 @@ async def open_vlm_displays(
                         color = 'bracket'
 
                     curve, _ = chart.draw_curve(
-                        # name='dolla_vlm',
                         name=name,
-                        data=shm.array,
+                        shm=shm,
                         array_key=name,
                         overlay=pi,
                         color=color,
@@ -812,7 +795,6 @@ async def open_vlm_displays(
                     # ``.draw_curve()``.
                     flow = chart._flows[name]
                     assert flow.plot is pi
-                    flow.shm = shm
 
             chart_curves(
                 fields,
@@ -847,7 +829,9 @@ async def open_vlm_displays(
             # liquidity events (well at least on low OHLC periods - 1s).
             vlm_curve.hide()
             chart.removeItem(vlm_curve)
-            chart._flows.pop('volume')
+            vflow = chart._flows['volume']
+            vflow.render = False
+
             # avoid range sorting on volume once disabled
             chart.view.disable_auto_yrange()
 
@@ -902,10 +886,10 @@ async def open_vlm_displays(
 
         # built-in vlm fsps
         for target, conf in {
-            tina_vwap: {
-                'overlay': 'ohlc',  # overlays with OHLCV (main) chart
-                'anchor': 'session',
-            },
+            # tina_vwap: {
+            #     'overlay': 'ohlc',  # overlays with OHLCV (main) chart
+            #     'anchor': 'session',
+            # },
         }.items():
             started = await admin.open_fsp_chart(
                 target,

piker/ui/_interaction.py

@@ -20,7 +20,6 @@ Chart view box primitives
 """
 from __future__ import annotations
 from contextlib import asynccontextmanager
-# import itertools
 import time
 from typing import Optional, Callable
 
@@ -35,10 +34,9 @@ import trio
 
 from ..log import get_logger
 from .._profile import pg_profile_enabled, ms_slower_then
-from ._style import _min_points_to_show
+# from ._style import _min_points_to_show
 from ._editors import SelectRect
 from . import _event
-from ._ohlc import BarItems
 
 
 log = get_logger(__name__)
@@ -486,15 +484,18 @@ class ChartView(ViewBox):
 
         # don't zoom more then the min points setting
         l, lbar, rbar, r = chart.bars_range()
-        vl = r - l
+        # vl = r - l
 
-        if ev.delta() > 0 and vl <= _min_points_to_show:
-            log.debug("Max zoom bruh...")
-            return
+        # if ev.delta() > 0 and vl <= _min_points_to_show:
+        #     log.debug("Max zoom bruh...")
+        #     return
 
-        if ev.delta() < 0 and vl >= len(chart._arrays[chart.name]) + 666:
-            log.debug("Min zoom bruh...")
-            return
+        # if (
+        #     ev.delta() < 0
+        #     and vl >= len(chart._flows[chart.name].shm.array) + 666
+        # ):
+        #     log.debug("Min zoom bruh...")
+        #     return
 
         # actual scaling factor
         s = 1.015 ** (ev.delta() * -1 / 20)  # self.state['wheelScaleFactor'])
@@ -568,6 +569,17 @@ class ChartView(ViewBox):
 
             self._resetTarget()
             self.scaleBy(s, focal)
+
+            # XXX: without this is seems as though sometimes
+            # when zooming in from far out (and maybe vice versa?)
+            # the signal isn't being fired enough since if you pan
+            # just after you'll see further downsampling code run
+            # (pretty noticeable on the OHLC ds curve) but with this
+            # that never seems to happen? Only question is how much this
+            # "double work" is causing latency when these missing event
+            # fires don't happen?
+            self.maybe_downsample_graphics()
+
             self.sigRangeChangedManually.emit(mask)
 
             # self._ic.set()
@@ -734,7 +746,7 @@ class ChartView(ViewBox):
 
         # flag to prevent triggering sibling charts from the same linked
         # set from recursion errors.
-        autoscale_linked_plots: bool = True,
+        autoscale_linked_plots: bool = False,
         name: Optional[str] = None,
         # autoscale_overlays: bool = False,
 
@@ -748,6 +760,7 @@ class ChartView(ViewBox):
 
         '''
         profiler = pg.debug.Profiler(
+            msg=f'`ChartView._set_yrange()`: `{self.name}`',
             disabled=not pg_profile_enabled(),
             gt=ms_slower_then,
             delayed=True,
@@ -777,9 +790,15 @@ class ChartView(ViewBox):
 
         # calculate max, min y values in viewable x-range from data.
         # Make sure min bars/datums on screen is adhered.
-        else:
-            br = bars_range or chart.bars_range()
-            profiler(f'got bars range: {br}')
+        # else:
+            # TODO: eventually we should point to the
+            # ``FlowsTable`` (or wtv) which should perform
+            # the group operations?
+
+            # flow = chart._flows[name or chart.name]
+            # br = bars_range or chart.bars_range()
+            # br = bars_range or chart.bars_range()
+            # profiler(f'got bars range: {br}')
 
             # TODO: maybe should be a method on the
             # chart widget/item?
@@ -795,7 +814,7 @@ class ChartView(ViewBox):
             #     for chart in plots:
             #         if chart and not chart._static_yrange:
             #             chart.cv._set_yrange(
-            #                 bars_range=br,
+            #                 # bars_range=br,
             #                 autoscale_linked_plots=False,
             #             )
             #     profiler('autoscaled linked plots')
@@ -809,6 +828,7 @@ class ChartView(ViewBox):
 
                 if yrange is None:
                     log.warning(f'No yrange provided for {self.name}!?')
+                    print(f"WTF NO YRANGE {self.name}")
                     return
 
             ylow, yhigh = yrange
@@ -830,6 +850,8 @@ class ChartView(ViewBox):
             self.setYRange(ylow, yhigh)
             profiler(f'set limits: {(ylow, yhigh)}')
 
+        profiler.finish()
+
     def enable_auto_yrange(
         self,
         src_vb: Optional[ChartView] = None,
@@ -843,17 +865,9 @@ class ChartView(ViewBox):
         if src_vb is None:
             src_vb = self
 
-        # such that when a linked chart changes its range
-        # this local view is also automatically changed and
-        # resized to data.
-        src_vb.sigXRangeChanged.connect(self._set_yrange)
-
         # splitter(s) resizing
         src_vb.sigResized.connect(self._set_yrange)
 
-        # mouse wheel doesn't emit XRangeChanged
-        src_vb.sigRangeChangedManually.connect(self._set_yrange)
-
         # TODO: a smarter way to avoid calling this needlessly?
         # 2 things i can think of:
         # - register downsample-able graphics specially and only
@@ -864,15 +878,16 @@ class ChartView(ViewBox):
             self.maybe_downsample_graphics
         )
 
-    def disable_auto_yrange(
-        self,
-    ) -> None:
+        # mouse wheel doesn't emit XRangeChanged
+        src_vb.sigRangeChangedManually.connect(self._set_yrange)
 
-        # self._chart._static_yrange = 'axis'
+        # src_vb.sigXRangeChanged.connect(self._set_yrange)
+        # src_vb.sigXRangeChanged.connect(
+        #     self.maybe_downsample_graphics
+        # )
+
+    def disable_auto_yrange(self) -> None:
 
-        self.sigXRangeChanged.disconnect(
-            self._set_yrange,
-        )
         self.sigResized.disconnect(
             self._set_yrange,
         )
@@ -883,6 +898,11 @@ class ChartView(ViewBox):
             self._set_yrange,
         )
 
+        # self.sigXRangeChanged.disconnect(self._set_yrange)
+        # self.sigXRangeChanged.disconnect(
+        #     self.maybe_downsample_graphics
+        # )
+
     def x_uppx(self) -> float:
         '''
         Return the "number of x units" within a single
@@ -890,7 +910,7 @@ class ChartView(ViewBox):
         graphics items which are our children.
 
         '''
-        graphics = list(self._chart._graphics.values())
+        graphics = [f.graphics for f in self._chart._flows.values()]
         if not graphics:
             return 0
 
@@ -903,23 +923,17 @@ class ChartView(ViewBox):
 
     def maybe_downsample_graphics(self):
 
-        uppx = self.x_uppx()
-        if (
-            # we probably want to drop this once we are "drawing in
-            # view" for downsampled flows..
-            uppx and uppx > 16
-            and self._ic is not None
-        ):
-            # don't bother updating since we're zoomed out bigly and
-            # in a pan-interaction, in which case we shouldn't be
-            # doing view-range based rendering (at least not yet).
-            # print(f'{uppx} exiting early!')
-            return
-
         profiler = pg.debug.Profiler(
+            msg=f'ChartView.maybe_downsample_graphics() for {self.name}',
             disabled=not pg_profile_enabled(),
-            gt=3,
-            delayed=True,
+
+            # XXX: important to avoid not seeing underlying
+            # ``.update_graphics_from_flow()`` nested profiling likely
+            # due to the way delaying works and garbage collection of
+            # the profiler in the delegated method calls.
+            delayed=False,
+            # gt=3,
+            gt=ms_slower_then,
         )
 
         # TODO: a faster single-loop-iterator way of doing this XD
@@ -928,19 +942,25 @@ class ChartView(ViewBox):
         plots = linked.subplots | {chart.name: chart}
         for chart_name, chart in plots.items():
             for name, flow in chart._flows.items():
-                graphics = flow.graphics
 
-                use_vr = False
-                if isinstance(graphics, BarItems):
-                    use_vr = True
+                if (
+                    not flow.render
+
+                    # XXX: super important to be aware of this.
+                    # or not flow.graphics.isVisible()
+                ):
+                    continue
 
                 # pass in no array which will read and render from the last
                 # passed array (normally provided by the display loop.)
-                chart.update_graphics_from_array(
+                chart.update_graphics_from_flow(
                     name,
-                    use_vr=use_vr,
+                    use_vr=True,
+
+                    # gets passed down into graphics obj
                     profiler=profiler,
                 )
+
                 profiler(f'range change updated {chart_name}:{name}')
 
         profiler.finish()

piker/ui/_ohlc.py

@@ -46,7 +46,7 @@ log = get_logger(__name__)
 
 def bar_from_ohlc_row(
     row: np.ndarray,
-    w: float
+    w: float = 0.43
 
 ) -> tuple[QLineF]:
     '''
@@ -158,8 +158,9 @@ def path_arrays_from_ohlc(
 
 def gen_qpath(
     data: np.ndarray,
-    start: int,  # XXX: do we need this?
-    w: float,
+    start: int = 0,  # XXX: do we need this?
+    # 0.5 is no overlap between arms, 1.0 is full overlap
+    w: float = 0.43,
     path: Optional[QtGui.QPainterPath] = None,
 
 ) -> QtGui.QPainterPath:
@@ -243,7 +244,7 @@ class BarItems(pg.GraphicsObject):
         self.fast_path = QtGui.QPainterPath()
 
         self._xrange: tuple[int, int]
-        self._yrange: tuple[float, float]
+        # self._yrange: tuple[float, float]
         self._vrange = None
 
         # TODO: don't render the full backing array each time
@@ -280,10 +281,10 @@ class BarItems(pg.GraphicsObject):
         # self.start_index = len(ohlc)
         index = ohlc['index']
         self._xrange = (index[0], index[-1])
-        self._yrange = (
-            np.nanmax(ohlc['high']),
-            np.nanmin(ohlc['low']),
-        )
+        # self._yrange = (
+        #     np.nanmax(ohlc['high']),
+        #     np.nanmin(ohlc['low']),
+        # )
 
         # up to last to avoid double draw of last bar
         self._last_bar_lines = bar_from_ohlc_row(last, self.w)
@@ -310,7 +311,7 @@ class BarItems(pg.GraphicsObject):
         self._pi.addItem(curve)
         self._ds_line = curve
 
-        self._ds_xrange = (index[0], index[-1])
+        # self._ds_xrange = (index[0], index[-1])
 
         # trigger render
         # https://doc.qt.io/qt-5/qgraphicsitem.html#update
@@ -324,289 +325,326 @@ class BarItems(pg.GraphicsObject):
         else:
             return 0
 
-    def update_from_array(
+    # def update_from_array(
+    #     self,
+
+    #     # full array input history
+    #     ohlc: np.ndarray,
+
+    #     # pre-sliced array data that's "in view"
+    #     ohlc_iv: np.ndarray,
+
+    #     view_range: Optional[tuple[int, int]] = None,
+    #     profiler: Optional[pg.debug.Profiler] = None,
+
+    # ) -> None:
+    #     '''
+    #     Update the last datum's bar graphic from input data array.
+
+    #     This routine should be interface compatible with
+    #     ``pg.PlotCurveItem.setData()``. Normally this method in
+    #     ``pyqtgraph`` seems to update all the data passed to the
+    #     graphics object, and then update/rerender, but here we're
+    #     assuming the prior graphics havent changed (OHLC history rarely
+    #     does) so this "should" be simpler and faster.
+
+    #     This routine should be made (transitively) as fast as possible.
+
+    #     '''
+    #     profiler = profiler or pg.debug.Profiler(
+    #         disabled=not pg_profile_enabled(),
+    #         gt=ms_slower_then,
+    #         delayed=True,
+    #     )
+
+    #     # index = self.start_index
+    #     istart, istop = self._xrange
+    #     # ds_istart, ds_istop = self._ds_xrange
+
+    #     index = ohlc['index']
+    #     first_index, last_index = index[0], index[-1]
+
+    #     # length = len(ohlc)
+    #     # prepend_length = istart - first_index
+    #     # append_length = last_index - istop
+
+    #     # ds_prepend_length = ds_istart - first_index
+    #     # ds_append_length = last_index - ds_istop
+
+    #     flip_cache = False
+
+    #     x_gt = 16
+    #     if self._ds_line:
+    #         uppx = self._ds_line.x_uppx()
+    #     else:
+    #         uppx = 0
+
+    #     should_line = self._in_ds
+    #     if (
+    #         self._in_ds
+    #         and uppx < x_gt
+    #     ):
+    #         should_line = False
+
+    #     elif (
+    #         not self._in_ds
+    #         and uppx >= x_gt
+    #     ):
+    #         should_line = True
+
+    #     profiler('ds logic complete')
+
+    #     if should_line:
+    #         # update the line graphic
+    #         # x, y = self._ds_line_xy = ohlc_flatten(ohlc_iv)
+    #         x, y = self._ds_line_xy = ohlc_flatten(ohlc)
+    #         x_iv, y_iv = self._ds_line_xy = ohlc_flatten(ohlc_iv)
+    #         profiler('flattening bars to line')
+
+    #         # TODO: we should be diffing the amount of new data which
+    #         # needs to be downsampled. Ideally we actually are just
+    #         # doing all the ds-ing in sibling actors so that the data
+    #         # can just be read and rendered to graphics on events of our
+    #         # choice.
+    #         # diff = do_diff(ohlc, new_bit)
+    #         curve = self._ds_line
+    #         curve.update_from_array(
+    #             x=x,
+    #             y=y,
+    #             x_iv=x_iv,
+    #             y_iv=y_iv,
+    #             view_range=None,  # hack
+    #             profiler=profiler,
+    #         )
+    #         profiler('updated ds line')
+
+    #         if not self._in_ds:
+    #             # hide bars and show line
+    #             self.hide()
+    #             # XXX: is this actually any faster?
+    #             # self._pi.removeItem(self)
+
+    #             # TODO: a `.ui()` log level?
+    #             log.info(
+    #                 f'downsampling to line graphic {self._name}'
+    #             )
+
+    #             # self._pi.addItem(curve)
+    #             curve.show()
+    #             curve.update()
+    #             self._in_ds = True
+
+    #         # stop here since we don't need to update bars path any more
+    #         # as we delegate to the downsample line with updates.
+
+    #     else:
+    #         # we should be in bars mode
+
+    #         if self._in_ds:
+    #             # flip back to bars graphics and hide the downsample line.
+    #             log.info(f'showing bars graphic {self._name}')
+
+    #             curve = self._ds_line
+    #             curve.hide()
+    #             # self._pi.removeItem(curve)
+
+    #             # XXX: is this actually any faster?
+    #             # self._pi.addItem(self)
+    #             self.show()
+    #             self._in_ds = False
+
+    #         # generate in_view path
+    #         self.path = gen_qpath(
+    #             ohlc_iv,
+    #             0,
+    #             self.w,
+    #             # path=self.path,
+    #         )
+
+    #         # TODO: to make the downsampling faster
+    #         # - allow mapping only a range of lines thus only drawing as
+    #         #   many bars as exactly specified.
+    #         # - move ohlc "flattening" to a shmarr
+    #         # - maybe move all this embedded logic to a higher
+    #         #   level type?
+
+    #         # if prepend_length:
+    #         #     # new history was added and we need to render a new path
+    #         #     prepend_bars = ohlc[:prepend_length]
+
+    #         # if ds_prepend_length:
+    #         #     ds_prepend_bars = ohlc[:ds_prepend_length]
+    #         #     pre_x, pre_y = ohlc_flatten(ds_prepend_bars)
+    #         #     fx = np.concatenate((pre_x, fx))
+    #         #     fy = np.concatenate((pre_y, fy))
+    #         #     profiler('ds line prepend diff complete')
+
+    #         # if append_length:
+    #         #     # generate new graphics to match provided array
+    #         #     # path appending logic:
+    #         #     # we need to get the previous "current bar(s)" for the time step
+    #         #     # and convert it to a sub-path to append to the historical set
+    #         #     # new_bars = ohlc[istop - 1:istop + append_length - 1]
+    #         #     append_bars = ohlc[-append_length - 1:-1]
+    #         #     # print(f'ohlc bars to append size: {append_bars.size}\n')
+
+    #         # if ds_append_length:
+    #         #     ds_append_bars = ohlc[-ds_append_length - 1:-1]
+    #         #     post_x, post_y = ohlc_flatten(ds_append_bars)
+    #         #     print(
+    #         #         f'ds curve to append sizes: {(post_x.size, post_y.size)}'
+    #         #     )
+    #         #     fx = np.concatenate((fx, post_x))
+    #         #     fy = np.concatenate((fy, post_y))
+
+    #         #     profiler('ds line append diff complete')
+
+    #         profiler('array diffs complete')
+
+    #         # does this work?
+    #         last = ohlc[-1]
+    #         # fy[-1] = last['close']
+
+    #         # # incremental update and cache line datums
+    #         # self._ds_line_xy = fx, fy
+
+    #         # maybe downsample to line
+    #         # ds = self.maybe_downsample()
+    #         # if ds:
+    #         #     # if we downsample to a line don't bother with
+    #         #     # any more path generation / updates
+    #         #     self._ds_xrange = first_index, last_index
+    #         #     profiler('downsampled to line')
+    #         #     return
+
+    #         # print(in_view.size)
+
+    #         # if self.path:
+    #         #     self.path = path
+    #         #     self.path.reserve(path.capacity())
+    #         #     self.path.swap(path)
+
+    #         # path updates
+    #         # if prepend_length:
+    #         #     # XXX: SOMETHING IS MAYBE FISHY HERE what with the old_path
+    #         #     # y value not matching the first value from
+    #         #     # ohlc[prepend_length + 1] ???
+    #         #     prepend_path = gen_qpath(prepend_bars, 0, self.w)
+    #         #     old_path = self.path
+    #         #     self.path = prepend_path
+    #         #     self.path.addPath(old_path)
+    #         #     profiler('path PREPEND')
+
+    #         # if append_length:
+    #         #     append_path = gen_qpath(append_bars, 0, self.w)
+
+    #         #     self.path.moveTo(
+    #         #         float(istop - self.w),
+    #         #         float(append_bars[0]['open'])
+    #         #     )
+    #         #     self.path.addPath(append_path)
+
+    #         #     profiler('path APPEND')
+    #         #     fp = self.fast_path
+    #         #     if fp is None:
+    #         #         self.fast_path = append_path
+
+    #         #     else:
+    #         #         fp.moveTo(
+    #         #             float(istop - self.w), float(new_bars[0]['open'])
+    #         #         )
+    #         #         fp.addPath(append_path)
+
+    #         #     self.setCacheMode(QtWidgets.QGraphicsItem.NoCache)
+    #         #     flip_cache = True
+
+    #         self._xrange = first_index, last_index
+
+    #         # trigger redraw despite caching
+    #         self.prepareGeometryChange()
+
+    #         self.draw_last(last)
+
+    #         # # generate new lines objects for updatable "current bar"
+    #         # self._last_bar_lines = bar_from_ohlc_row(last, self.w)
+
+    #         # # last bar update
+    #         # i, o, h, l, last, v = last[
+    #         #     ['index', 'open', 'high', 'low', 'close', 'volume']
+    #         # ]
+    #         # # assert i == self.start_index - 1
+    #         # # assert i == last_index
+    #         # body, larm, rarm = self._last_bar_lines
+
+    #         # # XXX: is there a faster way to modify this?
+    #         # rarm.setLine(rarm.x1(), last, rarm.x2(), last)
+
+    #         # # writer is responsible for changing open on "first" volume of bar
+    #         # larm.setLine(larm.x1(), o, larm.x2(), o)
+
+    #         # if l != h:  # noqa
+
+    #         #     if body is None:
+    #         #         body = self._last_bar_lines[0] = QLineF(i, l, i, h)
+    #         #     else:
+    #         #         # update body
+    #         #         body.setLine(i, l, i, h)
+
+    #         #     # XXX: pretty sure this is causing an issue where the bar has
+    #         #     # a large upward move right before the next sample and the body
+    #         #     # is getting set to None since the next bar is flat but the shm
+    #         #     # array index update wasn't read by the time this code runs. Iow
+    #         #     # we're doing this removal of the body for a bar index that is
+    #         #     # now out of date / from some previous sample. It's weird
+    #         #     # though because i've seen it do this to bars i - 3 back?
+
+    #         profiler('last bar set')
+
+    #         self.update()
+    #         profiler('.update()')
+
+    #         if flip_cache:
+    #             self.setCacheMode(QtWidgets.QGraphicsItem.DeviceCoordinateCache)
+
+    #         # profiler.finish()
+
+    def draw_last(
         self,
-
-        # full array input history
-        ohlc: np.ndarray,
-
-        # pre-sliced array data that's "in view"
-        ohlc_iv: np.ndarray,
-
-        view_range: Optional[tuple[int, int]] = None,
-        profiler: Optional[pg.debug.Profiler] = None,
+        last: np.ndarray,
 
     ) -> None:
-        '''
-        Update the last datum's bar graphic from input data array.
+        # generate new lines objects for updatable "current bar"
+        self._last_bar_lines = bar_from_ohlc_row(last, self.w)
 
-        This routine should be interface compatible with
-        ``pg.PlotCurveItem.setData()``. Normally this method in
-        ``pyqtgraph`` seems to update all the data passed to the
-        graphics object, and then update/rerender, but here we're
-        assuming the prior graphics havent changed (OHLC history rarely
-        does) so this "should" be simpler and faster.
+        # last bar update
+        i, o, h, l, last, v = last[
+            ['index', 'open', 'high', 'low', 'close', 'volume']
+        ]
+        # assert i == self.start_index - 1
+        # assert i == last_index
+        body, larm, rarm = self._last_bar_lines
 
-        This routine should be made (transitively) as fast as possible.
+        # XXX: is there a faster way to modify this?
+        rarm.setLine(rarm.x1(), last, rarm.x2(), last)
 
-        '''
-        profiler = profiler or pg.debug.Profiler(
-            disabled=not pg_profile_enabled(),
-            gt=ms_slower_then,
-            delayed=True,
-        )
+        # writer is responsible for changing open on "first" volume of bar
+        larm.setLine(larm.x1(), o, larm.x2(), o)
 
-        # index = self.start_index
-        istart, istop = self._xrange
-        ds_istart, ds_istop = self._ds_xrange
+        if l != h:  # noqa
 
-        index = ohlc['index']
-        first_index, last_index = index[0], index[-1]
+            if body is None:
+                body = self._last_bar_lines[0] = QLineF(i, l, i, h)
+            else:
+                # update body
+                body.setLine(i, l, i, h)
 
-        # length = len(ohlc)
-        # prepend_length = istart - first_index
-        # append_length = last_index - istop
-
-        # ds_prepend_length = ds_istart - first_index
-        # ds_append_length = last_index - ds_istop
-
-        flip_cache = False
-
-        x_gt = 16
-        if self._ds_line:
-            uppx = self._ds_line.x_uppx()
-        else:
-            uppx = 0
-
-        should_line = self._in_ds
-        if (
-            self._in_ds
-            and uppx < x_gt
-        ):
-            should_line = False
-
-        elif (
-            not self._in_ds
-            and uppx >= x_gt
-        ):
-            should_line = True
-
-        profiler('ds logic complete')
-
-        if should_line:
-            # update the line graphic
-            # x, y = self._ds_line_xy = ohlc_flatten(ohlc_iv)
-            x, y = self._ds_line_xy = ohlc_flatten(ohlc)
-            x_iv, y_iv = self._ds_line_xy = ohlc_flatten(ohlc_iv)
-            profiler('flattening bars to line')
-
-            # TODO: we should be diffing the amount of new data which
-            # needs to be downsampled. Ideally we actually are just
-            # doing all the ds-ing in sibling actors so that the data
-            # can just be read and rendered to graphics on events of our
-            # choice.
-            # diff = do_diff(ohlc, new_bit)
-            curve = self._ds_line
-            curve.update_from_array(
-                x=x,
-                y=y,
-                x_iv=x_iv,
-                y_iv=y_iv,
-                view_range=None,  # hack
-                profiler=profiler,
-            )
-            profiler('updated ds line')
-
-            if not self._in_ds:
-                # hide bars and show line
-                self.hide()
-                # XXX: is this actually any faster?
-                # self._pi.removeItem(self)
-
-                # TODO: a `.ui()` log level?
-                log.info(
-                    f'downsampling to line graphic {self._name}'
-                )
-
-                # self._pi.addItem(curve)
-                curve.show()
-                curve.update()
-                self._in_ds = True
-
-            # stop here since we don't need to update bars path any more
-            # as we delegate to the downsample line with updates.
-            profiler.finish()
-            # print('terminating early')
-            return
-
-        else:
-            # we should be in bars mode
-
-            if self._in_ds:
-                # flip back to bars graphics and hide the downsample line.
-                log.info(f'showing bars graphic {self._name}')
-
-                curve = self._ds_line
-                curve.hide()
-                # self._pi.removeItem(curve)
-
-                # XXX: is this actually any faster?
-                # self._pi.addItem(self)
-                self.show()
-                self._in_ds = False
-
-            # generate in_view path
-            self.path = gen_qpath(
-                ohlc_iv,
-                0,
-                self.w,
-                # path=self.path,
-            )
-
-            # TODO: to make the downsampling faster
-            # - allow mapping only a range of lines thus only drawing as
-            #   many bars as exactly specified.
-            # - move ohlc "flattening" to a shmarr
-            # - maybe move all this embedded logic to a higher
-            #   level type?
-
-            # if prepend_length:
-            #     # new history was added and we need to render a new path
-            #     prepend_bars = ohlc[:prepend_length]
-
-            # if ds_prepend_length:
-            #     ds_prepend_bars = ohlc[:ds_prepend_length]
-            #     pre_x, pre_y = ohlc_flatten(ds_prepend_bars)
-            #     fx = np.concatenate((pre_x, fx))
-            #     fy = np.concatenate((pre_y, fy))
-            #     profiler('ds line prepend diff complete')
-
-            # if append_length:
-            #     # generate new graphics to match provided array
-            #     # path appending logic:
-            #     # we need to get the previous "current bar(s)" for the time step
-            #     # and convert it to a sub-path to append to the historical set
-            #     # new_bars = ohlc[istop - 1:istop + append_length - 1]
-            #     append_bars = ohlc[-append_length - 1:-1]
-            #     # print(f'ohlc bars to append size: {append_bars.size}\n')
-
-            # if ds_append_length:
-            #     ds_append_bars = ohlc[-ds_append_length - 1:-1]
-            #     post_x, post_y = ohlc_flatten(ds_append_bars)
-            #     print(
-            #         f'ds curve to append sizes: {(post_x.size, post_y.size)}'
-            #     )
-            #     fx = np.concatenate((fx, post_x))
-            #     fy = np.concatenate((fy, post_y))
-
-            #     profiler('ds line append diff complete')
-
-            profiler('array diffs complete')
-
-            # does this work?
-            last = ohlc[-1]
-            # fy[-1] = last['close']
-
-            # # incremental update and cache line datums
-            # self._ds_line_xy = fx, fy
-
-            # maybe downsample to line
-            # ds = self.maybe_downsample()
-            # if ds:
-            #     # if we downsample to a line don't bother with
-            #     # any more path generation / updates
-            #     self._ds_xrange = first_index, last_index
-            #     profiler('downsampled to line')
-            #     return
-
-            # print(in_view.size)
-
-            # if self.path:
-            #     self.path = path
-            #     self.path.reserve(path.capacity())
-            #     self.path.swap(path)
-
-            # path updates
-            # if prepend_length:
-            #     # XXX: SOMETHING IS MAYBE FISHY HERE what with the old_path
-            #     # y value not matching the first value from
-            #     # ohlc[prepend_length + 1] ???
-            #     prepend_path = gen_qpath(prepend_bars, 0, self.w)
-            #     old_path = self.path
-            #     self.path = prepend_path
-            #     self.path.addPath(old_path)
-            #     profiler('path PREPEND')
-
-            # if append_length:
-            #     append_path = gen_qpath(append_bars, 0, self.w)
-
-            #     self.path.moveTo(
-            #         float(istop - self.w),
-            #         float(append_bars[0]['open'])
-            #     )
-            #     self.path.addPath(append_path)
-
-            #     profiler('path APPEND')
-            #     fp = self.fast_path
-            #     if fp is None:
-            #         self.fast_path = append_path
-
-            #     else:
-            #         fp.moveTo(
-            #             float(istop - self.w), float(new_bars[0]['open'])
-            #         )
-            #         fp.addPath(append_path)
-
-            #     self.setCacheMode(QtWidgets.QGraphicsItem.NoCache)
-            #     flip_cache = True
-
-            self._xrange = first_index, last_index
-
-            # trigger redraw despite caching
-            self.prepareGeometryChange()
-
-            # generate new lines objects for updatable "current bar"
-            self._last_bar_lines = bar_from_ohlc_row(last, self.w)
-
-            # last bar update
-            i, o, h, l, last, v = last[
-                ['index', 'open', 'high', 'low', 'close', 'volume']
-            ]
-            # assert i == self.start_index - 1
-            # assert i == last_index
-            body, larm, rarm = self._last_bar_lines
-
-            # XXX: is there a faster way to modify this?
-            rarm.setLine(rarm.x1(), last, rarm.x2(), last)
-
-            # writer is responsible for changing open on "first" volume of bar
-            larm.setLine(larm.x1(), o, larm.x2(), o)
-
-            if l != h:  # noqa
-
-                if body is None:
-                    body = self._last_bar_lines[0] = QLineF(i, l, i, h)
-                else:
-                    # update body
-                    body.setLine(i, l, i, h)
-
-                # XXX: pretty sure this is causing an issue where the bar has
-                # a large upward move right before the next sample and the body
-                # is getting set to None since the next bar is flat but the shm
-                # array index update wasn't read by the time this code runs. Iow
-                # we're doing this removal of the body for a bar index that is
-                # now out of date / from some previous sample. It's weird
-                # though because i've seen it do this to bars i - 3 back?
-
-            profiler('last bar set')
-
-            self.update()
-            profiler('.update()')
-
-            if flip_cache:
-                self.setCacheMode(QtWidgets.QGraphicsItem.DeviceCoordinateCache)
-
-            profiler.finish()
+            # XXX: pretty sure this is causing an issue where the bar has
+            # a large upward move right before the next sample and the body
+            # is getting set to None since the next bar is flat but the shm
+            # array index update wasn't read by the time this code runs. Iow
+            # we're doing this removal of the body for a bar index that is
+            # now out of date / from some previous sample. It's weird
+            # though because i've seen it do this to bars i - 3 back?
 
     def boundingRect(self):
         # Qt docs: https://doc.qt.io/qt-5/qgraphicsitem.html#boundingRect

requirements.txt

@@ -7,3 +7,7 @@
 # pin this to a dev branch that we have more control over especially
 # as more graphics stuff gets hashed out.
 -e git+https://github.com/pikers/pyqtgraph.git@piker_pin#egg=pyqtgraph
+
+
+# our async client for ``marketstore`` (the tsdb)
+-e git+https://github.com/pikers/anyio-marketstore.git@master#egg=anyio-marketstore

scripts/ib_data_reset.py

@@ -1,5 +1,5 @@
 # piker: trading gear for hackers
-# Copyright (C) Tyler Goodlet (in stewardship for piker0)
+# Copyright (C) Tyler Goodlet (in stewardship for pikers)
 
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License as published by
@@ -30,11 +30,13 @@ orig_win_id = t.find_focused().window
 # for tws
 win_names: list[str] = [
     'Interactive Brokers',  # tws running in i3
-    'IB Gateway.',  # gw running in i3
+    'IB Gateway',  # gw running in i3
+    # 'IB',  # gw running in i3 (newer version?)
 ]
 
 for name in win_names:
-    results = t.find_named(name)
+    results = t.find_titled(name)
+    print(f'results for {name}: {results}')
     if results:
         con = results[0]
         print(f'Resetting data feed for {name}')
@@ -47,22 +49,32 @@ for name in win_names:
         # https://github.com/rr-/pyxdotool
         # https://github.com/ShaneHutter/pyxdotool
         # https://github.com/cphyc/pyxdotool
-        subprocess.call([
-            'xdotool',
-            'windowactivate', '--sync', win_id,
 
-            # move mouse to bottom left of window (where there should
-            # be nothing to click).
-            'mousemove_relative', '--sync', str(w-4), str(h-4),
+        # TODO: only run the reconnect (2nd) kc on a detected
+        # disconnect?
+        for key_combo, timeout in [
+            # only required if we need a connection reset.
+            # ('ctrl+alt+r', 12),
+            # data feed reset.
+            ('ctrl+alt+f', 6)
+        ]:
+            subprocess.call([
+                'xdotool',
+                'windowactivate', '--sync', win_id,
 
-            # NOTE: we may need to stick a `--retry 3` in here..
-            'click', '--window', win_id, '--repeat', '3', '1',
+                # move mouse to bottom left of window (where there should
+                # be nothing to click).
+                'mousemove_relative', '--sync', str(w-4), str(h-4),
 
-            # hackzorzes
-            'key', 'ctrl+alt+f',
-            ],
-            timeout=1,
-        )
+                # NOTE: we may need to stick a `--retry 3` in here..
+                'click', '--window', win_id,
+                '--repeat', '3', '1',
+
+                # hackzorzes
+                'key', key_combo,
+                ],
+                timeout=timeout,
+            )
 
 # re-activate and focus original window
 subprocess.call([

setup.py

@@ -77,6 +77,14 @@ setup(
         # tsdbs
         'pymarketstore',
     ],
+    extras_require={
+
+        # serialization
+        'tsdb': [
+            'docker',
+        ],
+
+    },
     tests_require=['pytest'],
     python_requires=">=3.9",  # literally for ``datetime.datetime.fromisoformat``...
     keywords=["async", "trading", "finance", "quant", "charting"],