tractor/tractor/_runtime.py

1892 lines
64 KiB
Python

# tractor: structured concurrent "actors".
# Copyright 2018-eternity Tyler Goodlet.
# 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/>.
'''
The fundamental core machinery implementing every "actor"
including the process-local, or "python-interpreter (aka global)
singleton) `Actor` primitive(s) and its internal `trio` machinery
implementing the low level runtime system supporting the
discovery, communication, spawning, supervision and cancellation
of other actors in a hierarchincal process tree.
The runtime's main entry point: `async_main()` opens the top level
supervision and service `trio.Nursery`s which manage the tasks responsible
for running all lower level spawning, supervision and msging layers:
- lowlevel transport-protocol init and persistent connectivity on
top of `._ipc` primitives; the transport layer.
- bootstrapping of connection/runtime config from the spawning
parent (actor).
- starting and supervising IPC-channel msg processing loops around
tranport connections from parent/peer actors in order to deliver
SC-transitive RPC via scheduling of `trio` tasks.
- registration of newly spawned actors with the discovery sys.
'''
from __future__ import annotations
from contextlib import (
ExitStack,
)
from collections import defaultdict
from functools import partial
from itertools import chain
import importlib
import importlib.util
from pprint import pformat
import signal
import sys
from typing import (
Any,
Callable,
TYPE_CHECKING,
)
import uuid
from types import ModuleType
import os
import warnings
import trio
from trio import (
CancelScope,
Nursery,
TaskStatus,
)
from .msg import NamespacePath
from ._ipc import Channel
from ._context import (
mk_context,
Context,
)
from .log import get_logger
from ._exceptions import (
unpack_error,
ModuleNotExposed,
ContextCancelled,
TransportClosed,
)
from .devx import (
# pause,
maybe_wait_for_debugger,
_debug,
)
from ._discovery import get_registry
from ._portal import Portal
from . import _state
from . import _mp_fixup_main
from ._rpc import (
process_messages,
try_ship_error_to_remote,
)
if TYPE_CHECKING:
from ._supervise import ActorNursery
log = get_logger('tractor')
def _get_mod_abspath(module: ModuleType) -> str:
return os.path.abspath(module.__file__)
class Actor:
'''
The fundamental "runtime" concurrency primitive.
An *actor* is the combination of a regular Python process executing
a ``trio`` task tree, communicating with other actors through
"memory boundary portals" - which provide a native async API around
IPC transport "channels" which themselves encapsulate various
(swappable) network protocols.
Each "actor" is ``trio.run()`` scheduled "runtime" composed of
many concurrent tasks in a single thread. The "runtime" tasks
conduct a slew of low(er) level functions to make it possible
for message passing between actors as well as the ability to
create new actors (aka new "runtimes" in new processes which
are supervised via a nursery construct). Each task which sends
messages to a task in a "peer" (not necessarily a parent-child,
depth hierarchy) is able to do so via an "address", which maps
IPC connections across memory boundaries, and a task request id
which allows for per-actor tasks to send and receive messages
to specific peer-actor tasks with which there is an ongoing
RPC/IPC dialog.
'''
# ugh, we need to get rid of this and replace with a "registry" sys
# https://github.com/goodboy/tractor/issues/216
is_arbiter: bool = False
@property
def is_registrar(self) -> bool:
return self.is_arbiter
msg_buffer_size: int = 2**6
# nursery placeholders filled in by `async_main()` after fork
_root_n: Nursery|None = None
_service_n: Nursery|None = None
_server_n: Nursery|None = None
# Information about `__main__` from parent
_parent_main_data: dict[str, str]
_parent_chan_cs: CancelScope|None = None
# syncs for setup/teardown sequences
_server_down: trio.Event|None = None
# user toggled crash handling (including monkey-patched in
# `trio.open_nursery()` via `.trionics._supervisor` B)
_debug_mode: bool = False
# if started on ``asycio`` running ``trio`` in guest mode
_infected_aio: bool = False
# _ans: dict[
# tuple[str, str],
# list[ActorNursery],
# ] = {}
# Process-global stack closed at end on actor runtime teardown.
# NOTE: this is currently an undocumented public api.
lifetime_stack: ExitStack = ExitStack()
def __init__(
self,
name: str,
*,
enable_modules: list[str] = [],
uid: str|None = None,
loglevel: str|None = None,
registry_addrs: list[tuple[str, int]]|None = None,
spawn_method: str|None = None,
# TODO: remove!
arbiter_addr: tuple[str, int]|None = None,
) -> None:
'''
This constructor is called in the parent actor **before** the spawning
phase (aka before a new process is executed).
'''
self.name = name
self.uid = (
name,
uid or str(uuid.uuid4())
)
self._cancel_complete = trio.Event()
self._cancel_called_by_remote: tuple[str, tuple]|None = None
self._cancel_called: bool = False
# retreive and store parent `__main__` data which
# will be passed to children
self._parent_main_data = _mp_fixup_main._mp_figure_out_main()
# always include debugging tools module
enable_modules.append('tractor.devx._debug')
self.enable_modules: dict[str, str] = {}
for name in enable_modules:
mod: ModuleType = importlib.import_module(name)
self.enable_modules[name] = _get_mod_abspath(mod)
self._mods: dict[str, ModuleType] = {}
self.loglevel: str = loglevel
if arbiter_addr is not None:
warnings.warn(
'`Actor(arbiter_addr=<blah>)` is now deprecated.\n'
'Use `registry_addrs: list[tuple]` instead.',
DeprecationWarning,
stacklevel=2,
)
registry_addrs: list[tuple[str, int]] = [arbiter_addr]
# marked by the process spawning backend at startup
# will be None for the parent most process started manually
# by the user (currently called the "arbiter")
self._spawn_method: str = spawn_method
self._peers: defaultdict = defaultdict(list)
self._peer_connected: dict[tuple[str, str], trio.Event] = {}
self._no_more_peers = trio.Event()
self._no_more_peers.set()
self._ongoing_rpc_tasks = trio.Event()
self._ongoing_rpc_tasks.set()
# (chan, cid) -> (cancel_scope, func)
self._rpc_tasks: dict[
tuple[Channel, str],
tuple[Context, Callable, trio.Event]
] = {}
# map {actor uids -> Context}
self._contexts: dict[
tuple[
tuple[str, str], # .uid
str, # .cid
str, # .side
],
Context
] = {}
self._listeners: list[trio.abc.Listener] = []
self._parent_chan: Channel|None = None
self._forkserver_info: tuple|None = None
self._actoruid2nursery: dict[
tuple[str, str],
ActorNursery|None,
] = {} # type: ignore # noqa
# when provided, init the registry addresses property from
# input via the validator.
self._reg_addrs: list[tuple[str, int]] = []
if registry_addrs:
self.reg_addrs: list[tuple[str, int]] = registry_addrs
_state._runtime_vars['_registry_addrs'] = registry_addrs
@property
def reg_addrs(self) -> list[tuple[str, int]]:
'''
List of (socket) addresses for all known (and contactable)
registry actors.
'''
return self._reg_addrs
@reg_addrs.setter
def reg_addrs(
self,
addrs: list[tuple[str, int]],
) -> None:
if not addrs:
log.warning(
'Empty registry address list is invalid:\n'
f'{addrs}'
)
return
# always sanity check the input list since it's critical
# that addrs are correct for discovery sys operation.
for addr in addrs:
if not isinstance(addr, tuple):
raise ValueError(
'Expected `Actor.reg_addrs: list[tuple[str, int]]`\n'
f'Got {addrs}'
)
self._reg_addrs = addrs
async def wait_for_peer(
self, uid: tuple[str, str]
) -> tuple[trio.Event, Channel]:
'''
Wait for a connection back from a spawned actor with a `uid`
using a `trio.Event` for sync.
'''
log.runtime(f"Waiting for peer {uid} to connect")
event = self._peer_connected.setdefault(uid, trio.Event())
await event.wait()
log.runtime(f"{uid} successfully connected back to us")
return event, self._peers[uid][-1]
def load_modules(
self,
debug_mode: bool = False,
) -> None:
'''
Load enabled RPC py-modules locally (after process fork/spawn).
Since this actor may be spawned on a different machine from
the original nursery we need to try and load the local module
code (presuming it exists).
'''
try:
if self._spawn_method == 'trio':
parent_data = self._parent_main_data
if 'init_main_from_name' in parent_data:
_mp_fixup_main._fixup_main_from_name(
parent_data['init_main_from_name'])
elif 'init_main_from_path' in parent_data:
_mp_fixup_main._fixup_main_from_path(
parent_data['init_main_from_path'])
for modpath, filepath in self.enable_modules.items():
# XXX append the allowed module to the python path which
# should allow for relative (at least downward) imports.
sys.path.append(os.path.dirname(filepath))
log.runtime(f"Attempting to import {modpath}@{filepath}")
mod = importlib.import_module(modpath)
self._mods[modpath] = mod
if modpath == '__main__':
self._mods['__mp_main__'] = mod
except ModuleNotFoundError:
# it is expected the corresponding `ModuleNotExposed` error
# will be raised later
log.error(
f"Failed to import {modpath} in {self.name}"
)
raise
def _get_rpc_func(self, ns, funcname):
'''
Try to lookup and return a target RPC func from the
post-fork enabled module set.
'''
try:
return getattr(self._mods[ns], funcname)
except KeyError as err:
mne = ModuleNotExposed(*err.args)
if ns == '__main__':
modpath = '__name__'
else:
modpath = f"'{ns}'"
msg = (
"\n\nMake sure you exposed the target module, `{ns}`, "
"using:\n"
"ActorNursery.start_actor(<name>, enable_modules=[{mod}])"
).format(
ns=ns,
mod=modpath,
)
mne.msg += msg
raise mne
async def _stream_handler(
self,
stream: trio.SocketStream,
) -> None:
'''
Entry point for new inbound IPC connections on a specific
transport server.
'''
self._no_more_peers = trio.Event() # unset by making new
chan = Channel.from_stream(stream)
their_uid: tuple[str, str]|None = chan.uid
con_msg: str = ''
if their_uid:
# NOTE: `.uid` is only set after first contact
con_msg = (
'IPC Re-connection from already known peer? '
)
else:
con_msg = (
'New IPC connection to us '
)
con_msg += (
f'<= @{chan.raddr}\n'
f'|_{chan}\n'
# f' |_@{chan.raddr}\n\n'
)
# send/receive initial handshake response
try:
uid: tuple|None = await self._do_handshake(chan)
except (
# we need this for ``msgspec`` for some reason?
# for now, it's been put in the stream backend.
# trio.BrokenResourceError,
# trio.ClosedResourceError,
TransportClosed,
):
# XXX: This may propagate up from ``Channel._aiter_recv()``
# and ``MsgpackStream._inter_packets()`` on a read from the
# stream particularly when the runtime is first starting up
# inside ``open_root_actor()`` where there is a check for
# a bound listener on the "arbiter" addr. the reset will be
# because the handshake was never meant took place.
log.warning(
con_msg
+
' -> But failed to handshake? Ignoring..\n'
)
return
con_msg += (
f' -> Handshake with actor `{uid[0]}[{uid[1][-6:]}]` complete\n'
)
# IPC connection tracking for both peers and new children:
# - if this is a new channel to a locally spawned
# sub-actor there will be a spawn wait even registered
# by a call to `.wait_for_peer()`.
# - if a peer is connecting no such event will exit.
event: trio.Event|None = self._peer_connected.pop(
uid,
None,
)
if event:
con_msg += (
' -> Waking subactor spawn waiters: '
f'{event.statistics().tasks_waiting}\n'
f' -> Registered IPC chan for child actor {uid}@{chan.raddr}\n'
# f' {event}\n'
# f' |{event.statistics()}\n'
)
# wake tasks waiting on this IPC-transport "connect-back"
event.set()
else:
con_msg += (
f' -> Registered IPC chan for peer actor {uid}@{chan.raddr}\n'
) # type: ignore
chans: list[Channel] = self._peers[uid]
# if chans:
# # TODO: re-use channels for new connections instead
# # of always new ones?
# # => will require changing all the discovery funcs..
# append new channel
# TODO: can we just use list-ref directly?
chans.append(chan)
log.runtime(con_msg)
# Begin channel management - respond to remote requests and
# process received reponses.
disconnected: bool = False
try:
disconnected: bool = await process_messages(
self,
chan,
)
except trio.Cancelled:
log.cancel(
'IPC transport msg loop was cancelled for \n'
f'|_{chan}\n'
)
raise
finally:
local_nursery: (
ActorNursery|None
) = self._actoruid2nursery.get(uid)
# This is set in ``Portal.cancel_actor()``. So if
# the peer was cancelled we try to wait for them
# to tear down their side of the connection before
# moving on with closing our own side.
if local_nursery:
if chan._cancel_called:
log.cancel(
'Waiting on cancel request to peer\n'
f'`Portal.cancel_actor()` => {chan.uid}\n'
)
# XXX: this is a soft wait on the channel (and its
# underlying transport protocol) to close from the
# remote peer side since we presume that any channel
# which is mapped to a sub-actor (i.e. it's managed by
# one of our local nurseries) has a message is sent to
# the peer likely by this actor (which is now in
# a cancelled condition) when the local runtime here is
# now cancelled while (presumably) in the middle of msg
# loop processing.
with trio.move_on_after(0.5) as cs:
cs.shield = True
# attempt to wait for the far end to close the
# channel and bail after timeout (a 2-generals
# problem on closure).
assert chan.transport
async for msg in chan.transport.drain():
# try to deliver any lingering msgs
# before we destroy the channel.
# This accomplishes deterministic
# ``Portal.cancel_actor()`` cancellation by
# making sure any RPC response to that call is
# delivered the local calling task.
# TODO: factor this into a helper?
log.warning(
'Draining msg from disconnected peer\n'
f'{chan.uid}\n'
f'|_{chan}\n'
f' |_{chan.transport}\n\n'
f'{pformat(msg)}\n'
)
cid = msg.get('cid')
if cid:
# deliver response to local caller/waiter
await self._push_result(
chan,
cid,
msg,
)
# NOTE: when no call to `open_root_actor()` was
# made, we implicitly make that call inside
# the first `.open_nursery()`, in this case we
# can assume that we are the root actor and do
# not have to wait for the nursery-enterer to
# exit before shutting down the actor runtime.
#
# see matching note inside `._supervise.open_nursery()`
if not local_nursery._implicit_runtime_started:
log.runtime(
'Waiting on local actor nursery to exit..\n'
f'|_{local_nursery}\n'
)
await local_nursery.exited.wait()
if (
cs.cancelled_caught
and not local_nursery._implicit_runtime_started
):
log.warning(
'Failed to exit local actor nursery?\n'
f'|_{local_nursery}\n'
)
# await _debug.pause()
if disconnected:
# if the transport died and this actor is still
# registered within a local nursery, we report
# that the IPC layer may have failed
# unexpectedly since it may be the cause of
# other downstream errors.
entry = local_nursery._children.get(uid)
if entry:
proc: trio.Process
_, proc, _ = entry
if (
(poll := getattr(proc, 'poll', None))
and poll() is None
):
log.cancel(
f'Peer IPC broke but subproc is alive?\n\n'
f'<=x {chan.uid}@{chan.raddr}\n'
f' |_{proc}\n'
)
# ``Channel`` teardown and closure sequence
# drop ref to channel so it can be gc-ed and disconnected
log.runtime(
f'Disconnected IPC channel:\n'
f'uid: {chan.uid}\n'
f'|_{pformat(chan)}\n'
)
chans.remove(chan)
# TODO: do we need to be this pedantic?
if not chans:
log.runtime(
f'No more channels with {chan.uid}'
)
self._peers.pop(uid, None)
peers_str: str = ''
for uid, chans in self._peers.items():
peers_str += (
f'|_ uid: {uid}\n'
)
for i, chan in enumerate(chans):
peers_str += (
f' |_[{i}] {pformat(chan)}\n'
)
log.runtime(
f'Remaining IPC {len(self._peers)} peers:\n'
+ peers_str
)
# No more channels to other actors (at all) registered
# as connected.
if not self._peers:
log.runtime("Signalling no more peer channel connections")
self._no_more_peers.set()
# NOTE: block this actor from acquiring the
# debugger-TTY-lock since we have no way to know if we
# cancelled it and further there is no way to ensure the
# lock will be released if acquired due to having no
# more active IPC channels.
if _state.is_root_process():
pdb_lock = _debug.Lock
pdb_lock._blocked.add(uid)
# TODO: NEEEDS TO BE TESTED!
# actually, no idea if this ever even enters.. XD
pdb_user_uid: tuple = pdb_lock.global_actor_in_debug
if (
pdb_user_uid
and local_nursery
):
entry: tuple|None = local_nursery._children.get(pdb_user_uid)
if entry:
proc: trio.Process
_, proc, _ = entry
if (
(poll := getattr(proc, 'poll', None))
and poll() is None
):
log.cancel(
'Root actor reports no-more-peers, BUT '
'a DISCONNECTED child still has the debug '
'lock!\n'
f'root uid: {self.uid}\n'
f'last disconnected child uid: {uid}\n'
f'locking child uid: {pdb_user_uid}\n'
)
await maybe_wait_for_debugger(
child_in_debug=True
)
# TODO: just bc a child's transport dropped
# doesn't mean it's not still using the pdb
# REPL! so,
# -[ ] ideally we can check out child proc
# tree to ensure that its alive (and
# actually using the REPL) before we cancel
# it's lock acquire by doing the below!
# -[ ] create a way to read the tree of each actor's
# grandchildren such that when an
# intermediary parent is cancelled but their
# child has locked the tty, the grandparent
# will not allow the parent to cancel or
# zombie reap the child! see open issue:
# - https://github.com/goodboy/tractor/issues/320
# ------ - ------
# if a now stale local task has the TTY lock still
# we cancel it to allow servicing other requests for
# the lock.
db_cs: trio.CancelScope|None = pdb_lock._root_local_task_cs_in_debug
if (
db_cs
and not db_cs.cancel_called
and uid == pdb_user_uid
):
log.critical(
f'STALE DEBUG LOCK DETECTED FOR {uid}'
)
# TODO: figure out why this breaks tests..
db_cs.cancel()
# XXX: is this necessary (GC should do it)?
# XXX WARNING XXX
# Be AWARE OF THE INDENT LEVEL HERE
# -> ONLY ENTER THIS BLOCK WHEN ._peers IS
# EMPTY!!!!
if (
not self._peers
and chan.connected()
):
# if the channel is still connected it may mean the far
# end has not closed and we may have gotten here due to
# an error and so we should at least try to terminate
# the channel from this end gracefully.
log.runtime(
'Terminating channel with `None` setinel msg\n'
f'|_{chan}\n'
)
try:
# send msg loop terminate sentinel which
# triggers cancellation of all remotely
# started tasks.
await chan.send(None)
# XXX: do we want this? no right?
# causes "[104] connection reset by peer" on other end
# await chan.aclose()
except trio.BrokenResourceError:
log.runtime(f"Channel {chan.uid} was already closed")
# TODO: rename to `._deliver_payload()` since this handles
# more then just `result` msgs now obvi XD
async def _push_result(
self,
chan: Channel,
cid: str,
msg: dict[str, Any],
) -> None|bool:
'''
Push an RPC msg-payload to the local consumer peer-task's
queue.
'''
uid: tuple[str, str] = chan.uid
assert uid, f"`chan.uid` can't be {uid}"
try:
ctx: Context = self._contexts[(
uid,
cid,
# TODO: how to determine this tho?
# side,
)]
except KeyError:
log.warning(
'Ignoring invalid IPC ctx msg!\n\n'
f'<= sender: {uid}\n'
f'=> cid: {cid}\n\n'
f'{msg}\n'
)
return
return await ctx._deliver_msg(msg)
def get_context(
self,
chan: Channel,
cid: str,
nsf: NamespacePath,
# TODO: support lookup by `Context.side: str` ?
# -> would allow making a self-context which might have
# certain special use cases where RPC isolation is wanted
# between 2 tasks running in the same process?
# => prolly needs some deeper though on the real use cases
# and whether or not such things should be better
# implemented using a `TaskManager` style nursery..
#
# side: str|None = None,
msg_buffer_size: int|None = None,
allow_overruns: bool = False,
) -> Context:
'''
Look-up (existing) or create a new
inter-actor-SC-linked task "context" (a `Context`) which
encapsulates the local RPC task's execution enviroment
around `Channel` relayed msg handling including,
- a dedicated `trio` cancel scope (`Context._scope`),
- a pair of IPC-msg-relay "feeder" mem-channels
(`Context._recv/send_chan`),
- and a "context id" (cid) unique to the task-pair
msging session's lifetime.
'''
actor_uid = chan.uid
assert actor_uid
try:
ctx = self._contexts[(
actor_uid,
cid,
# side,
)]
log.runtime(
f'Retreived cached IPC ctx for\n'
f'peer: {chan.uid}\n'
f'cid:{cid}\n'
)
ctx._allow_overruns = allow_overruns
# adjust buffer size if specified
state = ctx._send_chan._state # type: ignore
if msg_buffer_size and state.max_buffer_size != msg_buffer_size:
state.max_buffer_size = msg_buffer_size
except KeyError:
log.runtime(
f'Creating NEW IPC ctx for\n'
f'peer: {chan.uid}\n'
f'cid: {cid}\n'
)
ctx = mk_context(
chan,
cid,
nsf=nsf,
msg_buffer_size=msg_buffer_size or self.msg_buffer_size,
_allow_overruns=allow_overruns,
)
self._contexts[(
actor_uid,
cid,
# side,
)] = ctx
return ctx
async def start_remote_task(
self,
chan: Channel,
nsf: NamespacePath,
kwargs: dict,
# IPC channel config
msg_buffer_size: int|None = None,
allow_overruns: bool = False,
load_nsf: bool = False,
) -> Context:
'''
Send a `'cmd'` msg to a remote actor, which requests the
start and schedule of a remote task-as-function's
entrypoint.
Synchronously validates the endpoint type and returns
a (caller side) `Context` that can be used to accept
delivery of msg payloads from the local runtime's
processing loop: `._rpc.process_messages()`.
'''
cid: str = str(uuid.uuid4())
assert chan.uid
ctx = self.get_context(
chan=chan,
cid=cid,
nsf=nsf,
# side='caller',
msg_buffer_size=msg_buffer_size,
allow_overruns=allow_overruns,
)
if (
'self' in nsf
or not load_nsf
):
ns, _, func = nsf.partition(':')
else:
# TODO: pass nsf directly over wire!
# -[ ] but, how to do `self:<Actor.meth>`??
ns, func = nsf.to_tuple()
log.runtime(
'Sending cmd to\n'
f'peer: {chan.uid} => \n'
'\n'
f'=> {ns}.{func}({kwargs})\n'
)
await chan.send(
{'cmd': (
ns,
func,
kwargs,
self.uid,
cid,
)}
)
# Wait on first response msg and validate; this should be
# immediate.
first_msg: dict = await ctx._recv_chan.receive()
functype: str = first_msg.get('functype')
if 'error' in first_msg:
raise unpack_error(first_msg, chan)
elif functype not in (
'asyncfunc',
'asyncgen',
'context',
):
raise ValueError(f"{first_msg} is an invalid response packet?")
ctx._remote_func_type = functype
return ctx
async def _from_parent(
self,
parent_addr: tuple[str, int]|None,
) -> tuple[
Channel,
list[tuple[str, int]]|None,
]:
'''
Bootstrap this local actor's runtime config from its parent by
connecting back via the IPC transport, handshaking and then
`Channel.recv()`-ing seeded data.
'''
try:
# Connect back to the parent actor and conduct initial
# handshake. From this point on if we error, we
# attempt to ship the exception back to the parent.
chan = Channel(
destaddr=parent_addr,
)
await chan.connect()
# TODO: move this into a `Channel.handshake()`?
# Initial handshake: swap names.
await self._do_handshake(chan)
accept_addrs: list[tuple[str, int]]|None = None
if self._spawn_method == "trio":
# Receive runtime state from our parent
parent_data: dict[str, Any]
parent_data = await chan.recv()
log.runtime(
'Received state from parent:\n\n'
# TODO: eventually all these msgs as
# `msgspec.Struct` with a special mode that
# pformats them in multi-line mode, BUT only
# if "trace"/"util" mode is enabled?
f'{pformat(parent_data)}\n'
)
accept_addrs: list[tuple[str, int]] = parent_data.pop('bind_addrs')
rvs = parent_data.pop('_runtime_vars')
if rvs['_debug_mode']:
try:
log.info('Enabling `stackscope` traces on SIGUSR1')
from .devx import enable_stack_on_sig
enable_stack_on_sig()
except ImportError:
log.warning(
'`stackscope` not installed for use in debug mode!'
)
log.runtime(f"Runtime vars are: {rvs}")
rvs['_is_root'] = False
_state._runtime_vars.update(rvs)
for attr, value in parent_data.items():
if (
attr == 'reg_addrs'
and value
):
# XXX: ``msgspec`` doesn't support serializing tuples
# so just cash manually here since it's what our
# internals expect.
# TODO: we don't really NEED these as
# tuples so we can probably drop this
# casting since apparently in python lists
# are "more efficient"?
self.reg_addrs = [tuple(val) for val in value]
else:
setattr(self, attr, value)
return chan, accept_addrs
except OSError: # failed to connect
log.warning(
f'Failed to connect to parent!?\n\n'
'Closing IPC [TCP] transport server to\n'
f'{parent_addr}\n'
f'|_{self}\n\n'
)
await self.cancel(chan=None) # self cancel
raise
async def _serve_forever(
self,
handler_nursery: Nursery,
*,
# (host, port) to bind for channel server
listen_sockaddrs: list[tuple[str, int]]|None = None,
task_status: TaskStatus[Nursery] = trio.TASK_STATUS_IGNORED,
) -> None:
'''
Start the IPC transport server, begin listening for new connections.
This will cause an actor to continue living (and thus
blocking at the process/OS-thread level) until
`.cancel_server()` is called.
'''
if listen_sockaddrs is None:
listen_sockaddrs = [(None, 0)]
self._server_down = trio.Event()
try:
async with trio.open_nursery() as server_n:
for host, port in listen_sockaddrs:
listeners: list[trio.abc.Listener] = await server_n.start(
partial(
trio.serve_tcp,
handler=self._stream_handler,
port=port,
host=host,
# NOTE: configured such that new
# connections will stay alive even if
# this server is cancelled!
handler_nursery=handler_nursery,
)
)
sockets: list[trio.socket] = [
getattr(listener, 'socket', 'unknown socket')
for listener in listeners
]
log.runtime(
'Started TCP server(s)\n'
f'|_{sockets}\n'
)
self._listeners.extend(listeners)
task_status.started(server_n)
finally:
# signal the server is down since nursery above terminated
self._server_down.set()
def cancel_soon(self) -> None:
'''
Cancel this actor asap; can be called from a sync context.
Schedules runtime cancellation via `Actor.cancel()` inside
the RPC service nursery.
'''
assert self._service_n
self._service_n.start_soon(
self.cancel,
None, # self cancel all rpc tasks
)
async def cancel(
self,
# chan whose lifetime limits the lifetime of its remotely
# requested and locally spawned RPC tasks - similar to the
# supervision semantics of a nursery wherein the actual
# implementation does start all such tasks in
# a sub-nursery.
req_chan: Channel|None,
) -> bool:
'''
Cancel this actor's runtime, eventually resulting in
termination of its containing OS process.
The ideal "deterministic" teardown sequence in order is:
- cancel all ongoing rpc tasks by cancel scope.
- cancel the channel server to prevent new inbound
connections.
- cancel the "service" nursery reponsible for
spawning new rpc tasks.
- return control the parent channel message loop.
'''
(
requesting_uid,
requester_type,
req_chan,
log_meth,
) = (
req_chan.uid,
'peer',
req_chan,
log.cancel,
) if req_chan else (
# a self cancel of ALL rpc tasks
self.uid,
'self',
self,
log.runtime,
)
# TODO: just use the new `Context.repr_rpc: str` (and
# other) repr fields instead of doing this all manual..
msg: str = (
f'Runtime cancel request from {requester_type}:\n\n'
f'<= .cancel(): {requesting_uid}\n'
)
# TODO: what happens here when we self-cancel tho?
self._cancel_called_by_remote: tuple = requesting_uid
self._cancel_called = True
# cancel all ongoing rpc tasks
with CancelScope(shield=True):
# kill any debugger request task to avoid deadlock
# with the root actor in this tree
dbcs = _debug.Lock._debugger_request_cs
if dbcs is not None:
msg += (
'>> Cancelling active debugger request..\n'
f'|_{_debug.Lock}\n'
)
dbcs.cancel()
# self-cancel **all** ongoing RPC tasks
await self.cancel_rpc_tasks(
req_uid=requesting_uid,
parent_chan=None,
)
# stop channel server
self.cancel_server()
if self._server_down is not None:
await self._server_down.wait()
else:
log.warning(
'Transport[TCP] server was cancelled start?'
)
# cancel all rpc tasks permanently
if self._service_n:
self._service_n.cancel_scope.cancel()
log_meth(msg)
self._cancel_complete.set()
return True
# XXX: hard kill logic if needed?
# def _hard_mofo_kill(self):
# # If we're the root actor or zombied kill everything
# if self._parent_chan is None: # TODO: more robust check
# root = trio.lowlevel.current_root_task()
# for n in root.child_nurseries:
# n.cancel_scope.cancel()
async def _cancel_task(
self,
cid: str,
parent_chan: Channel,
requesting_uid: tuple[str, str]|None,
ipc_msg: dict|None|bool = False,
) -> bool:
'''
Cancel a local (RPC) task by context-id/channel by calling
`trio.CancelScope.cancel()` on it's surrounding cancel
scope.
'''
# this ctx based lookup ensures the requested task to be
# cancelled was indeed spawned by a request from its
# parent (or some grandparent's) channel
ctx: Context
func: Callable
is_complete: trio.Event
try:
(
ctx,
func,
is_complete,
) = self._rpc_tasks[(
parent_chan,
cid,
)]
scope: CancelScope = ctx._scope
except KeyError:
# NOTE: during msging race conditions this will often
# emit, some examples:
# - callee returns a result before cancel-msg/ctxc-raised
# - callee self raises ctxc before caller send request,
# - callee errors prior to cancel req.
log.cancel(
'Cancel request invalid, RPC task already completed?\n\n'
f'<= canceller: {requesting_uid}\n\n'
f'=> {cid}@{parent_chan.uid}\n'
f' |_{parent_chan}\n'
)
return True
log.cancel(
'Cancel request for RPC task\n\n'
f'<= Actor._cancel_task(): {requesting_uid}\n\n'
f'=> {ctx._task}\n'
f' |_ >> {ctx.repr_rpc}\n'
# f' >> Actor._cancel_task() => {ctx._task}\n'
# f' |_ {ctx._task}\n\n'
# TODO: better ascii repr for "supervisor" like
# a nursery or context scope?
# f'=> {parent_chan}\n'
# f' |_{ctx._task}\n'
# TODO: simplified `Context.__repr__()` fields output
# shows only application state-related stuff like,
# - ._stream
# - .closed
# - .started_called
# - .. etc.
# f' >> {ctx.repr_rpc}\n'
# f' |_ctx: {cid}\n'
# f' >> {ctx._nsf}()\n'
)
if (
ctx._canceller is None
and requesting_uid
):
ctx._canceller: tuple = requesting_uid
# TODO: pack the RPC `{'cmd': <blah>}` msg into a ctxc and
# then raise and pack it here?
if (
ipc_msg
and ctx._cancel_msg is None
):
# assign RPC msg directly from the loop which usually
# the case with `ctx.cancel()` on the other side.
ctx._cancel_msg = ipc_msg
# don't allow cancelling this function mid-execution
# (is this necessary?)
if func is self._cancel_task:
log.error('Do not cancel a cancel!?')
return True
# TODO: shouldn't we eventually be calling ``Context.cancel()``
# directly here instead (since that method can handle both
# side's calls into it?
# await ctx.cancel()
scope.cancel()
# wait for _invoke to mark the task complete
flow_info: str = (
f'<= canceller: {requesting_uid}\n'
f'=> ipc-parent: {parent_chan}\n'
f' |_{ctx}\n'
)
log.runtime(
'Waiting on RPC task to cancel\n'
f'{flow_info}'
)
await is_complete.wait()
log.runtime(
f'Sucessfully cancelled RPC task\n'
f'{flow_info}'
)
return True
async def cancel_rpc_tasks(
self,
req_uid: tuple[str, str],
# NOTE: when None is passed we cancel **all** rpc
# tasks running in this actor!
parent_chan: Channel|None,
) -> None:
'''
Cancel all ongoing RPC tasks owned/spawned for a given
`parent_chan: Channel` or simply all tasks (inside
`._service_n`) when `parent_chan=None`.
'''
tasks: dict = self._rpc_tasks
if not tasks:
log.runtime(
'Actor has no cancellable RPC tasks?\n'
f'<= canceller: {req_uid}\n'
)
return
# TODO: seriously factor this into some helper funcs XD
tasks_str: str = ''
for (ctx, func, _) in tasks.values():
# TODO: std repr of all primitives in
# a hierarchical tree format, since we can!!
# like => repr for funcs/addrs/msg-typing:
#
# -[ ] use a proper utf8 "arm" like
# `stackscope` has!
# -[ ] for typed msging, show the
# py-type-annot style?
# - maybe auto-gen via `inspect` / `typing` type-sig:
# https://stackoverflow.com/a/57110117
# => see ex. code pasted into `.msg.types`
#
# -[ ] proper .maddr() for IPC primitives?
# - `Channel.maddr() -> str:` obvi!
# - `Context.maddr() -> str:`
tasks_str += (
f' |_@ /ipv4/tcp/cid="{ctx.cid[-16:]} .."\n'
f' |>> {ctx._nsf}() -> dict:\n'
)
descr: str = (
'all' if not parent_chan
else
"IPC channel's "
)
rent_chan_repr: str = (
f'|_{parent_chan}'
if parent_chan
else ''
)
log.cancel(
f'Cancelling {descr} {len(tasks)} rpc tasks\n\n'
f'<= `Actor.cancel_rpc_tasks()`: {req_uid}\n'
f' {rent_chan_repr}\n'
# f'{self}\n'
# f'{tasks_str}'
)
for (
(task_caller_chan, cid),
(ctx, func, is_complete),
) in tasks.copy().items():
if (
# maybe filter to specific IPC channel?
(parent_chan
and
task_caller_chan != parent_chan)
# never "cancel-a-cancel" XD
or (func == self._cancel_task)
):
continue
# TODO: this maybe block on the task cancellation
# and so should really done in a nursery batch?
await self._cancel_task(
cid,
task_caller_chan,
requesting_uid=req_uid,
)
if tasks:
log.cancel(
'Waiting for remaining rpc tasks to complete\n'
f'|_{tasks}'
)
await self._ongoing_rpc_tasks.wait()
def cancel_server(self) -> None:
'''
Cancel the internal IPC transport server nursery thereby
preventing any new inbound IPC connections establishing.
'''
if self._server_n:
log.runtime("Shutting down channel server")
self._server_n.cancel_scope.cancel()
@property
def accept_addrs(self) -> list[tuple[str, int]]:
'''
All addresses to which the IPC-transport-channel server
binds and listens for new connections.
'''
# throws OSError on failure
return [
listener.socket.getsockname()
for listener in self._listeners
] # type: ignore
@property
def accept_addr(self) -> tuple[str, int]:
'''
Primary address to which the IPC transport server is
bound.
'''
# throws OSError on failure
return self.accept_addrs[0]
def get_parent(self) -> Portal:
'''
Return a `Portal` to our parent.
'''
assert self._parent_chan, "No parent channel for this actor?"
return Portal(self._parent_chan)
def get_chans(self, uid: tuple[str, str]) -> list[Channel]:
'''
Return all IPC channels to the actor with provided `uid`.
'''
return self._peers[uid]
# TODO: move to `Channel.handshake(uid)`
async def _do_handshake(
self,
chan: Channel
) -> tuple[str, str]:
'''
Exchange `(name, UUIDs)` identifiers as the first
communication step with any (peer) remote `Actor`.
These are essentially the "mailbox addresses" found in
"actor model" parlance.
'''
await chan.send(self.uid)
value: tuple = await chan.recv()
uid: tuple[str, str] = (str(value[0]), str(value[1]))
if not isinstance(uid, tuple):
raise ValueError(f"{uid} is not a valid uid?!")
chan.uid = str(uid[0]), str(uid[1])
return uid
def is_infected_aio(self) -> bool:
'''
If `True`, this actor is running `trio` in guest mode on
the `asyncio` event loop and thus can use the APIs in
`.to_asyncio` to coordinate tasks running in each
framework but within the same actor runtime.
'''
return self._infected_aio
async def async_main(
actor: Actor,
accept_addrs: tuple[str, int]|None = None,
# XXX: currently ``parent_addr`` is only needed for the
# ``multiprocessing`` backend (which pickles state sent to
# the child instead of relaying it over the connect-back
# channel). Once that backend is removed we can likely just
# change this to a simple ``is_subactor: bool`` which will
# be False when running as root actor and True when as
# a subactor.
parent_addr: tuple[str, int]|None = None,
task_status: TaskStatus[None] = trio.TASK_STATUS_IGNORED,
) -> None:
'''
Main `Actor` runtime entrypoint; start the transport-specific
IPC channel server, (maybe) connect back to parent (to receive
additional config), startup all core `trio` machinery for
delivering RPCs, register with the discovery system.
The "root" (or "top-level") and "service" `trio.Nursery`s are
opened here and when cancelled/terminated effectively shutdown
the actor's "runtime" and all thus all ongoing RPC tasks.
'''
# attempt to retreive ``trio``'s sigint handler and stash it
# on our debugger lock state.
_debug.Lock._trio_handler = signal.getsignal(signal.SIGINT)
is_registered: bool = False
try:
# establish primary connection with immediate parent
actor._parent_chan: Channel|None = None
if parent_addr is not None:
(
actor._parent_chan,
set_accept_addr_says_rent,
) = await actor._from_parent(parent_addr)
# either it's passed in because we're not a child or
# because we're running in mp mode
if (
set_accept_addr_says_rent
and set_accept_addr_says_rent is not None
):
accept_addrs = set_accept_addr_says_rent
# The "root" nursery ensures the channel with the immediate
# parent is kept alive as a resilient service until
# cancellation steps have (mostly) occurred in
# a deterministic way.
async with trio.open_nursery() as root_nursery:
actor._root_n = root_nursery
assert actor._root_n
async with trio.open_nursery() as service_nursery:
# This nursery is used to handle all inbound
# connections to us such that if the TCP server
# is killed, connections can continue to process
# in the background until this nursery is cancelled.
actor._service_n = service_nursery
assert actor._service_n
# load exposed/allowed RPC modules
# XXX: do this **after** establishing a channel to the parent
# but **before** starting the message loop for that channel
# such that import errors are properly propagated upwards
actor.load_modules()
# XXX TODO XXX: figuring out debugging of this
# would somemwhat guarantee "self-hosted" runtime
# debugging (since it hits all the ede cases?)
#
# `tractor.pause()` right?
# try:
# actor.load_modules()
# except ModuleNotFoundError as err:
# _debug.pause_from_sync()
# import pdbp; pdbp.set_trace()
# raise
# Startup up the transport(-channel) server with,
# - subactor: the bind address is sent by our parent
# over our established channel
# - root actor: the ``accept_addr`` passed to this method
assert accept_addrs
try:
actor._server_n = await service_nursery.start(
partial(
actor._serve_forever,
service_nursery,
listen_sockaddrs=accept_addrs,
)
)
except OSError as oserr:
# NOTE: always allow runtime hackers to debug
# tranport address bind errors - normally it's
# something silly like the wrong socket-address
# passed via a config or CLI Bo
entered_debug = await _debug._maybe_enter_pm(oserr)
if entered_debug:
log.runtime('Exited debug REPL..')
raise
accept_addrs: list[tuple[str, int]] = actor.accept_addrs
# NOTE: only set the loopback addr for the
# process-tree-global "root" mailbox since
# all sub-actors should be able to speak to
# their root actor over that channel.
if _state._runtime_vars['_is_root']:
for addr in accept_addrs:
host, _ = addr
# TODO: generic 'lo' detector predicate
if '127.0.0.1' in host:
_state._runtime_vars['_root_mailbox'] = addr
# Register with the arbiter if we're told its addr
log.runtime(
f'Registering `{actor.name}` ->\n'
f'{pformat(accept_addrs)}'
)
# TODO: ideally we don't fan out to all registrars
# if addresses point to the same actor..
# So we need a way to detect that? maybe iterate
# only on unique actor uids?
for addr in actor.reg_addrs:
try:
assert isinstance(addr, tuple)
assert addr[1] # non-zero after bind
except AssertionError:
await _debug.pause()
async with get_registry(*addr) as reg_portal:
for accept_addr in accept_addrs:
if not accept_addr[1]:
await _debug.pause()
assert accept_addr[1]
await reg_portal.run_from_ns(
'self',
'register_actor',
uid=actor.uid,
sockaddr=accept_addr,
)
is_registered: bool = True
# init steps complete
task_status.started()
# Begin handling our new connection back to our
# parent. This is done last since we don't want to
# start processing parent requests until our channel
# server is 100% up and running.
if actor._parent_chan:
await root_nursery.start(
partial(
process_messages,
actor,
actor._parent_chan,
shield=True,
)
)
log.runtime(
'Actor runtime is up!'
# 'Blocking on service nursery to exit..\n'
)
log.runtime(
"Service nursery complete\n"
"Waiting on root nursery to complete"
)
# Blocks here as expected until the root nursery is
# killed (i.e. this actor is cancelled or signalled by the parent)
except Exception as err:
log.runtime("Closing all actor lifetime contexts")
actor.lifetime_stack.close()
if not is_registered:
# TODO: I guess we could try to connect back
# to the parent through a channel and engage a debugger
# once we have that all working with std streams locking?
log.exception(
f"Actor errored and failed to register with arbiter "
f"@ {actor.reg_addrs[0]}?")
log.error(
"\n\n\t^^^ THIS IS PROBABLY AN INTERNAL `tractor` BUG! ^^^\n\n"
"\t>> CALMLY CALL THE AUTHORITIES AND HIDE YOUR CHILDREN <<\n\n"
"\tIf this is a sub-actor hopefully its parent will keep running "
"correctly presuming this error was safely ignored..\n\n"
"\tPLEASE REPORT THIS TRACEBACK IN A BUG REPORT: "
"https://github.com/goodboy/tractor/issues\n"
)
if actor._parent_chan:
await try_ship_error_to_remote(
actor._parent_chan,
err,
)
# always!
match err:
case ContextCancelled():
log.cancel(
f'Actor: {actor.uid} was task-context-cancelled with,\n'
f'str(err)'
)
case _:
log.exception("Actor errored:")
raise
finally:
log.runtime(
'Runtime nursery complete'
'-> Closing all actor lifetime contexts..'
)
# tear down all lifetime contexts if not in guest mode
# XXX: should this just be in the entrypoint?
actor.lifetime_stack.close()
# TODO: we can't actually do this bc the debugger
# uses the _service_n to spawn the lock task, BUT,
# in theory if we had the root nursery surround this finally
# block it might be actually possible to debug THIS
# machinery in the same way as user task code?
# if actor.name == 'brokerd.ib':
# with CancelScope(shield=True):
# await _debug.breakpoint()
# Unregister actor from the registry-sys / registrar.
if (
is_registered
and not actor.is_registrar
):
failed: bool = False
for addr in actor.reg_addrs:
assert isinstance(addr, tuple)
with trio.move_on_after(0.5) as cs:
cs.shield = True
try:
async with get_registry(
*addr,
) as reg_portal:
await reg_portal.run_from_ns(
'self',
'unregister_actor',
uid=actor.uid
)
except OSError:
failed = True
if cs.cancelled_caught:
failed = True
if failed:
log.warning(
f'Failed to unregister {actor.name} from '
f'registar @ {addr}'
)
# Ensure all peers (actors connected to us as clients) are finished
if not actor._no_more_peers.is_set():
if any(
chan.connected() for chan in chain(*actor._peers.values())
):
log.runtime(
f"Waiting for remaining peers {actor._peers} to clear")
with CancelScope(shield=True):
await actor._no_more_peers.wait()
log.runtime("All peer channels are complete")
log.runtime("Runtime completed")
# TODO: rename to `Registry` and move to `._discovery`!
class Arbiter(Actor):
'''
A special registrar actor who can contact all other actors
within its immediate process tree and possibly keeps a registry
of others meant to be discoverable in a distributed
application. Normally the registrar is also the "root actor"
and thus always has access to the top-most-level actor
(process) nursery.
By default, the registrar is always initialized when and if no
other registrar socket addrs have been specified to runtime
init entry-points (such as `open_root_actor()` or
`open_nursery()`). Any time a new main process is launched (and
thus thus a new root actor created) and, no existing registrar
can be contacted at the provided `registry_addr`, then a new
one is always created; however, if one can be reached it is
used.
Normally a distributed app requires at least registrar per
logical host where for that given "host space" (aka localhost
IPC domain of addresses) it is responsible for making all other
host (local address) bound actors *discoverable* to external
actor trees running on remote hosts.
'''
is_arbiter = True
def __init__(
self,
*args,
**kwargs,
) -> None:
self._registry: dict[
tuple[str, str],
tuple[str, int],
] = {}
self._waiters: dict[
str,
# either an event to sync to receiving an actor uid (which
# is filled in once the actor has sucessfully registered),
# or that uid after registry is complete.
list[trio.Event | tuple[str, str]]
] = {}
super().__init__(*args, **kwargs)
async def find_actor(
self,
name: str,
) -> tuple[str, int]|None:
for uid, sockaddr in self._registry.items():
if name in uid:
return sockaddr
return None
async def get_registry(
self
) -> dict[str, tuple[str, int]]:
'''
Return current name registry.
This method is async to allow for cross-actor invocation.
'''
# NOTE: requires ``strict_map_key=False`` to the msgpack
# unpacker since we have tuples as keys (not this makes the
# arbiter suscetible to hashdos):
# https://github.com/msgpack/msgpack-python#major-breaking-changes-in-msgpack-10
return {
'.'.join(key): val
for key, val in self._registry.items()
}
async def wait_for_actor(
self,
name: str,
) -> list[tuple[str, int]]:
'''
Wait for a particular actor to register.
This is a blocking call if no actor by the provided name is currently
registered.
'''
sockaddrs: list[tuple[str, int]] = []
sockaddr: tuple[str, int]
for (aname, _), sockaddr in self._registry.items():
log.runtime(
f'Actor mailbox info:\n'
f'aname: {aname}\n'
f'sockaddr: {sockaddr}\n'
)
if name == aname:
sockaddrs.append(sockaddr)
if not sockaddrs:
waiter = trio.Event()
self._waiters.setdefault(name, []).append(waiter)
await waiter.wait()
for uid in self._waiters[name]:
if not isinstance(uid, trio.Event):
sockaddrs.append(self._registry[uid])
return sockaddrs
async def register_actor(
self,
uid: tuple[str, str],
sockaddr: tuple[str, int]
) -> None:
uid = name, hash = (str(uid[0]), str(uid[1]))
addr = (host, port) = (
str(sockaddr[0]),
int(sockaddr[1]),
)
if port == 0:
await _debug.pause()
assert port # should never be 0-dynamic-os-alloc
self._registry[uid] = addr
# pop and signal all waiter events
events = self._waiters.pop(name, [])
self._waiters.setdefault(name, []).append(uid)
for event in events:
if isinstance(event, trio.Event):
event.set()
async def unregister_actor(
self,
uid: tuple[str, str]
) -> None:
uid = (str(uid[0]), str(uid[1]))
entry: tuple = self._registry.pop(uid, None)
if entry is None:
log.warning(f'Request to de-register {uid} failed?')