Also adjust sizing such that the history buffer will backfill the last
six years by default (in 1m OHLC) and the hft buffer will do only 3 days
worth. Also ensure the fsp layer passes the src shm's buffer size when
allocating since the size is now required by allocators in the shm apis.
Avoid unnecessarily re-rendering the wrong (1min OHLC history) chart
and/or other such charts with update tasks listening to the sampler
stream. Instead only redraw in tasks which are updating vizs which match
the actual details of the backfill event.
We can probably also eventually match against a range tuple (emitted in
the msg) and then have the task further only update the formatter layer
unless the range is actually in view?
It's no longer part of the default OHLCV array-buffer schema and just
generally we should be processing and managing **any** non source data
in the FSP subsystem(s) despite it maybe being provided as a default by
some backends.
Explains why stuff always seemed wrong before XD
Previously whenever a time-gappy asset (like a stock due to it's venue
operating hours) was being loaded, we weren't querying for a "durations
worth" of bars and this was causing all sorts of actual gaps in our
data set that shouldn't exist..
Fix that by always attempting to retrieve a min aggregate-time's
worth/duration of bars/datums in the history manager. Actually,
i implemented this in both the feed and api layers for this backend
since it doesn't seem to strictly work just implementing it at the
`Client.bars()` level, not sure why but..
Also, buncha `ruff` linting cleanups and fix the logger nameeee, lel.
For now, just detect and fill in gaps (via fresh backend queries)
*in the shm buffer* but eventually i'm pretty sure we can just write
these direct to the parquet file as well.
Use the new `.data._timeseries.detect_null_time_gap()` to find and fill
in the `ShmArray` index range, re-check it and enter a prompt if it
didn't totally fill.
Also,
- do a massive cleanup and removal of all unused/commented code.
- drop the duplicate frames tracking, don't think we need it after
removing multi-frame concurrent queries.
- change backfill loop variable `end_dt` -> `last_start_dt` which is
more semantically correct.
- fix logic to backfill any missing sub-sequence portion for any frame
query that overruns the shm buffer prependable space by detecting
the available rows left to insert and only push those.
- add a new `shm_push_in_between()` helper to match.
It took a little while (and a lot of commenting out of old no longer
needed code) but, this gets tsdb (from parquet file) loading *before*
final backfilling from the most recent history frame until the most
recent tsdb time stamp!
More or less all the convoluted concurrency shit we had for coping with
`marketstore` IPC junk is no longer needed, particularly all the query
size limits and accompanying load loops.. The recent frame loading
technique/order *has* now changed though since we'd like to show charts
asap once tsdb history loads.
The new load sequence is as follows:
- load mr (most recent) frame from backend.
- load existing history (one shot) from the "tsdb" aka parquet files
with `polars`.
- backfill the gap part from the mr frame back to the tsdb start
incrementally by making (hacky) `ShmArray.push(start=<blah>)` calls
and *not* updating the `._first.value` while doing it XD
Dirtier deatz:
- make `tsdb_backfill()` run per timeframe in a separate task.
- drop all the loop through timeframes and insert `dts_per_tf` crap.
- only spawn a subtask for the `start_backfill()` call which in turn
only does the gap backfilling as mentioned above.
- mask out all the code related to being limited to certain query sizes
(over gRPC) as was restricted by marketstore.. not gonna go through
what all of that was since it's probably getting deleted in a follow
up commit.
- buncha off-by-one tweaks to do with backfilling the gap from mr frame
to tsdb start.. mostly tinkered it to get it all right but seems to be
working correctly B)
- still use the `broadcast_all()` msg stuff when doing the gap backfill
though don't have it really working yet on the UI side (since
previously we were relying on the shm first/last values.. so this will
be "coming soon" :)
For OHLCV time series we normally presume a uniform sampling period
(1s or 60s by default) and it's handy to have tools to ensure a series
is gapless or contains expected gaps based on (legacy) market hours.
For this we leverage `polars`:
- add `.nativedb.with_dts()` a datetime-from-epoch-time-column frame
"column-expander" which inserts datetime-casted, epoch-diff and
dt-diff columns.
- add `.nativedb.detect_time_gaps()` which filters to any larger then
expected sampling period rows.
- wrap the above (for now) in a `piker store anal` (analysis) cmd which
atm always enters a breakpoint for tinkering.
Supporting storage client additions:
- add a `detect_period()` helper for extracting expected OHLC time step.
- add new `NativedbStorageClient` methods and attrs to provide for the above:
- `.mk_path()` to **only** deliver a parquet-file path for use in
other methods.
- `._dfs` to house cached `pl.DataFrame`s loaded from `.parquet` files.
- `.as_df()` which loads cached frames or loads them from disk and
then caches (for next use).
- `_write_ohlcv()` a private-sync version of the public equivalent
meth since we don't currently have any actual async file IO
underneath; add a flag for whether to return as a `numpy.ndarray`.
- drop buncha cruft from `store ls` cmd and make it work for
multi-backend fqme listing.
- including adding an `.address` to the mkts client which shows the
grpc socketaddr details.
- change defauls to new `'nativedb'.
- drop 'marketstore' from built-in backend list (for now)
It was a concurrency-hack mess somewhat due to all sorts of limitations
imposed by marketstore (query size limits, strange datetime/timestamp
errors, slow table loads for large queries..) and we can drastically
simplify. There's still some issues with getting new backfills (not yet
in storage) correctly prepended: there's sometimes little gaps due to shm
races when reading history indexing vs. when the live-feed startup
finishes.
We generally need tests for all this and likely a better rework of the
feed layer's init such that we're showing history in chart afap instead
of waiting on backfills or the live feed to come up.
Much more to come B)
Turns out no backend (including kraken) requires it and really this
kinda of measure should be implemented and recorded from our fsp layer
instead of (hackily) sometimes expecting it to be in "source data".
After much frustration with a particular tsdb (cough) this instead
implements a new native-file (and apache tech based) backend which
stores time series in parquet files (for now) using the `polars` apis
(since we plan to use that lib as well for processing).
Note this code is currently **very** rough and in draft mode.
Details:
- add conversion routines for going from `polars.DataFrame` to
`numpy.ndarray` and back.
- lay out a simple file-name as series key symbology:
`fqme.<datadescriptions>.parquet`, though probably it will evolve.
- implement the entire `StorageClient` interface as it stands.
- adjust `storage.cli` cmds to instead expect to use this new backend,
which means it's a complete mess XD
Main benefits/motivation:
- wayy faster load times with no "datums to load limit" required.
- smaller space footprint and we haven't even touched compression
settings yet!
- wayyy more compatible with other systems which can lever the apache
ecosystem.
- gives us finer grained control over the filesystem usage so we can
choose to swap out stuff like the replication system or networking
access.
Turns out just (over)writing `.parquet` files with >= 1M datums is like
less then a second, and we can likely speed up appends using
`fastparquet` (usage coming soon).
Includes:
- a new `clone` CLI subcmd to test this all out by ad-hoc copy of
(literally hardcoded to a daemon-actor specific shm allocation X) an
existing `/dev/shm/<ShmArray>` and push to `.parquet` file.
- code to convert from our `ShmArray.array: np.ndarray` ->
`polars.DataFrame` (thanks SO).
- timing checks around the file IO and np -> polars conversion.
- a `read` subcmd which i was using to test the sync `pymarketstore`
client against our async one to see if the issues from
https://github.com/pikers/piker/issues/443 were resolved, but nope!
Turns out trying to switch to the old sync client and going back to
using the old json-RPC API (after having had to patch the upstream repo
to not import gRPC machinery to avoid crashes..) I'm basically getting
the exact same issues.
New tinkering results does possibly tell some new stuff:
- the EOF error seems to indeed be due to trying fetch records which haven't been
written (properly) - like asking for a `end=<epoch_int>` that is
earlier then the earliest record.
- the "snappy input corrupt" error seems to have something to do with
the `Params.end` field not being an `int` and/or the int precision not
being chosen correctly?
- toying with this a bunch manually shows that the internals of the
client (particularly `.build_query()` stuff) is parsing/calcing the
`Epoch` and `Nanoseconds` values out incorrectly.. which is likely
part of the problem.
- we also changed `anyio_marketstore.MarketStoreclient.build_query()`
logic when removing `pandas` a while back, which also seems to be
part of the problem on the async side, however reverting those
changes also didn't fix the issue entirely; likely something else
more subtle going on (maybe with the write vs. read `Epoch` field
type we pass?).
Despite all this malarky, we're already underway more or less obsoleting
this whole thing with a much less complex approach of using apache
parquet files and modern filesystem tools to get a more flexible and
numerics-native dataframe-oriented tsdb B)
Turns out the reason we were originally making the `time: float` column in our
ohlcv arrays was bc that's what **only** ib uses XD (and/or 🤦)
Instead we changed the default field type to be an `int` (which is also
more correct to avoid `float` rounding/precision discrepancies) and thus
**do not need to override it** in all other (crypto) backends (except
`ib`). Now we only do the customization (via `._ohlc_dtype`) to `float`
only for `ib` for now (though pretty sure we can also not do that
eventually as well..)!
Use `def_iohlcv_fields` for a name and instead of copying and inserting
the index field pop it for the non-index version. Drop creating
`np.dtype()` instances since `numpy`'s apis accept both input forms so
this is simpler on our end.
Turns out you can mix and match `click` with `typer` so this moves what
was the `.data.cli` stuff into `storage.cli` and uses the integration
api to make it all work B)
New subcmd: `piker store`
- add `piker store ls` which lists all fqme keyed time-series from backend.
- add `store delete` to remove any such key->time-series.
- now uses a nursery for multi-timeframe concurrency B)
Mask out all the old `marketstore` specific subcmds for now (streaming,
ingest, storesh, etc..) in anticipation of moving them into
a subpkg-module and make sure to import the sub-cmd module in our top
level cli package.
Other `.storage` api tweaks:
- drop the reraising with custom error (for now).
- rename `Storage` -> `StorageClient` (or should it be API?).
To kick off our (tsdb) storage backends this adds our first implementing
a new `Storage(Protocol)` client interface. Going foward, the top level
`.storage` pkg-module will now expose backend agnostic APIs and helpers
whilst specific backend implementations will adhere to that middle-ware
layer.
Deats:
- add `.storage.marketstore.Storage` as the first client implementation,
moving all needed (import) dependencies out from
`.service.marketstore` as well as `.ohlc_key_map` and `get_client()`.
- move root `conf.toml` loading from `.data.history` into
`.storage.__init__.open_storage_client()` which now takes in a `name:
str` and does all the work of loading the correct backend module, its
config, and determining if a service-instance can be contacted and
a client loaded; in the case where this fails we raise a new
`StorageConnectionError`.
- add a new `.storage.get_storagemod()` just like we have for brokers.
- make `open_storage_client()` also return the backend module such that
the history-data layer can make backend specific calls as needed (eg.
ohlc_key_map).
- fall back to a basic non-tsdb backfill when `open_storage_client()`
raises the new connection error.
The plan is to offer multiple tsdb and other storage backends (for
a variety of use cases) and expose them similarly to how we do for
broker and data providers B)
Tyler Goodlet