tractor/tests/test_resource_cache.py

632 lines
19 KiB
Python

'''
Suites for our `.trionics.maybe_open_context()` multi-task
shared-cached `@acm` API.
'''
from contextlib import asynccontextmanager as acm
import platform
from typing import Awaitable
import pytest
import trio
import tractor
from tractor.trionics import (
maybe_open_context,
collapse_eg,
)
from tractor.log import (
get_console_log,
get_logger,
)
log = get_logger()
_resource: int = 0
@acm
async def maybe_increment_counter(task_name: str):
global _resource
_resource += 1
await trio.lowlevel.checkpoint()
yield _resource
await trio.lowlevel.checkpoint()
_resource -= 1
@pytest.mark.parametrize(
'key_on',
['key_value', 'kwargs'],
ids="key_on={}".format,
)
def test_resource_only_entered_once(key_on):
global _resource
_resource = 0
key = None
if key_on == 'key_value':
key = 'some_common_key'
async def main():
cache_active: bool = False
async def enter_cached_mngr(name: str):
nonlocal cache_active
if key_on == 'kwargs':
# make a common kwargs input to key on it
kwargs = {'task_name': 'same_task_name'}
assert key is None
else:
# different task names per task will be used
kwargs = {'task_name': name}
async with maybe_open_context(
maybe_increment_counter,
kwargs=kwargs,
key=key,
) as (cache_hit, resource):
if cache_hit:
try:
cache_active = True
assert resource == 1
await trio.sleep_forever()
finally:
cache_active = False
else:
assert resource == 1
await trio.sleep_forever()
with trio.move_on_after(0.5):
async with (
tractor.open_root_actor(),
trio.open_nursery() as tn,
):
for i in range(10):
tn.start_soon(
enter_cached_mngr,
f'task_{i}',
)
await trio.sleep(0.001)
trio.run(main)
@tractor.context
async def streamer(
ctx: tractor.Context,
seq: list[int] = list(range(1000)),
) -> None:
await ctx.started()
async with ctx.open_stream() as stream:
for val in seq:
await stream.send(val)
await trio.sleep(0.001)
print('producer finished')
@acm
async def open_stream() -> Awaitable[
tuple[
tractor.ActorNursery,
tractor.MsgStream,
]
]:
try:
async with tractor.open_nursery() as an:
portal = await an.start_actor(
'streamer',
enable_modules=[__name__],
)
try:
async with (
portal.open_context(streamer) as (ctx, first),
ctx.open_stream() as stream,
):
print('Entered open_stream() caller')
yield an, stream
print('Exited open_stream() caller')
finally:
print(
'Cancelling streamer with,\n'
'=> `Portal.cancel_actor()`'
)
await portal.cancel_actor()
print('Cancelled streamer')
except Exception as err:
print(
f'`open_stream()` errored?\n'
f'{err!r}\n'
)
await tractor.pause(shield=True)
raise err
@acm
async def maybe_open_stream(taskname: str):
async with maybe_open_context(
# NOTE: all secondary tasks should cache hit on the same key
acm_func=open_stream,
) as (
cache_hit,
(an, stream)
):
# when the actor + portal + ctx + stream has already been
# allocated we want to just bcast to this task.
if cache_hit:
print(f'{taskname} loaded from cache')
# add a new broadcast subscription for the quote stream
# if this feed is already allocated by the first
# task that entereed
async with stream.subscribe() as bstream:
yield an, bstream
print(
f'cached task exited\n'
f')>\n'
f' |_{taskname}\n'
)
# we should always unreg the "cloned" bcrc for this
# consumer-task
assert id(bstream) not in bstream._state.subs
else:
# yield the actual stream
try:
yield an, stream
finally:
print(
f'NON-cached task exited\n'
f')>\n'
f' |_{taskname}\n'
)
first_bstream = stream._broadcaster
bcrx_state = first_bstream._state
subs: dict[int, int] = bcrx_state.subs
if len(subs) == 1:
assert id(first_bstream) in subs
# ^^TODO! the bcrx should always de-allocate all subs,
# including the implicit first one allocated on entry
# by the first subscribing peer task, no?
#
# -[ ] adjust `MsgStream.subscribe()` to do this mgmt!
# |_ allows reverting `MsgStream.receive()` to the
# non-bcaster method.
# |_ we can decide whether to reset `._broadcaster`?
#
# await tractor.pause(shield=True)
def test_open_local_sub_to_stream(
debug_mode: bool,
):
'''
Verify a single inter-actor stream can can be fanned-out shared to
N local tasks using `trionics.maybe_open_context()`.
'''
timeout: float = 3.6
if platform.system() == "Windows":
timeout: float = 10
if debug_mode:
timeout = 999
print(f'IN debug_mode, setting large timeout={timeout!r}..')
async def main():
full = list(range(1000))
an: tractor.ActorNursery|None = None
num_tasks: int = 10
async def get_sub_and_pull(taskname: str):
nonlocal an
stream: tractor.MsgStream
async with (
maybe_open_stream(taskname) as (
an,
stream,
),
):
if '0' in taskname:
assert isinstance(stream, tractor.MsgStream)
else:
assert isinstance(
stream,
tractor.trionics.BroadcastReceiver
)
first = await stream.receive()
print(f'{taskname} started with value {first}')
seq: list[int] = []
async for msg in stream:
seq.append(msg)
assert set(seq).issubset(set(full))
# end of @acm block
print(f'{taskname} finished')
root: tractor.Actor
with trio.fail_after(timeout) as cs:
# TODO: turns out this isn't multi-task entrant XD
# We probably need an indepotent entry semantic?
async with tractor.open_root_actor(
debug_mode=debug_mode,
# maybe_enable_greenback=True,
#
# ^TODO? doesn't seem to mk breakpoint() usage work
# bc each bg task needs to open a portal??
# - [ ] we should consider making this part of
# our taskman defaults?
# |_see https://github.com/goodboy/tractor/pull/363
#
) as root:
assert root.is_registrar
async with (
trio.open_nursery() as tn,
):
for i in range(num_tasks):
tn.start_soon(
get_sub_and_pull,
f'task_{i}',
)
await trio.sleep(0.001)
print('all consumer tasks finished!')
# ?XXX, ensure actor-nursery is shutdown or we might
# hang here due to a minor task deadlock/race-condition?
#
# - seems that all we need is a checkpoint to ensure
# the last suspended task, which is inside
# `.maybe_open_context()`, can do the
# `Portal.cancel_actor()` call?
#
# - if that bg task isn't resumed, then this blocks
# timeout might hit before that?
#
if root.ipc_server.has_peers():
await trio.lowlevel.checkpoint()
# alt approach, cancel the entire `an`
# await tractor.pause()
# await an.cancel()
# end of runtime scope
print('root actor terminated.')
if cs.cancelled_caught:
pytest.fail(
'Should NOT time out in `open_root_actor()` ?'
)
print('exiting main.')
trio.run(main)
@acm
async def maybe_cancel_outer_cs(
cs: trio.CancelScope|None = None,
delay: float = 0,
):
# on first task delay this enough to block
# the 2nd task but then cancel it mid sleep
# so that the tn.start() inside the key-err handler block
# is cancelled and would previously corrupt the
# mutext state.
log.info(f'task entering sleep({delay})')
await trio.sleep(delay)
if cs:
log.info('task calling cs.cancel()')
cs.cancel()
yield
if cs:
await trio.sleep_forever()
# XXX, if not cancelled we'll leak this inf-blocking
# subtask to the actor's service tn..
else:
await trio.lowlevel.checkpoint()
@pytest.mark.parametrize(
'delay',
[0.05, 0.5, 1],
ids="pre_sleep_delay={}".format,
)
@pytest.mark.parametrize(
'cancel_by_cs',
[True, False],
ids="cancel_by_cs={}".format,
)
def test_lock_not_corrupted_on_fast_cancel(
delay: float,
cancel_by_cs: bool,
debug_mode: bool,
loglevel: str,
):
'''
Verify that if the caching-task (the first to enter
`maybe_open_context()`) is cancelled mid-cache-miss, the embedded
mutex can never be left in a corrupted state.
That is, the lock is always eventually released ensuring a peer
(cache-hitting) task will never,
- be left to inf-block/hang on the `lock.acquire()`.
- try to release the lock when still owned by the caching-task
due to it having erronously exited without calling
`lock.release()`.
'''
async def use_moc(
delay: float,
cs: trio.CancelScope|None = None,
):
log.info('task entering moc')
async with maybe_open_context(
maybe_cancel_outer_cs,
kwargs={
'cs': cs,
'delay': delay,
},
) as (cache_hit, _null):
if cache_hit:
log.info('2nd task entered')
else:
log.info('1st task entered')
if cs:
await trio.sleep_forever()
else:
await trio.sleep(delay)
# ^END, exit shared ctx.
async def main():
with trio.fail_after(delay + 2):
async with (
tractor.open_root_actor(
debug_mode=debug_mode,
loglevel=loglevel,
),
# ?TODO, pass this as the parent tn?
trio.open_nursery() as tn,
):
get_console_log('info')
log.info('yo starting')
cs = tn.cancel_scope
tn.start_soon(
use_moc,
delay,
cs if cancel_by_cs else None,
name='child',
)
with trio.CancelScope() as rent_cs:
await use_moc(
delay=delay,
cs=rent_cs if cancel_by_cs else None,
)
trio.run(main)
@acm
async def acm_with_resource(resource_id: str):
'''
Yield `resource_id` as the cached value.
Used to verify per-`ctx_key` isolation when the same
`acm_func` is called with different kwargs.
'''
yield resource_id
def test_per_ctx_key_resource_lifecycle(
debug_mode: bool,
loglevel: str,
):
'''
Verify that `maybe_open_context()` correctly isolates resource
lifecycle **per `ctx_key`** when the same `acm_func` is called
with different kwargs.
Previously `_Cache.users` was a single global `int` and
`_Cache.locks` was keyed on `fid` (function ID), so calling
the same `acm_func` with different kwargs (producing different
`ctx_key`s) meant:
- teardown for one key was skipped bc the *other* key's users
kept the global count > 0,
- and re-entry could hit the old
`assert not resources.get(ctx_key)` crash during the
teardown window.
This was the root cause of a long-standing bug in piker's
`brokerd.kraken` backend.
'''
timeout: float = 6
if debug_mode:
timeout = 999
async def main():
a_ready = trio.Event()
a_exit = trio.Event()
async def hold_resource_a():
'''
Open resource 'a' and keep it alive until signalled.
'''
async with maybe_open_context(
acm_with_resource,
kwargs={'resource_id': 'a'},
) as (cache_hit, value):
assert not cache_hit
assert value == 'a'
log.info("resource 'a' entered (holding)")
a_ready.set()
await a_exit.wait()
log.info("resource 'a' exiting")
with trio.fail_after(timeout):
async with (
tractor.open_root_actor(
debug_mode=debug_mode,
loglevel=loglevel,
),
trio.open_nursery() as tn,
):
# Phase 1: bg task holds resource 'a' open.
tn.start_soon(hold_resource_a)
await a_ready.wait()
# Phase 2: open resource 'b' (different kwargs,
# same acm_func) then exit it while 'a' is still
# alive.
async with maybe_open_context(
acm_with_resource,
kwargs={'resource_id': 'b'},
) as (cache_hit, value):
assert not cache_hit
assert value == 'b'
log.info("resource 'b' entered")
log.info("resource 'b' exited, waiting for teardown")
await trio.lowlevel.checkpoint()
# Phase 3: re-open 'b'; must be a fresh cache MISS
# proving 'b' was torn down independently of 'a'.
#
# With the old global `_Cache.users` counter this
# would be a stale cache HIT (leaked resource) or
# trigger `assert not resources.get(ctx_key)`.
async with maybe_open_context(
acm_with_resource,
kwargs={'resource_id': 'b'},
) as (cache_hit, value):
assert not cache_hit, (
"resource 'b' was NOT torn down despite "
"having zero users! (global user count bug)"
)
assert value == 'b'
log.info(
"resource 'b' re-entered "
"(cache miss, correct)"
)
# Phase 4: let 'a' exit, clean shutdown.
a_exit.set()
trio.run(main)
# @pytest.mark.xfail(
# reason=(
# 'Demonstrates the `_Cache.run_ctx` teardown race: '
# 'a re-entering task hits '
# '`assert not resources.get(ctx_key)` because '
# '`values` was popped but `resources` was not yet '
# '(acm `__aexit__` checkpoint in between). '
# 'Fixed by per-`ctx_key` locking in 9e49eddd.'
# ),
# raises=AssertionError,
# )
def test_moc_reentry_during_teardown(
debug_mode: bool,
loglevel: str,
):
'''
Reproduce the piker `open_cached_client('kraken')` race:
- same `acm_func`, NO kwargs (identical `ctx_key`)
- multiple tasks share the cached resource
- all users exit -> teardown starts
- a NEW task enters during `_Cache.run_ctx.__aexit__`
- `values[ctx_key]` is gone (popped in inner finally)
but `resources[ctx_key]` still exists (outer finally
hasn't run yet bc the acm cleanup has checkpoints)
- old code: `assert not resources.get(ctx_key)` FIRES
This models the real-world scenario where `brokerd.kraken`
tasks concurrently call `open_cached_client('kraken')`
(same `acm_func`, empty kwargs, shared `ctx_key`) and
the teardown/re-entry race triggers intermittently.
'''
async def main():
in_aexit = trio.Event()
@acm
async def cached_client():
'''
Simulates `kraken.api.get_client()`:
- no params (all callers share one `ctx_key`)
- slow-ish cleanup to widen the race window
between `values.pop()` and `resources.pop()`
inside `_Cache.run_ctx`.
'''
yield 'the-client'
# Signal that we're in __aexit__ — at this
# point `values` has already been popped by
# `run_ctx`'s inner finally, but `resources`
# is still alive (outer finally hasn't run).
in_aexit.set()
await trio.sleep(10)
first_done = trio.Event()
async def use_and_exit():
async with maybe_open_context(
cached_client,
) as (cache_hit, value):
assert value == 'the-client'
first_done.set()
async def reenter_during_teardown():
'''
Wait for the acm's `__aexit__` to start (meaning
`values` is popped but `resources` still exists),
then re-enter — triggering the assert.
'''
await in_aexit.wait()
async with maybe_open_context(
cached_client,
) as (cache_hit, value):
assert value == 'the-client'
with trio.fail_after(5):
async with (
tractor.open_root_actor(
debug_mode=debug_mode,
loglevel=loglevel,
),
collapse_eg(),
trio.open_nursery() as tn,
):
tn.start_soon(use_and_exit)
tn.start_soon(reenter_during_teardown)
trio.run(main)