forked from goodboy/tractor
1
0
Fork 0

Be ultra-correct in `Portal.open_context()`

This took way too long to get right but hopefully will give us grok-able
and correct context exit semantics going forward B)

The main fixes were:
- always shielding the `MsgStream.aclose()` call on teardown to avoid
  bubbling a `Cancelled`.
- properly absorbing any `ContextCancelled` in cases due to "self
  cancellation" using the new `Context.canceller` in the logic.
- capturing any error raised by the `Context.result()` call in the
  "normal exit, result received" case and setting it as the
  `Context._local_error` so that self-cancels can be easily measured via
  `Context.cancelled_caught` in same way as remote-error caused
  cancellations.
- extremely detailed comments around all of the cancellation-error cases
  to avoid ever getting confused about the control flow in the future XD
multihomed
Tyler Goodlet 2023-10-23 17:34:28 -04:00
parent 131674eabd
commit df31047ecb
1 changed files with 118 additions and 52 deletions

View File

@ -48,6 +48,7 @@ from ._exceptions import (
unpack_error,
NoResult,
ContextCancelled,
MessagingError,
)
from ._context import Context
from ._streaming import MsgStream
@ -71,11 +72,6 @@ def _unwrap_msg(
raise unpack_error(msg, channel) from None
# TODO: maybe move this to ._exceptions?
class MessagingError(Exception):
'Some kind of unexpected SC messaging dialog issue'
class Portal:
'''
A 'portal' to a memory-domain-separated `Actor`.
@ -220,14 +216,18 @@ class Portal:
try:
# send cancel cmd - might not get response
# XXX: sure would be nice to make this work with a proper shield
# XXX: sure would be nice to make this work with
# a proper shield
with trio.move_on_after(
timeout
or self.cancel_timeout
) as cs:
cs.shield = True
await self.run_from_ns('self', 'cancel')
await self.run_from_ns(
'self',
'cancel',
)
return True
if cs.cancelled_caught:
@ -462,10 +462,14 @@ class Portal:
try:
# the "first" value here is delivered by the callee's
# ``Context.started()`` call.
first = msg['started']
first: Any = msg['started']
ctx._started_called: bool = True
except KeyError:
# TODO: can we maybe factor this into the new raiser
# `_streaming._raise_from_no_yield_msg()` and make that
# helper more generic, say with a `_no_<blah>_msg()`?
if not (cid := msg.get('cid')):
raise MessagingError(
'Received internal error at context?\n'
@ -517,54 +521,102 @@ class Portal:
# started in the ctx nursery.
ctx._scope.cancel()
# XXX: (maybe) shield/mask context-cancellations that were
# initiated by any of the context's 2 tasks. There are
# subsequently 2 operating cases for a "graceful cancel"
# of a `Context`:
# XXX NOTE XXX: maybe shield against
# self-context-cancellation (which raises a local
# `ContextCancelled`) when requested (via
# `Context.cancel()`) by the same task (tree) which entered
# THIS `.open_context()`.
#
# 1.*this* side's task called `Context.cancel()`, in
# which case we mask the `ContextCancelled` from bubbling
# to the opener (much like how `trio.Nursery` swallows
# any `trio.Cancelled` bubbled by a call to
# NOTE: There are 2 operating cases for a "graceful cancel"
# of a `Context`. In both cases any `ContextCancelled`
# raised in this scope-block came from a transport msg
# relayed from some remote-actor-task which our runtime set
# as a `Context._remote_error`
#
# the CASES:
#
# - if that context IS THE SAME ONE that called
# `Context.cancel()`, we want to absorb the error
# silently and let this `.open_context()` block to exit
# without raising.
#
# - if it is from some OTHER context (we did NOT call
# `.cancel()`), we want to re-RAISE IT whilst also
# setting our own ctx's "reason for cancel" to be that
# other context's cancellation condition; we set our
# `.canceller: tuple[str, str]` to be same value as
# caught here in a `ContextCancelled.canceller`.
#
# Again, there are 2 cases:
#
# 1-some other context opened in this `.open_context()`
# block cancelled due to a self or peer cancellation
# request in which case we DO let the error bubble to the
# opener.
#
# 2-THIS "caller" task somewhere invoked `Context.cancel()`
# and received a `ContextCanclled` from the "callee"
# task, in which case we mask the `ContextCancelled` from
# bubbling to this "caller" (much like how `trio.Nursery`
# swallows any `trio.Cancelled` bubbled by a call to
# `Nursery.cancel_scope.cancel()`)
#
# 2.*the other* side's (callee/spawned) task cancelled due
# to a self or peer cancellation request in which case we
# DO let the error bubble to the opener.
except ContextCancelled as ctxc:
scope_err = ctxc
# CASE 1: this context was never cancelled
# via a local task's call to `Context.cancel()`.
if not ctx._cancel_called:
raise
# CASE 2: context was cancelled by local task calling
# `.cancel()`, we don't raise and the exit block should
# exit silently.
else:
if (
ctx._cancel_called
and (
ctxc is ctx._remote_error
or
ctxc.canceller is self.canceller
)
):
log.debug(
f'Context {ctx} cancelled gracefully with:\n'
f'{ctxc}'
)
# CASE 1: this context was never cancelled via a local
# task (tree) having called `Context.cancel()`, raise
# the error since it was caused by someone else!
else:
raise
# the above `._scope` can be cancelled due to:
# 1. an explicit self cancel via `Context.cancel()` or
# `Actor.cancel()`,
# 2. any "callee"-side remote error, possibly also a cancellation
# request by some peer,
# 3. any "caller" (aka THIS scope's) local error raised in the above `yield`
except (
# - a standard error in the caller/yieldee
# CASE 3: standard local error in this caller/yieldee
Exception,
# - a runtime teardown exception-group and/or
# cancellation request from a caller task.
BaseExceptionGroup,
trio.Cancelled,
# CASES 1 & 2: normally manifested as
# a `Context._scope_nursery` raised
# exception-group of,
# 1.-`trio.Cancelled`s, since
# `._scope.cancel()` will have been called and any
# `ContextCancelled` absorbed and thus NOT RAISED in
# any `Context._maybe_raise_remote_err()`,
# 2.-`BaseExceptionGroup[ContextCancelled | RemoteActorError]`
# from any error raised in the "callee" side with
# a group only raised if there was any more then one
# task started here in the "caller" in the
# `yield`-ed to task.
BaseExceptionGroup, # since overrun handler tasks may have been spawned
trio.Cancelled, # NOTE: NOT from inside the ctx._scope
KeyboardInterrupt,
) as err:
scope_err = err
# XXX: request cancel of this context on any error.
# NOTE: `Context.cancel()` is conversely NOT called in
# the `ContextCancelled` "cancellation requested" case
# above.
# XXX: ALWAYS request the context to CANCEL ON any ERROR.
# NOTE: `Context.cancel()` is conversely NEVER CALLED in
# the `ContextCancelled` "self cancellation absorbed" case
# handled in the block above!
log.cancel(
'Context cancelled for task due to\n'
f'{err}\n'
@ -583,7 +635,7 @@ class Portal:
raise # duh
# no scope error case
# no local scope error, the "clean exit with a result" case.
else:
if ctx.chan.connected():
log.info(
@ -597,15 +649,27 @@ class Portal:
# `Context._maybe_raise_remote_err()`) IFF
# a `Context._remote_error` was set by the runtime
# via a call to
# `Context._maybe_cancel_and_set_remote_error()`
# which IS SET any time the far end fails and
# causes "caller side" cancellation via
# a `ContextCancelled` here.
# `Context._maybe_cancel_and_set_remote_error()`.
# As per `Context._deliver_msg()`, that error IS
# ALWAYS SET any time "callee" side fails and causes "caller
# side" cancellation via a `ContextCancelled` here.
# result = await ctx.result()
try:
result = await ctx.result()
log.runtime(
f'Context {fn_name} returned value from callee:\n'
f'`{result}`'
)
except BaseException as berr:
# on normal teardown, if we get some error
# raised in `Context.result()` we still want to
# save that error on the ctx's state to
# determine things like `.cancelled_caught` for
# cases where there was remote cancellation but
# this task didn't know until final teardown
# / value collection.
scope_err = berr
raise
finally:
# though it should be impossible for any tasks
@ -655,12 +719,14 @@ class Portal:
with trio.CancelScope(shield=True):
await ctx._recv_chan.aclose()
# XXX: since we always (maybe) re-raise (and thus also
# mask runtime machinery related
# multi-`trio.Cancelled`s) any scope error which was
# the underlying cause of this context's exit, add
# different log msgs for each of the (2) cases.
# XXX: we always raise remote errors locally and
# generally speaking mask runtime-machinery related
# multi-`trio.Cancelled`s. As such, any `scope_error`
# which was the underlying cause of this context's exit
# should be stored as the `Context._local_error` and
# used in determining `Context.cancelled_caught: bool`.
if scope_err is not None:
ctx._local_error: BaseException = scope_err
etype: Type[BaseException] = type(scope_err)
# CASE 2
@ -690,7 +756,7 @@ class Portal:
await maybe_wait_for_debugger()
# FINALLY, remove the context from runtime tracking and
# exit Bo
# exit!
self.actor._contexts.pop(
(self.channel.uid, ctx.cid),
None,