Unwind process opening and shield hard reap

immediate_remote_cancels
Tyler Goodlet 2021-10-12 23:59:39 -04:00
parent bb9d9c74b1
commit 46ff558556
1 changed files with 72 additions and 91 deletions

View File

@ -174,11 +174,30 @@ async def do_hard_kill(
proc.kill()
@asynccontextmanager
async def spawn_subactor(
subactor: 'Actor',
async def new_proc(
name: str,
actor_nursery: 'ActorNursery', # type: ignore # noqa
subactor: Actor,
errors: Dict[Tuple[str, str], Exception],
# passed through to actor main
bind_addr: Tuple[str, int],
parent_addr: Tuple[str, int],
):
_runtime_vars: Dict[str, Any], # serialized and sent to _child
*,
task_status: TaskStatus[Portal] = trio.TASK_STATUS_IGNORED
) -> None:
"""
Create a new ``multiprocessing.Process`` using the
spawn method as configured using ``try_set_start_method()``.
"""
cancel_scope = None
# mark the new actor with the global spawn method
subactor._spawn_method = _spawn_method
if _spawn_method == 'trio':
spawn_cmd = [
sys.executable,
"-m",
@ -202,46 +221,9 @@ async def spawn_subactor(
subactor.loglevel
]
proc = await trio.open_process(spawn_cmd)
try:
yield proc
proc = await trio.open_process(spawn_cmd)
finally:
log.runtime(f"Attempting to kill {proc}")
# XXX: do this **after** cancellation/tearfown
# to avoid killing the process too early
# since trio does this internally on ``__aexit__()``
await do_hard_kill(proc)
async def new_proc(
name: str,
actor_nursery: 'ActorNursery', # type: ignore # noqa
subactor: Actor,
errors: Dict[Tuple[str, str], Exception],
# passed through to actor main
bind_addr: Tuple[str, int],
parent_addr: Tuple[str, int],
_runtime_vars: Dict[str, Any], # serialized and sent to _child
*,
task_status: TaskStatus[Portal] = trio.TASK_STATUS_IGNORED
) -> None:
"""Create a new ``multiprocessing.Process`` using the
spawn method as configured using ``try_set_start_method()``.
"""
cancel_scope = None
# mark the new actor with the global spawn method
subactor._spawn_method = _spawn_method
if _spawn_method == 'trio':
async with trio.open_nursery() as nursery:
async with spawn_subactor(
subactor,
parent_addr,
) as proc:
log.runtime(f"Started {proc}")
# wait for actor to spawn and connect back to us
@ -274,6 +256,7 @@ async def new_proc(
with trio.CancelScope(shield=True):
await actor_nursery._join_procs.wait()
async with trio.open_nursery() as nursery:
if portal in actor_nursery._cancel_after_result_on_exit:
cancel_scope = await nursery.start(
cancel_on_completion,
@ -285,25 +268,10 @@ async def new_proc(
# Wait for proc termination but **dont' yet** call
# ``trio.Process.__aexit__()`` (it tears down stdio
# which will kill any waiting remote pdb trace).
# TODO: No idea how we can enforce zombie
# reaping more stringently without the shield
# we used to have below...
# with trio.CancelScope(shield=True):
# async with proc:
# Always "hard" join sub procs since no actor zombies
# are allowed!
# this is a "light" (cancellable) join, the hard join is
# in the enclosing scope (see above).
# This is a "soft" (cancellable) join/reap.
await proc.wait()
log.debug(f"Joined {proc}")
# pop child entry to indicate we no longer managing this subactor
subactor, proc, portal = actor_nursery._children.pop(subactor.uid)
finally:
# cancel result waiter that may have been spawned in
# tandem if not done already
if cancel_scope:
@ -311,6 +279,19 @@ async def new_proc(
"Cancelling existing result waiter task for "
f"{subactor.uid}")
cancel_scope.cancel()
log.runtime(f"Attempting to kill {proc}")
# The "hard" reap since no actor zombies are allowed!
# XXX: do this **after** cancellation/tearfown to avoid
# killing the process too early.
with trio.CancelScope(shield=True):
await do_hard_kill(proc)
log.debug(f"Joined {proc}")
# pop child entry to indicate we no longer managing this subactor
subactor, proc, portal = actor_nursery._children.pop(subactor.uid)
else:
# `multiprocessing`
# async with trio.open_nursery() as nursery: