Compare commits

..

No commits in common. "fec2ba004c633be3782917a0fecc339b928d2e22" and "0ab49cd244ed8e74628512e44e3e17831b6d80f3" have entirely different histories.

4 changed files with 185 additions and 216 deletions

View File

@ -74,7 +74,7 @@ jobs:
run: pip install -U . -r requirements-test.txt -r requirements-docs.txt --upgrade-strategy eager
- name: Run tests
run: pytest tests/ --spawn-backend=${{ matrix.spawn_backend }} -rs -v
run: pytest tests/ --spawn-backend=${{ matrix.spawn_backend }} -rs -v --full-trace
# We skip 3.10 on windows for now due to
# https://github.com/pytest-dev/pytest/issues/8733

View File

@ -18,10 +18,6 @@ import time
import pytest
import pexpect
from pexpect.exceptions import (
TIMEOUT,
EOF,
)
from conftest import repodir
@ -376,11 +372,10 @@ def test_multi_subactors(
spawn,
ctlc: bool,
):
'''
Multiple subactors, both erroring and
breakpointing as well as a nested subactor erroring.
'''
"""
Multiple subactors, both erroring and breakpointing as well as
a nested subactor erroring.
"""
child = spawn(r'multi_subactors')
# scan for the pdbpp prompt
@ -419,16 +414,14 @@ def test_multi_subactors(
# 2nd name_error failure
child.expect(r"\(Pdb\+\+\)")
# XXX: lol honestly no idea why CI is cuck but
# seems like this likely falls into our unhandled nested
# case and isn't working in that env due to raciness..
from conftest import _ci_env
if not ctlc and _ci_env:
name = 'name_error' if ctlc else 'name_error_1'
assert_before(child, [
f"Attaching to pdb in crashed actor: ('{name}'",
"NameError",
])
assert_before(child, [
"Attaching to pdb in crashed actor: ('name_error_1'",
"NameError",
])
# before = str(child.before.decode())
# assert "Attaching to pdb in crashed actor: ('name_error_1'" in before
# assert "NameError" in before
if ctlc:
do_ctlc(child)
@ -458,11 +451,11 @@ def test_multi_subactors(
do_ctlc(child)
# 2nd depth nursery should trigger
if not ctlc:
assert_before(child, [
spawn_err,
"RemoteActorError: ('name_error_1'",
])
# child.sendline('c')
# child.expect(r"\(Pdb\+\+\)")
# before = str(child.before.decode())
assert spawn_err in before
assert "RemoteActorError: ('name_error_1'" in before
# now run some "continues" to show re-entries
for _ in range(5):
@ -471,50 +464,31 @@ def test_multi_subactors(
# quit the loop and expect parent to attach
child.sendline('q')
child.expect(r"\(Pdb\+\+\)")
before = str(child.before.decode())
try:
child.expect(r"\(Pdb\+\+\)")
except TIMEOUT:
if _ci_env and not ctlc:
raise
# in ci seems like this can sometimes just result
# in full tree death?
print('tree died?')
else:
before = str(child.before.decode())
assert_before(child, [
# debugger attaches to root
"Attaching to pdb in crashed actor: ('root'",
# expect a multierror with exceptions for each sub-actor
"RemoteActorError: ('breakpoint_forever'",
"RemoteActorError: ('name_error'",
"RemoteActorError: ('spawn_error'",
"RemoteActorError: ('name_error_1'",
'bdb.BdbQuit',
])
# debugger attaches to root
assert "Attaching to pdb in crashed actor: ('root'" in before
# expect a multierror with exceptions for each sub-actor
assert "RemoteActorError: ('breakpoint_forever'" in before
assert "RemoteActorError: ('name_error'" in before
assert "RemoteActorError: ('spawn_error'" in before
assert "RemoteActorError: ('name_error_1'" in before
assert 'bdb.BdbQuit' in before
if ctlc:
do_ctlc(child)
# process should exit
child.sendline('c')
try:
child.expect(pexpect.EOF)
except TIMEOUT:
child.expect(r"\(Pdb\+\+\)")
child.expect(pexpect.EOF)
# repeat of previous multierror for final output
assert_before(child, [
"RemoteActorError: ('breakpoint_forever'",
"RemoteActorError: ('name_error'",
"RemoteActorError: ('spawn_error'",
"RemoteActorError: ('name_error_1'",
'bdb.BdbQuit',
])
before = str(child.before.decode())
assert "RemoteActorError: ('breakpoint_forever'" in before
assert "RemoteActorError: ('name_error'" in before
assert "RemoteActorError: ('spawn_error'" in before
assert "RemoteActorError: ('name_error_1'" in before
assert 'bdb.BdbQuit' in before
def test_multi_daemon_subactors(
@ -570,7 +544,7 @@ def test_multi_daemon_subactors(
# now the root actor won't clobber the bp_forever child
# during it's first access to the debug lock, but will instead
# wait for the lock to release, by the edge triggered
# ``_debug.Lock.no_remote_has_tty`` event before sending cancel messages
# ``_debug._no_remote_has_tty`` event before sending cancel messages
# (via portals) to its underlings B)
# at some point here there should have been some warning msg from
@ -603,7 +577,7 @@ def test_multi_daemon_subactors(
child.sendline('c')
child.expect(pexpect.EOF)
except TIMEOUT:
except pexpect.exceptions.TIMEOUT:
# Failed to exit using continue..?
child.sendline('q')
child.expect(pexpect.EOF)
@ -633,45 +607,32 @@ def test_multi_subactors_root_errors(
# continue again to catch 2nd name error from
# actor 'name_error_1' (which is 2nd depth).
child.sendline('c')
try:
child.expect(r"\(Pdb\+\+\)")
except TIMEOUT:
child.sendline('')
# XXX: lol honestly no idea why CI is cuck but
# seems like this likely falls into our unhandled nested
# case and isn't working in that env due to raciness..
from conftest import _ci_env
if not ctlc and _ci_env:
name = 'name_error' if ctlc else 'name_error_1'
assert_before(child, [
f"Attaching to pdb in crashed actor: ('{name}'",
"NameError",
])
child.expect(r"\(Pdb\+\+\)")
before = str(child.before.decode())
assert "Attaching to pdb in crashed actor: ('name_error_1'" in before
assert "NameError" in before
if ctlc:
do_ctlc(child)
child.sendline('c')
child.expect(r"\(Pdb\+\+\)")
assert_before(child, [
"Attaching to pdb in crashed actor: ('spawn_error'",
# boxed error from previous step
"RemoteActorError: ('name_error_1'",
"NameError",
])
before = str(child.before.decode())
assert "Attaching to pdb in crashed actor: ('spawn_error'" in before
# boxed error from previous step
assert "RemoteActorError: ('name_error_1'" in before
assert "NameError" in before
if ctlc:
do_ctlc(child)
child.sendline('c')
child.expect(r"\(Pdb\+\+\)")
assert_before(child, [
"Attaching to pdb in crashed actor: ('root'",
# boxed error from previous step
"RemoteActorError: ('name_error'",
"NameError",
])
before = str(child.before.decode())
assert "Attaching to pdb in crashed actor: ('root'" in before
# boxed error from first level failure
assert "RemoteActorError: ('name_error'" in before
assert "NameError" in before
# warnings assert we probably don't need
# assert "Cancelling nursery in ('spawn_error'," in before
@ -715,7 +676,7 @@ def test_multi_nested_subactors_error_through_nurseries(
child.sendline('c')
time.sleep(0.1)
except EOF:
except pexpect.exceptions.EOF:
# race conditions on how fast the continue is sent?
print(f"Failed early on {i}?")
@ -761,8 +722,8 @@ def test_root_nursery_cancels_before_child_releases_tty_lock(
child.expect(r"\(Pdb\+\+\)")
except (
EOF,
TIMEOUT,
pexpect.exceptions.EOF,
pexpect.exceptions.TIMEOUT,
):
# races all over..
@ -783,11 +744,11 @@ def test_root_nursery_cancels_before_child_releases_tty_lock(
child.sendline('c')
time.sleep(0.1)
for i in range(3):
for i in range(10):
try:
child.expect(pexpect.EOF)
break
except TIMEOUT:
except pexpect.exceptions.TIMEOUT:
child.sendline('c')
time.sleep(0.1)
print('child was able to grab tty lock again?')

View File

@ -967,7 +967,7 @@ class Actor:
# don't start entire actor runtime
# cancellation if this actor is in debug
# mode
pdb_complete = _debug.Lock.local_pdb_complete
pdb_complete = _debug._local_pdb_complete
if pdb_complete:
await pdb_complete.wait()
@ -1413,7 +1413,7 @@ class Actor:
# kill any debugger request task to avoid deadlock
# with the root actor in this tree
dbcs = _debug.Lock._debugger_request_cs
dbcs = _debug._debugger_request_cs
if dbcs is not None:
log.cancel("Cancelling active debugger request")
dbcs.cancel()

View File

@ -60,81 +60,26 @@ log = get_logger(__name__)
__all__ = ['breakpoint', 'post_mortem']
class Lock:
'''
Actor global debug lock state.
# TODO: wrap all these in a static global class: ``DebugLock`` maybe?
Mostly to avoid a lot of ``global`` declarations for now XD.
# placeholder for function to set a ``trio.Event`` on debugger exit
_pdb_release_hook: Optional[Callable] = None
'''
# placeholder for function to set a ``trio.Event`` on debugger exit
pdb_release_hook: Optional[Callable] = None
# actor-wide variable pointing to current task name using debugger
_local_task_in_debug: Optional[str] = None
# actor-wide variable pointing to current task name using debugger
local_task_in_debug: Optional[str] = None
# actor tree-wide actor uid that supposedly has the tty lock
_global_actor_in_debug: Optional[Tuple[str, str]] = None
# actor tree-wide actor uid that supposedly has the tty lock
global_actor_in_debug: Optional[Tuple[str, str]] = None
# lock in root actor preventing multi-access to local tty
_debug_lock: trio.StrictFIFOLock = trio.StrictFIFOLock()
_local_pdb_complete: Optional[trio.Event] = None
_no_remote_has_tty: Optional[trio.Event] = None
local_pdb_complete: Optional[trio.Event] = None
no_remote_has_tty: Optional[trio.Event] = None
# lock in root actor preventing multi-access to local tty
_debug_lock: trio.StrictFIFOLock = trio.StrictFIFOLock()
# XXX: set by the current task waiting on the root tty lock
# and must be cancelled if this actor is cancelled via message
# otherwise deadlocks with the parent actor may ensure
_debugger_request_cs: Optional[trio.CancelScope] = None
_orig_sigint_handler: Optional[Callable] = None
@classmethod
def shield_sigint(cls):
cls._orig_sigint_handler = signal.signal(
signal.SIGINT,
shield_sigint,
)
@classmethod
def unshield_sigint(cls):
if cls._orig_sigint_handler is not None:
# restore original sigint handler
signal.signal(
signal.SIGINT,
cls._orig_sigint_handler
)
cls._orig_sigint_handler = None
@classmethod
def maybe_release(cls):
cls.local_task_in_debug = None
if cls.pdb_release_hook:
cls.pdb_release_hook()
@classmethod
def root_release(cls):
try:
cls._debug_lock.release()
except RuntimeError:
# uhhh makes no sense but been seeing the non-owner
# release error even though this is definitely the task
# that locked?
owner = cls._debug_lock.statistics().owner
if owner:
raise
cls.global_actor_in_debug = None
cls.local_task_in_debug = None
try:
# sometimes the ``trio`` might already be terminated in
# which case this call will raise.
cls.local_pdb_complete.set()
finally:
# restore original sigint handler
cls.unshield_sigint()
# XXX: set by the current task waiting on the root tty lock
# and must be cancelled if this actor is cancelled via message
# otherwise deadlocks with the parent actor may ensure
_debugger_request_cs: Optional[trio.CancelScope] = None
class TractorConfig(pdbpp.DefaultConfig):
@ -163,13 +108,13 @@ class MultiActorPdb(pdbpp.Pdb):
try:
super().set_continue()
finally:
Lock.maybe_release()
maybe_release()
def set_quit(self):
try:
super().set_quit()
finally:
Lock.maybe_release()
maybe_release()
# TODO: will be needed whenever we get to true remote debugging.
@ -208,6 +153,14 @@ class MultiActorPdb(pdbpp.Pdb):
# log.info("Closing stdin hijack")
# break
# TODO: make this method on a global lock type!
def maybe_release():
global _local_task_in_debug, _pdb_release_hook
_local_task_in_debug = None
if _pdb_release_hook:
_pdb_release_hook()
@acm
async def _acquire_debug_lock(
uid: Tuple[str, str]
@ -222,6 +175,8 @@ async def _acquire_debug_lock(
to the ``pdb`` repl.
'''
global _debug_lock, _global_actor_in_debug, _no_remote_has_tty
task_name = trio.lowlevel.current_task().name
log.runtime(
@ -235,15 +190,15 @@ async def _acquire_debug_lock(
f"entering lock checkpoint, remote task: {task_name}:{uid}"
)
we_acquired = True
await Lock._debug_lock.acquire()
await _debug_lock.acquire()
if Lock.no_remote_has_tty is None:
if _no_remote_has_tty is None:
# mark the tty lock as being in use so that the runtime
# can try to avoid clobbering any connection from a child
# that's currently relying on it.
Lock.no_remote_has_tty = trio.Event()
_no_remote_has_tty = trio.Event()
Lock.global_actor_in_debug = uid
_global_actor_in_debug = uid
log.runtime(f"TTY lock acquired, remote task: {task_name}:{uid}")
# NOTE: critical section: this yield is unshielded!
@ -256,32 +211,32 @@ async def _acquire_debug_lock(
# surrounding caller side context should cancel normally
# relaying back to the caller.
yield Lock._debug_lock
yield _debug_lock
finally:
# if Lock.global_actor_in_debug == uid:
# if _global_actor_in_debug == uid:
if (
we_acquired
and Lock._debug_lock.locked()
and _debug_lock.locked()
):
Lock._debug_lock.release()
_debug_lock.release()
# IFF there are no more requesting tasks queued up fire, the
# "tty-unlocked" event thereby alerting any monitors of the lock that
# we are now back in the "tty unlocked" state. This is basically
# and edge triggered signal around an empty queue of sub-actor
# tasks that may have tried to acquire the lock.
stats = Lock._debug_lock.statistics()
stats = _debug_lock.statistics()
if (
not stats.owner
):
log.runtime(f"No more tasks waiting on tty lock! says {uid}")
if Lock.no_remote_has_tty is not None:
Lock.no_remote_has_tty.set()
Lock.no_remote_has_tty = None
if _no_remote_has_tty is not None:
_no_remote_has_tty.set()
_no_remote_has_tty = None
Lock.global_actor_in_debug = None
_global_actor_in_debug = None
log.runtime(
f"TTY lock released, remote task: {task_name}:{uid}"
@ -316,8 +271,11 @@ async def _hijack_stdin_for_child(
)
log.debug(f"Actor {subactor_uid} is WAITING on stdin hijack lock")
Lock.shield_sigint()
orig_handler = signal.signal(
signal.SIGINT,
shield_sigint,
)
try:
with (
trio.CancelScope(shield=True),
@ -369,7 +327,10 @@ async def _hijack_stdin_for_child(
return "pdb_unlock_complete"
finally:
Lock.unshield_sigint()
signal.signal(
signal.SIGINT,
orig_handler
)
async def wait_for_parent_stdin_hijack(
@ -387,8 +348,10 @@ async def wait_for_parent_stdin_hijack(
debug (see below inside ``maybe_wait_for_debugger()``).
'''
global _debugger_request_cs
with trio.CancelScope(shield=True) as cs:
Lock._debugger_request_cs = cs
_debugger_request_cs = cs
try:
async with get_root() as portal:
@ -408,9 +371,9 @@ async def wait_for_parent_stdin_hijack(
# unblock local caller
try:
assert Lock.local_pdb_complete
assert _local_pdb_complete
task_status.started(cs)
await Lock.local_pdb_complete.wait()
await _local_pdb_complete.wait()
finally:
# TODO: shielding currently can cause hangs...
@ -427,7 +390,8 @@ async def wait_for_parent_stdin_hijack(
finally:
log.pdb(f"Exiting debugger for actor {actor_uid}")
Lock.local_task_in_debug = None
global _local_task_in_debug
_local_task_in_debug = None
log.pdb(f"Child {actor_uid} released parent stdio lock")
@ -435,8 +399,10 @@ def mk_mpdb() -> tuple[MultiActorPdb, Callable]:
pdb = MultiActorPdb()
# signal.signal = pdbpp.hideframe(signal.signal)
Lock.shield_sigint()
orig_handler = signal.signal(
signal.SIGINT,
partial(shield_sigint, pdb_obj=pdb),
)
# XXX: These are the important flags mentioned in
# https://github.com/python-trio/trio/issues/1155
@ -444,7 +410,15 @@ def mk_mpdb() -> tuple[MultiActorPdb, Callable]:
pdb.allow_kbdint = True
pdb.nosigint = True
return pdb, Lock.unshield_sigint
# TODO: add this as method on our pdb obj?
def undo_sigint():
# restore original sigint handler
signal.signal(
signal.SIGINT,
orig_handler
)
return pdb, undo_sigint
async def _breakpoint(
@ -466,6 +440,9 @@ async def _breakpoint(
actor = tractor.current_actor()
task_name = trio.lowlevel.current_task().name
global _local_pdb_complete, _pdb_release_hook
global _local_task_in_debug, _global_actor_in_debug
# TODO: is it possible to debug a trio.Cancelled except block?
# right now it seems like we can kinda do with by shielding
# around ``tractor.breakpoint()`` but not if we move the shielded
@ -473,14 +450,14 @@ async def _breakpoint(
# with trio.CancelScope(shield=shield):
# await trio.lowlevel.checkpoint()
if not Lock.local_pdb_complete or Lock.local_pdb_complete.is_set():
Lock.local_pdb_complete = trio.Event()
if not _local_pdb_complete or _local_pdb_complete.is_set():
_local_pdb_complete = trio.Event()
# TODO: need a more robust check for the "root" actor
if actor._parent_chan and not is_root_process():
if Lock.local_task_in_debug:
if Lock.local_task_in_debug == task_name:
if _local_task_in_debug:
if _local_task_in_debug == task_name:
# this task already has the lock and is
# likely recurrently entering a breakpoint
return
@ -490,18 +467,18 @@ async def _breakpoint(
# support for recursive entries to `tractor.breakpoint()`
log.warning(f"{actor.uid} already has a debug lock, waiting...")
await Lock.local_pdb_complete.wait()
await _local_pdb_complete.wait()
await trio.sleep(0.1)
# mark local actor as "in debug mode" to avoid recurrent
# entries/requests to the root process
Lock.local_task_in_debug = task_name
_local_task_in_debug = task_name
def child_release():
try:
# sometimes the ``trio`` might already be termianated in
# which case this call will raise.
Lock.local_pdb_complete.set()
_local_pdb_complete.set()
finally:
# restore original sigint handler
undo_sigint()
@ -509,7 +486,7 @@ async def _breakpoint(
# _local_task_in_debug = None
# assign unlock callback for debugger teardown hooks
Lock.pdb_release_hook = child_release
_pdb_release_hook = child_release
# this **must** be awaited by the caller and is done using the
# root nursery so that the debugger can continue to run without
@ -526,39 +503,66 @@ async def _breakpoint(
actor.uid,
)
except RuntimeError:
Lock.pdb_release_hook()
_pdb_release_hook()
raise
elif is_root_process():
# we also wait in the root-parent for any child that
# may have the tty locked prior
global _debug_lock
# TODO: wait, what about multiple root tasks acquiring it though?
# root process (us) already has it; ignore
if Lock.global_actor_in_debug == actor.uid:
if _global_actor_in_debug == actor.uid:
return
# XXX: since we need to enter pdb synchronously below,
# we have to release the lock manually from pdb completion
# callbacks. Can't think of a nicer way then this atm.
if Lock._debug_lock.locked():
if _debug_lock.locked():
log.warning(
'Root actor attempting to shield-acquire active tty lock'
f' owned by {Lock.global_actor_in_debug}')
f' owned by {_global_actor_in_debug}')
# must shield here to avoid hitting a ``Cancelled`` and
# a child getting stuck bc we clobbered the tty
with trio.CancelScope(shield=True):
await Lock._debug_lock.acquire()
await _debug_lock.acquire()
else:
# may be cancelled
await Lock._debug_lock.acquire()
await _debug_lock.acquire()
Lock.global_actor_in_debug = actor.uid
Lock.local_task_in_debug = task_name
_global_actor_in_debug = actor.uid
_local_task_in_debug = task_name
# the lock must be released on pdb completion
Lock.pdb_release_hook = Lock.root_release
def root_release():
global _local_pdb_complete, _debug_lock
global _global_actor_in_debug, _local_task_in_debug
try:
_debug_lock.release()
except RuntimeError:
# uhhh makes no sense but been seeing the non-owner
# release error even though this is definitely the task
# that locked?
owner = _debug_lock.statistics().owner
if owner:
raise
_global_actor_in_debug = None
_local_task_in_debug = None
try:
# sometimes the ``trio`` might already be termianated in
# which case this call will raise.
_local_pdb_complete.set()
finally:
# restore original sigint handler
undo_sigint()
_pdb_release_hook = root_release
try:
# block here one (at the appropriate frame *up*) where
@ -567,7 +571,7 @@ async def _breakpoint(
debug_func(actor, pdb)
except bdb.BdbQuit:
Lock.maybe_release()
maybe_release()
raise
# XXX: apparently we can't do this without showing this frame
@ -603,7 +607,8 @@ def shield_sigint(
'''
__tracebackhide__ = True
uid_in_debug = Lock.global_actor_in_debug
global _local_task_in_debug, _global_actor_in_debug
uid_in_debug = _global_actor_in_debug
actor = tractor.current_actor()
@ -676,7 +681,7 @@ def shield_sigint(
)
return do_cancel()
task = Lock.local_task_in_debug
task = _local_task_in_debug
if task:
log.pdb(
f"Ignoring SIGINT while task in debug mode: `{task}`"
@ -749,7 +754,8 @@ def _set_trace(
pdb, undo_sigint = mk_mpdb()
# we entered the global ``breakpoint()`` built-in from sync code?
Lock.local_task_in_debug = 'sync'
global _local_task_in_debug, _pdb_release_hook
_local_task_in_debug = 'sync'
pdb.set_trace(frame=frame)
@ -824,7 +830,7 @@ async def _maybe_enter_pm(err):
):
log.debug("Actor crashed, entering debug mode")
await post_mortem()
Lock.maybe_release()
maybe_release()
return True
else:
@ -869,6 +875,8 @@ async def maybe_wait_for_debugger(
if (
is_root_process()
):
global _no_remote_has_tty, _global_actor_in_debug, _wait_all_tasks_lock
# If we error in the root but the debugger is
# engaged we don't want to prematurely kill (and
# thus clobber access to) the local tty since it
@ -880,8 +888,8 @@ async def maybe_wait_for_debugger(
for _ in range(poll_steps):
if Lock.global_actor_in_debug:
sub_in_debug = tuple(Lock.global_actor_in_debug)
if _global_actor_in_debug:
sub_in_debug = tuple(_global_actor_in_debug)
# alive = tractor.current_actor().child_alive(sub_in_debug)
# if not alive:
# break
@ -897,7 +905,7 @@ async def maybe_wait_for_debugger(
# XXX: doesn't seem to work
# await trio.testing.wait_all_tasks_blocked(cushion=0)
debug_complete = Lock.no_remote_has_tty
debug_complete = _no_remote_has_tty
if (
(debug_complete and
not debug_complete.is_set())