To handle both remote cancellation this adds `ContextCanceled.canceller:
tuple` the uid of the cancel requesting actor and is expected to be set
by the runtime when servicing any remote cancel request. This makes it
possible for `ContextCancelled` receivers to know whether "their actor
runtime" is the source of the cancellation.
Also add an explicit `RemoteActor.src_actor_uid` which better formalizes
the notion of "which remote actor" the error originated from.
Both of these new attrs are expected to be packed in the `.msgdata` when
the errors are loaded locally.
Previously we were leaking our (pdb++) override into the Python runtime
which would always result in a runtime error whenever `breakpoint()` is
called outside our runtime; after exit of the root actor . This
explicitly restores any previous hook override (detected during startup)
or deletes the hook and restores the environment if none existed prior.
Also adds a new WIP debugging example script to ensure breakpointing
works as normal after runtime close; this will be added to the test
suite.
Makes the broadcast test suite not hang xD, and is our expected default
behaviour. Also removes a ton of commented legacy cruft from before the
refactor to remove the `.receive()` recursion and fixes some typing.
Oh right, and in the case where there's only one subscriber left we warn
log about it since in theory we could actually entirely unwind the
bcaster back to the original underlying, though not sure if that's sane
or works for some use cases (like wanting to have some other subscriber
get added dynamically later).
Since one-way streaming can be accomplished by just *not* sending on one
side (and/or thus wrapping such usage in a more restrictive API), we
just drop the recv-only parent type. The only method different was
`MsgStream.send()`, now merged in. Further in usage of `.subscribe()`
we monkey patch the underlying stream's `.send()` onto the delivered
broadcast receiver so that subscriber tasks can two-way stream as though
using the stream directly.
This allows us to more definitively drop `tractor.open_stream_from()` in
the longer run if we so choose as well; note currently this will
potentially create an issue if a caller tries to `.send()` on such a one
way stream.
Driven by a bug found in `piker` where we'd get an inf recursion error
due to `BroadcastReceiver.receive()` being called when consumer tasks
are awoken but no value is ready to `.nowait_receive()`.
This new rework takes an approach closer to the interface and internals
of `trio.MemoryReceiveChannel` particularly in terms of,
- implementing a `BroadcastReceiver.receive_nowait()` and using it
within the async `.receive()`.
- failing over to an internal `._receive_from_underlying()` when the
`_nowait()` call raises `trio.WouldBlock`.
- adding `BroadcastState.statistics()` for debugging and testing
dropping recursion from `.receive()`.
We weren't doing this originally I *think* just because of the path
dependent nature of the way the code was developed (originally being
mega pedantic about one-way vs. bidirectional streams) but, it doesn't
seem like there's any issue just calling the stream's `.aclose()`; also
have the benefit of just being less code and logic checks B)
When backpressure is used and a feeder mem chan breaks during msg
delivery (usually because the IPC allocating task already terminated)
instead of raising we simply warn as we do for the non-backpressure
case.
Also, add a proper `Actor.is_arbiter` test inside `._invoke()` to avoid
doing an arbiter-registry lookup if the current actor **is** the
registrar.
The stdlib has all sorts of muckery with ignoring SIGINT in the
`Pdb._cmdloop()` but here we just override all that since we don't trust
their decisions about cancellation handling whatsoever. Adds
a `Lock.repl: MultiActorPdb` attr which is set by any task which
acquires root TTY lock indicating (via actor global state) that the
current actor is using the debugger REPL and can be expected to re-draw
the prompt on SIGINT. Further we mask out log messages from any actor
who also has the `shield_sigint_handler()` enabled to avoid logging
noise when debugging.
If we pack the nursery parent task's error into the `errors` table
directly in the handler, we don't need to specially handle packing that
same error into any exception group raised while handling sub-actor
cancellation; drops some ugly indentation ;)
Pretty sure this is the final touch to alleviate all our debug lock
headaches! Instead of trying to revert to the "last" handler (as `pdb`
does internally in the stdlib) we always just revert to the handler
`trio` registers during startup. Further this seems to allow cancelling
the root-side locking task if it's detected as stale IFF we only do this
when the root actor is in a "no more IPC peers" state.
Deatz:
- (always) set `._debug.Lock._trio_handler` as the `trio` version, not
some last used handler to make sure we're getting the ctrl-c handling
we want when not in debug mode.
- assign the trio handler in `open_root_actor()`
`._runtime._async_main()` to be sure it's applied in subactors as well
as the root.
- only do debug lock blocking and root-side-locking-task cancels when
a "no peers" condition is detected in the root actor: i.e. no IPC
channels are detected by the root meaning it's impossible any actor
has a sane lock-state ongoing for debug mode.
This is a lingering debugger locking race case we needed to handle:
- child crashes acquires TTY lock in root and attaches to `pdb`
- child IPC goes down such that all channels to the root are broken
/ non-functional.
- root is stuck thinking the child is still in debug even though it
can't be contacted and the child actor machinery hasn't been
cancelled by its parent.
- root get's stuck in deadlock with child since it won't send a cancel
request until the child is finished debugging, but the child can't
unlock the debugger bc IPC is down.
To avoid this scenario add debug lock blocking list via
`._debug.Lock._blocked: set[tuple]` which holds actor uids for any actor
that is detected by the root as having no transport channel connections
with said root (of which at least one should exist if this sub-actor at
some point acquired the debug lock). The root consequently checks this
list for any actor that tries to (re)acquire the lock and blocks with
a `ContextCancelled`. When a debug condition is tested in
`._runtime._invoke` the context's `._enter_debugger_on_cancel` which
is set to `False` if the actor is on the block list in which case the
post-mortem entry is skipped.
Further this adds a root-locking-task side cancel scope to
`Lock._root_local_task_cs_in_debug` which can be cancelled by the root
runtime when a stale lock is detected after all IPC channels for the
actor have been torn down. NOTE: right now we're NOT doing this since it
seems to cause test failures likely due because it may cause pre-mature
cancellation and maybe needs a bit more experimenting?
In the case of a callee-side context cancelling itself it can be handy
to let the caller-side task know (even if through logging) that the
cancel was due to some known reason. Make `.cancel()` accept such
a message on the callee side and have it included in the
`._runtime._invoke()` raised `ContextCancelled` emission.
Also add a `Context._trigger_debugger_on_cancel: bool` flag which can be
set to `False` to avoid the debugger post-mortem crash mode from
engaging on cross-context tasks which cancel themselves for a known
reason (as is needed for blocked tasks in the debug TTY-lock machinery).
Turns out the lifetime mgmt of separate nurseries per delegate manager
is tricky; a new nursery can't be naively allocated on cache-misses since
it may get closed by some early terminating task instead of by the "last
using" consumer task. In theory if we allocate using the same logic as
that used for the last-task-triggers-exit then this should work?
For now just go back to a single global nursery per `_Cache` which still
avoids use of the internal actor service nursery.
Instead of sticking all `trionics.maybe_open_context()` tasks inside the
actor's (root) service nursery, open a unique one per manager function
instance (id).
Further, accept a callable for the `key` such that a user can have
more flexible control on the caching logic and move the
`maybe_open_nursery()` helper out of the portal mod and into this
trionics "managers" module.
Instead of the logic branching create a table `._spawn._methods`
which is used to lookup the desired backend framework (in this case
still only one of `multiprocessing` or `trio`) and make the top level
`.new_proc()` do the lookup and any common logic. Use a `typing.Literal`
to define the lookup table's key set.
Repair and ignore a bunch of type-annot related stuff todo with `mypy`
updates and backend-specific process typing.
The common logic to both remove our custom SIGINT handler as well
as signal the actor global event that pdb is complete. Call this
whenever we exit a post mortem call and thus any time some rpc task
get's debugged inside `._actor._invoke()`.
Further, we have to manually print the REPL prompt on 3.9 for some wack
reason, so stick a version guard in the sigint handler for that..
When in an uncertain teardown state and in debug mode a context can be
popped from actor runtime before a child finished debugging (the case
when the parent is tearing down but the child hasn't closed/completed
its tty lock IPC exit phase) and the child sends the "stop" message to
unlock the debugger but it's ignored bc the parent has already dropped
the ctx. Instead we call `._debug.maybe_wait_for_deugger()` before these
context removals to avoid the root getting stuck thinking the lock was
never released.
Further, add special `Actor._cancel_task()` handling code inside
`_invoke()` which continues to execute the method despite the IPC
channel to the caller being broken and thus avoiding potential hangs due
to a target (child) actor task remaining alive.
Ensure that even when `pdb` resumption methods are called during a crash
where `trio`'s runtime has already terminated (eg. `Event.set()` will
raise) we always revert our sigint handler to the original. Further
inside the handler if we hit a case where a child is in debug and
(thinks it) has the global pdb lock, if it has no IPC connection to
a parent, simply presume tty sync-coordination is now lost and cancel
the child immediately.
A hopefully significant fix here is to always avoid suppressing a SIGINT
when the root actor can not detect an active IPC connections (via
a connected channel) to the supposed debug lock holding actor. In that
case it is most likely that the actor has either terminated or has lost
its connection for debugger control and there is no way the root can
verify the lock is in use; thus we choose to allow KBI cancellation.
Drop the (by comment) `try`-`finally` block in
`_hijoack_stdin_for_child()` around the `_acquire_debug_lock()` call
since all that logic should now be handled internal to that locking
manager. Try to catch a weird error around the `.do_longlist()` method
call that seems to sometimes break on py3.10 and latest `pdbpp`.
The method now returns a `bool` which flags whether the transport died
to the caller and allows for reporting a disconnect in the
channel-transport handler task. This is something a user will normally
want to know about on the caller side especially after seeing
a traceback from the peer (if in tree) on console.
There's no point in sending a cancel message to the remote linked task
and especially no reason to block waiting on a result from that task if
the transport layer is detected to be disconnected. We expect that the
transport shouldn't go down at the layer of the message loop
(reconnection logic should be handled in the transport layer itself) so
if we detect the channel is not connected we don't bother requesting
cancels nor waiting on a final result message.
Why?
- if the connection goes down in error the caller side won't have a way
to know "how long" it should block to wait for a cancel ack or result
and causes a potential hang that may require an additional ctrl-c from
the user especially if using the debugger or if the traceback is not
seen on console.
- obviously there's no point in waiting for messages when there's no
transport to deliver them XD
Further, add some more detailed cancel logging detailing the task and
actor ids.
There's a bug that's triggered in the stdlib without latest `pdb++`
installed; add a note for that.
Further inside `wait_for_parent_stdin_hijack()` don't `.started()` until
the interactor stream has been opened to avoid races when debugging this
`._debug.py` module (at the least) since we usually don't want the
spawning (parent) task to resume until we know for sure the tty lock has
been acquired. Also, drop the random checkpoint we had inside
`_breakpoint()`, not sure it was actually adding anything useful since
we're (mostly) carefully shielded throughout this func.
None of it worked (you still will see `.__exit__()` frames on debugger
entry - you'd think this would have been solved by now but, shrug) so
instead wrap the debugger entry-point in a `try:` and put the SIGINT
handler restoration inside `MultiActorPdb` teardown hooks.
This seems to restore the UX as it was prior but with also giving the
desired SIGINT override handler behaviour.
Using either of `@pdb.hideframe` or `__tracebackhide__` on stdlib
methods doesn't seem to work either.. This all seems to have something
to do with async generator usage I think ?
This gets very close to avoiding any possible hangs to do with tty
locking and SIGINT handling minus a special case that will be detailed
below.
Summary of implementation changes:
- convert `_mk_pdb()` -> `with _open_pdb() as pdb:` which implicitly
handles the `bdb.BdbQuit` case such that debugger teardown hooks are
always called.
- rename the handler to `shield_sigint()` and handle a variety of new
cases:
* the root is in debug but hasn't been cancelled -> call
`Actor.cancel_soon()`
* the root is in debug but *has* been called (`Actor.cancel_soon()`
already called) -> raise KBI
* a child is in debug *and* has a task locking the debugger -> ignore
SIGINT in child *and* the root actor.
- if the debugger instance is provided to the handler at acquire time,
on SIGINT handling completion re-print the last pdb++ REPL output so
that the user realizes they are still actively in debug.
- ignore the unlock case where a race condition of "no task" holding the
lock causes the `RuntimeError` normally associated with the "wrong
task" doing so (not sure if this is a `trio` bug?).
- change debug logs to runtime level.
Unhandled case(s):
- a child is maybe in debug mode but does not itself have any task using
the debugger.
* ToDo: we need a way to decide what to do with
"intermediate" child actors who themselves either are not in
`debug_mode=True` but have children who *are* such that a SIGINT
won't cause cancellation of that child-as-parent-of-another-child
**iff** any of their children are in in debug mode.
This fixes an previously undetected bug where if an
`.open_channel_from()` spawned task errored the error would not be
propagated to the `trio` side and instead would fail silently with
a console log error. What was most odd is that it only seems easy to
trigger when you put a slight task sleep before the error is raised
(:eyeroll:). This patch adds a few things to address this and just in
general improve iter-task lifetime syncing:
- add `LinkedTaskChannel._trio_exited: bool` a flag set from the `trio`
side when the channel block exits.
- add a `wait_on_aio_task: bool` flag to `translate_aio_errors` which
toggles whether to wait the `asyncio` task termination event on exit.
- cancel the `asyncio` task if the trio side has ended, when
`._trio_exited == True`.
- always close the `trio` mem channel when the task exits such that
the `asyncio` side can error on any next `.send()` call.
Sometimes it's handy to just have a non-`Portal` yielding way
to figure out if a "service" actor is up, so add this discovery
helper for that. We'll prolly just leave it undocumented for
now until we figure out a longer-term/better discovery system.
When an `asyncio` side task errors or is cancelled we now explicitly
report the traceback and task name if possible as well as the source
reason for the error (some come from the `trio` side).
Further, properly set any `trio` side exception (after unwrapping it
from the `outcome.Error`) on the future that runs the `trio` guest run.
Python 3.9's new object resolver + a `str` is much simpler then mucking
with tuples (and easier to serialize). Include a `.to_tuple()` formatter
since we still are passing the module namespace and function name
separately inside the runtime's message format but in theory we might be
able to simplify this depending on how we would change the support for
`enable_modules:list[str]` in the spawn API.
Thanks to @Fuyukai for pointing `resolve_name()` which I didn't know
about before!
Adjust the `soft_wait()` strategy to avoid sending needless cancel
requests if it is known that a child process is already terminated or
does so before the cancel request times out. This should be no slower
and should avoid needless waits on either closure-in-progress or already
closed channels.
Basic strategy is,
- request child actor to cancel
- if process termination is detected, cancel the cancel
- if the process is still alive after a cancel request timeout warn the
user and yield back to the hard reap handling
Better encapsulate all the mem-chan, Queue, sync-primitives inside our
linked task channel in order to avoid `mypy`'s complaints about monkey
patching. This also sets footing for adding an `asyncio`-side channel
API that can be used more like this `trio`-side API.
For whatever reason `trio` seems to be swallowing this exception when
raised in the `trio` task so instead wrap it in our own non-base
exception type: `AsyncioCancelled` and raise that when the `asyncio`
task cancels itself internally using `raise <err> from <src_err>` style.
Further don't bother cancelling the `trio` task (via cancel scope)
since we we can just use the recv mem chan closure error as a signal
and explicitly lookup any set asyncio error.
Wraps the pairs of underlying `trio` mem chans and the `asyncio.Queue`
with this new composite which will be delivered from `open_channel_from()`.
This allows for both sending and receiving values from the `asyncio`
task (2 way msg passing) as well controls for cancelling or waiting on
the task.
Factor `asyncio` translation and re-raising logic into a new closure
which is run on both `trio` side error handling as well as on normal
termination to avoid missing `asyncio` errors even when `trio` task
cancellation is handled first.
Only close the `trio` mem chans on `trio` task termination *iff*
the task was spawned using `open_channel_from()`:
- on `open_channel_from()` exit, mem chan closure is the desired semantic
- on `run_task()` we normally only return a single value or error and
if the channel is closed before the error is raised we may propagate
a `trio.EndOfChannel` instead of the desired underlying `asyncio`
task's error
Pull the common `asyncio` -> `trio` error translation logic into
a common context manager and don't expect a final result to be captured
when using `open_channel_from()` since it's a manager interface and it
would be clunky to try and deliver some "final result" after exit.
Close the mem chan before cancelling the `trio` task in order to ensure
we retrieve whatever error is shuttled from `asyncio` before the channel
read is potentially cancelled (previously a race?).
Handle `asyncio.CancelledError` specially such that we raise it directly
(instead of `raise aio_cancelled from other_err`) since it *is* the
source error in the case where the cancellation is `asyncio` internal.
Clearly this wasn't developed against a task that spawned just an async
func in `asyncio`.. Fix all that and remove a bunch of unnecessary func
layers. Add provisional support for the target receiving the `to_trio`
and `from_trio` channels and for the @tractor.stream marker.
This should mostly maintain top level SC principles for any task spawned
using `tractor.to_asyncio.run()`. When the `asyncio` task completes make
sure to cancel the pertaining `trio` cancel scope and raise any error
that may have resulted. This interface uses `trio`'s "guest-mode" to run
`asyncio` loop using a special entrypoint which is handed to Python
during process spawn.
If the one side of an inter-actor context cancels the other then that
side should always expect back a `ContextCancelled` message. However we
should not set this error in this case (where the cancel request was
sent and a `ContextCancelled` msg was received back) since it may
override some other error that caused the cancellation request to be
sent out in the first place. As an example when a context opens another
context to a peer and some error happens which causes the second peer
context to be cancelled but we want to propagate the original error.
Fixes the issue found in https://github.com/pikers/piker/issues/244
After more extensive testing I realized that keying on the context
manager *instance id* isn't going to work since each entering task is
going to create a unique key XD
Instead pass the manager function as `acm_func` and optionally allow
keying the resource on the passed `kwargs` (if hashable) or the
`key:str`. Further, pass the key to the enterer task and avoid
a separate keying scheme for the manager versus the value it delivers.
Don't bother with checking and releasing the lock in `finally:` block,
it should be an error if it's still locked.
Without this wakeup you can have tasks which re-enter `.receive()`
and get stuck waiting on the wakeup event indefinitely. Whenever
a ``trio.EndOfChannel`` arrives we want to make sure all consumers
at least know about it and don't block. This previous behaviour was
basically a bug.
Add some state flags for tracking if the broadcaster was either
cancelled or terminated via EOC mostly for testing and debugging
purposes though this info might be useful if we decide to offer
a `.statistics()` like API in the future.
This commit obviously denotes a re-license of all applicable parts of
the code base. Acknowledgement of this change was completed in #274 by
the majority of the current set of contributors. From here henceforth
all changes will be AGPL licensed and distributed. This is purely an
effort to maintain the same copy-left policy whilst closing the
(perceived) SaaS loophole the GPL allows for. It is merely for this
loophole: to avoid code hiding by any potential "network providers" who
are attempting to use the project to make a profit without either
compensating the authors or re-distributing their changes.
I thought quite a bit about this change and can't see a reason not to
close the SaaS loophole in our current license. We still are (hard)
copy-left and I plan to keep the code base this way for a couple
reasons:
- The code base produces income/profit through parent projects and is
demonstrably of high value.
- I believe firms should not get free lunch for the sake of
"contributions from their employees" or "usage as a service" which
I have found to be a dubious argument at best.
- If a firm who intends to profit from the code base wants to use it
they can propose a secondary commercial license to purchase with the
proceeds going to the project's authors under some form of well
defined contract.
- Many successful projects like Qt use this model; I see no reason it
can't work in this case until such a time as the authors feel it
should be loosened.
There has been detailed discussion in #103 on licensing alternatives.
The main point of this AGPL change is to protect the code base for the
time being from exploitation while it grows and as we move into the next
phase of development which will include extension into the multi-host
distributed software space.
If we make it too fast a nursery with debug mode children can cancel
too fast and causes some test failures. It's likely not a huge deal
anyway since the purpose of this poll/check is for human interaction
and the current delay isn't really that noticeable.
Decrease log levels in the debug module to avoid console noise when in
use. Toss in some more detailed comments around the new debugger lock
points.
A context method handling all this logic makes the most sense since it
contains all the state related to whether the error should be raised in
a nursery scope or is expected to be raised by a consumer task which
reads and processes the msg directly (via a `Portal` API call). This
also makes it easy to always process remote errors even when there is no
(stream) overrun condition.
A context stream overrun should normally never take place since if
a stream is opened (via ``Context.open_stream()``) backpressure is
applied on the message buffer (unless explicitly disabled by the
``backpressure=False`` flag) such that an overrun on the receiving task
should result in blocking the (remote) sender task (eventually depending
on the underlying ``MsgStream`` transport).
Here we add a special error message that reports if one side never
opened a stream and let's the user know in the overrun error message
that they may be trying to push messages to a task that isn't ready to
receive them.
Further fixes / details:
- pop any `Context` at the end of any `_invoke()` task that creates
one and registers with the runtime.
- ignore but warn about messages received for a context that either
no longer exists or is unknown (guarding against crashes by malicious
packets in the latter case)
Keeping it disabled on context open will help with detecting any stream
connection which was never opened on one side of the task pair. In that
case we can report that there was an overrun **and** a stream wasn't
opened versus if the stream is explicitly configured not to use bp then
we throw the standard overflow.
Use `trio.Nursery._closed` to detect "closure" XD since it seems to be
the most reliable way to determine if a spawn call will trigger
a runtime error.
Half of portal API usage requires a 1 message response (`.run()`,
`.run_in_actor()`) and the streaming APIs should probably be explicitly
enabled for backpressure if desired by the user. This makes more sense
in (psuedo) realtime systems where it's better to notify on a block then
freeze without notice. Make this default behaviour with a new error to
be raised: `tractor._exceptions.StreamOverrun` when a sender overruns
a stream by the default size (2**6 for now). The old behavior can be
enabled with `Context.open_stream(backpressure=True)` but now with
warning log messages when there are overruns.
Add task-linked-context error propagation using a "nursery raising"
technique such that if either end of context linked pair of tasks
errors, that error can be relayed to other side and raised as a form of
interrupt at the receiving task's next `trio` checkpoint. This enables
reliable error relay without expecting the (error) receiving task to
call an API which would raise the remote exception (which it might never
currently if using `tractor.MsgStream` APIs).
Further internal implementation details:
- define the default msg buffer size as `Actor.msg_buffer_size`
- expose a `msg_buffer_size: int` kwarg from `Actor.get_context()`
- maybe raise aforementioned context errors using
`Context._maybe_error_from_remote_msg()` inside `Actor._push_result()`
- support optional backpressure on a stream when pushing messages
in `Actor._push_result()`
- in `_invote()` handle multierrors raised from a `@tractor.context`
entrypoint as being potentially caused by a relayed error from the
remote caller task, if `Context._error` has been set then raise that
error inside the `RemoteActorError` that will be relayed back to that
caller more or less proxying through the source side error back to its
origin.
In preparation for supporting both backpressure detection (through an
optional error) as well as control over the msg channel buffer size, add
internal configuration flags for both to contexts. Also adjust
`Context._err_on_from_remote_msg()` -> `._maybe..` such that it can be
called and will only raise if a scope nursery has been set. Add
a `Context._error` for stashing the remote task's error that may be
delivered in an `'error'` message.
This more formally declares the runtime's remote task startingn API
and uses it throughout all the dependent `Portal` API methods.
Allows dropping `Portal._submit()` and simplifying `.run_in_actor()`
style result waiting to be delegated to the context APIs at remote
task `return` response time. We now also track the remote entrypoint
"type` as `Context._remote_func_type`.
Instead of tracking feeder mem chans per RPC dialog, store `Context`
instances which (now) hold refs to the underlying RPC-task feeder chans
and track them inside a `Actor._contexts` map. This begins a transition
to making the "context" idea the primitive abstraction for representing
messaging dialogs between tasks in different memory domains (i.e.
usually separate processes).
A slew of changes made this possible:
- change `Actor.get_memchans()` -> `.get_context()`.
- Add new `Context._send_chan` and `._recv_chan` vars.
- implicitly create a new context on every `Actor.send_cmd()` call.
- use the context created by `.send_cmd()` in `Portal.open_context()`
instead of manually creating one.
- call `Actor.get_context()` inside tasks run from `._invoke()`
such that feeder chans are implicitly created for callee tasks
thus fixing the bug #265.
NB: We might change some of the internal semantics to do with *when* the
feeder chans are actually created to denote whether or not a far end
task is actually *read to receive* messages. For example, in the cases
where it **never** will be ready to receive messages (one-way streaming,
a context that never opens a stream, etc.) we will likely want some kind
of error or at least warning to the caller that messages can't be sent
(yet).
Previously we were ignoring a race where the callee an opened task
context could enter `Context.open_stream()` before calling `.started().
Disallow this as well as calling `.started()` more then once.
We don't need to any more presuming you get ideal remote cancellation
conditions where the remote actor should teardown and kill the streams
from its end.
On msg loop termination we now check and see if a channel is associated
with a child-actor registered in some local task's nursery. If so, we
attempt to wait on channel closure initiated from the child side (by
draining the underlying msg stream) so as to avoid closing it too early
resulting in the child not relaying its termination status response. This
means we now support the ideal case in 2-general's where we get back the
ack to the closure request instead of just ignoring it and timing out XD
The main implementation detail is that when `Portal.cancel_actor()`
remotely calls `Actor.cancel()` we actually wait for the RPC response
from that request before allowing the channel shutdown sequence to
engage. The new msg stream draining support enables this.
Also, factor child-to-parent error propagation logic into a helper func
and improve some docs (yeah yeah y'all don't like the ''', i don't
care - it makes my eyes not hurt).
Use a `trio.Event` to enable nursery closure detection such that core
runtime tasks can be notified when a local nursery exits and allow
shutdown protocols to operate without close-before-terminate issues
(such as IPC channel closure during remote peer cancellation).
Enables "draining" the last set of messages after a channel/stream has
been terminated mostly for the purposes of receiving a final ACK to
a remote cancel command. Also, add an internal `Channel._cancel_called`
flag which can be set by `Portal.cancel_actor()`.
It's definitely possible to have a nursery spawn task be cancelled
before a `trio.Process` handle is ever returned; we now handle this
case as a cancelled-during-spawn scenario. Zombie collection logic
also is bypassed in this case.
Thanks to @richardsheridan for pointing out the limitations of using
*any* kind of value as the result-cached-flag and how it might cause
problems for anyone returning pickled blob-data. This changes the
`Portal` internal result value tracking to stash the full message from
which the value can be retrieved by any `Portal.result()` caller.
The internal change is that `Portal._return_once()` now returns a tuple
of the message *and* its value.
Fixes the issue where if the main remote task returns `None`,
`Portal.result()` would erroneously wait again on the underlying feeder
mem chan since `None` was being used as the cache flag. Instead set the
flag as the channel uid and consider the result collected when set to
anything else (since it would be odd to return that value from a remote
task when you already can read it as part of portal/channel apis).
The api we've made here is actually closer to `asyncio.gather()` but
with opening async context managers instead of funcs. Use another event
to allow for graceful teardown of children on non-cancellation exits
and add a doc string.
Since it seems we're building out more and more higher level primitives
in order to support certain parallel style actor trees and messaging
patterns (eg. task broadcast channels), we might as well start a new
sub-package for purely `trio` constructions. We hereby dub this
the realm of `trionics` (like electronics but for trios instead of
electrons).
To kick things off, add an `async_enter_all()` concurrent
exit-stack-like context manager API which will concurrently spawn
a sequence of provided async context managers and deliver their ordered
results but with proper support for `trio` cancellation semantics.
The stdlib's `AsyncExitStack` is not compatible with nurseries not
`trio` tasks (which are cancelled) since as task will be suspended on
the stack after push and does not ever hit a checkpoint until the stack
is closed.
This is actually surprisingly easy to grok having gone through a lot of
pain understanding edge cases in the zombie lord dev branch. Basically
we just need to make sure actors are managed in a 2 step reap sequence.
In the "soft" reap phase we wait for the process to terminate on its own
concurrently with (maybe) waiting for its portal's final result (if it's
a `.run_in_actor()`). If this path is cancelled or errors, then we do
a "hard" reap where we timeout and send a signal to the proc to
terminate immediately. The only last remaining trick is to tie in the
root-is-debugger-aware logic to yet again avoid tty clobbers.
As for `Actor.cancel()` requests, do the same for
`Actor._cancel_task()` but use `_invoke()` to ensure
correct msg transactions with caller. Don't cancel task
cancels on a cancel-all-tasks operation in attempt at
more determinism.
Now that we're on our way to a (somewhat) serious beta release I think
it's about time to start de-noising the logging emissions. Since we're
trying out this approach of "stack layer oriented" log levels, I figured
this is a good time to move most of the "warnings" to what they should
be: cancellation monitoring status messages. The level is set to 16
which is just above our "runtime" level but just below the traditional
"info" level. I think this will be a decent approach since usually if
you're confused about why your `tractor` app is behaving unlike you
expect, it's 90% of the time going to be to do with cancellation or
error propagation. This this setup a user can specify the 'cancel' level
and see all the msgs pertaining to both actor and task-in-actor
cancellation mechanics.
The stdlib's `logging.LoggingAdapter` doesn't currently pass through
`stacklevel: int` down to its wrapped logger instance. Hack it here
and get our msgs looking like they would if using a built-in level.
In an effort to have some kind of more formal interface around the
transport layer, add a `MsgTransport` protocol type and use with
the channel composition of message streams. Start a little "key map"
of `(<codec>, <protocol>)` to `MsgTransport` types which can be
dynamically loaded. Add a `Channel.from_stream()` constructor thus
cleaning up the mangled logic that was in the constructor based on
inputs. Drop all the "auto reconnect" channel logic for now since
nothing is using it (internally) and it's likely it will need rework
once we bring in a protocol besides TCP.
`msgspec` sends python lists over the wire
(https://github.com/jcrist/msgspec/issues/30) which is fine and dandy
but we use them as lookup keys so we need to be sure we tuple-cast
first.
This change some super old (and bad) code from the project's very early
days. For some redic reason i must have thought masking `trio`'s
internal stream / transport errors and a TCP EOF as `StopAsyncIteration`
somehow a good idea. The reality is you probably
want to know the difference between an unexpected transport error
and a simple EOF lol. This begins to resolve that by adding our own
special `TransportClosed` error to signal the "graceful" termination of
a channel's underlying transport. Oh, and this builds on the `msgspec`
integration which helped shed light on the core issues here B)
Add a `tractor._ipc.MsgspecStream` type which can be swapped in for
`msgspec` serialization transparently. A small msg-length-prefix framing
is implemented as part of the type and we use
`tricycle.BufferedReceieveStream` to handle buffering logic for the
underlying transport.
Notes:
- had to force cast a few more list -> tuple spots due to no native
`tuple`decode-by-default in `msgspec`: https://github.com/jcrist/msgspec/issues/30
- the framing can be understood by this protobuf walkthrough:
https://eli.thegreenplace.net/2011/08/02/length-prefix-framing-for-protocol-buffers
- `tricycle` becomes a new dependency
Can only really use an encoder currently since there is no streaming api
in `msgspec` as of currently. See jcrist/msgspec#27.
Not sure if any encoding speedups are currently noticeable especially
without any validation going on yet XD.
First experiments toward #196
- drop `shield` input to `MsgStream`
- check for cancel called prior to loading the feeder mem chan
in `Context.open_stream()`
- warn on a timeout when trying to cancel a remote task from
`Context.cancel()`
- drop noop endofchannel handler block
The `collections.deque` takes care of array length truncation of values
for us implicitly but in the future we'll likely want this value exposed
to alternate array implementations. This patch is to provide for that as
well as make `mypy` happy since the `dequeu.maxlen` can also be `None`.
Get rid of all the (requirements for) clones of the underlying
receivable. We can just use a uuid generated key for each instance
(thinking now this can probably just be `id(self)`). I'm fully convinced
now that channel cloning is only a source of confusion and anti-patterns
when we already have nurseries to define resource lifetimes. There is no
benefit in particular when you allocate subscriptions using a context
manager (not sure why `trio.open_memory_channel()` doesn't enforce
this).
Further refinements:
- add a `._closed` state that will error the receiver on reuse
- drop module script section; it's been moved to a real test
- call the "receiver" duck-type stub a new name
This allows for wrapping an existing stream by re-assigning its receive
method to the allocated broadcaster's `.receive()` so as to avoid
expecting any original consumer(s) of the stream to now know about the
broadcaster; this instead mutates the stream to delegate to the new
receive call behind the scenes any time `.subscribe()` is called.
Add a `typing.Protocol` for so called "cloneable channels" until we
decide/figure out a better keying system for each subscription and
mask all undesired typing failures.
Add `ReceiveMsgStream.subscribe()` which allows allocating a broadcast
receiver around the stream for use by multiple actor-local consumer
tasks. Entering this context manager idempotently mutates the stream's
receive machinery which for now can not be undone. Move `.clone()` to
the receive stream type.
Resolves#204
For every set of broadcast receivers which pull from the same producer,
we need a singleton state for all of,
- subscriptions
- the sender ready event
- the queue
Add a `BroadcastState` dataclass for this and pass it to all
subscriptions. This makes the design much more like the built-in memory
channels which do something very similar with `MemoryChannelState`.
Use a `filter()` on the subs list in the sequence update step, plus some
other commented approaches we can try for speed.
Using the current task as a subscription key fails horribly as soon as
you hand off new subscription receiver to another task you've spawned..
Instead use the underlying ``trio.abc.ReceiveChannel.clone()`` as a key
(so i guess we're assuming cloning is supported by the underlying?)
which makes this all work just like default mem chans. As a bonus, now
we can just close the underlying rx (which may be a clone) on
`.aclose()` and everything should just work in terms of the underlying
channels lifetime (i think?).
Change `.subscribe()` to be async since the receive channel type
interface only expects `.aclose()` and it actually ends up being
nicer for 3.9+ style `async with` parentheses style anyway.
Tyler Goodlet
goodboy
overclockworked64