Fold `CancelPendingDelayedTasks()` into `UnregisterIsolate()` and
make implementing it optional.
It makes sense for these two operations to happen at the same time,
so it is sufficient to provide a single operation instead of two
separate ones.
PR-URL: https://github.com/nodejs/node/pull/30034
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Shelley Vohr <codebytere@gmail.com>
- Since this was introduced, V8 has effectively started requiring
that the platform knows of the `Isolate*` before we (or an embedder)
create our `IsolateData` structure; therefore, (un)registering it
from the `IsolateData` constructor/destructor doesn’t make much
sense anymore.
- Instead, we can require that the register/unregister functions
are only called once, simplifying the implementation a bit.
- Add a callback that we can use to know when the platform has
cleaned up its resources associated with a given `Isolate`.
In particular, this means that in the Worker code, we don’t need
to rely on what are essentially guesses about the number of event
loop turns that we need in order to have everything cleaned up.
PR-URL: https://github.com/nodejs/node/pull/26384
Reviewed-By: Daniel Bevenius <daniel.bevenius@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com>
Node first calls `Isolate::Dispose`, then
`NodePlatform::UnregisterIsolate`.
This again calls `PerIsolatePlatformData::Shutdown`, which (before this
patch) called `FlushForegroundTasksInternal`, which might call
`RunForegroundTask` if it finds foreground tasks to be executed. This
will fail however, since `Isolate::GetCurrent` was already reset during
`Isolate::Dispose`.
Hence remove the check to `FlushForegroundTasksInternal` and add checks
instead that no more foreground tasks are scheduled.
Refs: https://github.com/v8/node/pull/86
PR-URL: https://github.com/nodejs/node/pull/25653
Reviewed-By: Gus Caplan <me@gus.host>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Metadata trace-events should be held in secondary storage so that they
can be periodically reemitted. This change establishes the secondary
storage and ensures that events are reemitted on each flush.
PR-URL: https://github.com/nodejs/node/pull/20900
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
For safer shutdown, we should destroy the platform – and background
threads - before the tracing infrastructure is destroyed. This change
fixes the relative order of NodePlatform disposition and the tracing
agent shutting down. This matches the nesting order for startup.
Make the tracing agent own the tracing controller instead of platform
to match the above.
Fixes: https://github.com/nodejs/node/issues/22865
PR-URL: https://github.com/nodejs/node/pull/22938
Reviewed-By: Eugene Ostroukhov <eostroukhov@google.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
We use the `Isolate*` pointer as the sole identifier
for a V8 Isolate. In some environments (e.g. multi-threaded),
Isolates may be destroyed and new ones created; then, it
may happen that the memory that was previously used for
one `Isolate` can be re-used for another `Isolate`
after the first one has been disposed of.
This check is a little guard against accidentally
re-using the same per-Isolate platform data structure
in such cases, i.e. making sure (to the degree to which
that is possible) that the old `Isolate*` has been properly
unregistered before one at the same memory address is added.
(It’s not 100 % foolproof because the `uv_loop_t*`
pointer value could theoretically be the same as well.)
PR-URL: https://github.com/nodejs/node/pull/21156
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
V8 uses a thread pool provided by the host to schedule background tasks
for concurrent GC and compiation. Emit trace events to identify the
background threads. Ensure that the tracing infrastructure is started
before the thread pool is initialized.
PR-URL: https://github.com/nodejs/node/pull/20823
Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Use std::unique_ptr for delayed tasks in the scheduled
delayed tasks vector. This makes it clear that the vector
has ownership of the delayed tasks and is responsible for
deleting them.
Use a custom deleter for the pointers because libuv
needs to close the handle and then delete the data. Provide
the handle when creating the pointer instead of invoking the
special delete action everytime an element is removed from the vector.
PR-URL: https://github.com/nodejs/node/pull/17083
Reviewed-By: Anna Henningsen <anna@addaleax.net>
Worker threads need an event loop without active libuv handles in
order to shut down. One source of handles that was previously
not accounted for were delayed V8 tasks; these create timers
that would be standing in the way of clearing the event loop.
To solve this, keep track of the scheduled tasks in a list
and close their timer handles before the corresponding isolate/loop
is removed from the platform.
It is not clear from the V8 documentation what the expectation is
with respect to pending background tasks at the end of the
isolate lifetime; however, an alternative approach of executing
these scheduled tasks when flushing them led to an infinite loop
of tasks scheduling each other; so it seems safe to assume that
the behaviour implemented in this patch is at least acceptable.
Original-PR-URL: https://github.com/ayojs/ayo/pull/120
Original-Reviewed-By: Stephen Belanger <admin@stephenbelanger.com>
PR-URL: https://github.com/nodejs/node/pull/16700
Reviewed-By: James M Snell <jasnell@gmail.com>
V8 platform tasks may schedule other tasks (both background and
foreground), and may perform asynchronous operations like
resolving Promises.
To address that:
- Run the task queue drain call inside a callback scope.
This makes sure asynchronous operations inside it, like
resolving promises, lead to the microtask queue and any
subsequent operations not being silently forgotten.
- Move the task queue drain call before `EmitBeforeExit()`
and only run `EmitBeforeExit()` if there is no new event
loop work.
- Account for possible new foreground tasks scheduled by
background tasks in `DrainBackgroundTasks()`.
PR-URL: https://github.com/nodejs/node/pull/15428
Reviewed-By: Franziska Hinkelmann <franziska.hinkelmann@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Matthew Loring <mattloring@google.com>
Marcel Laverdet