-
-
Notifications
You must be signed in to change notification settings - Fork 638
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[pantsd] Address pantsd-runner hang on Linux and re-enable integration test. #4407
Conversation
@@ -112,7 +112,8 @@ impl Entry { | |||
let state = | |||
match self.node.clone() { | |||
EntryKey::Valid(n) => { | |||
let pool = context_factory.pool(); | |||
let pool_opt = context_factory.pool(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One line for this is probably fine.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
not according to the compiler.
error: borrowed value does not live long enough
--> src/rust/engine/src/graph.rs:115:88
|
115 | let pool = context_factory.pool().as_ref().expect("Uninitialized CpuPool!");
| ---------------------- temporary value created here ^ temporary value dropped here while still borrowed
...
122 | },
| - temporary value needs to live until here
|
= note: consider using a `let` binding to increase its lifetime
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ah. Yea, makes sense. Thanks.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks!
@@ -130,6 +130,11 @@ def run(self): | |||
"""Fork, daemonize and invoke self.post_fork_child() (via ProcessManager).""" | |||
self.daemonize(write_pid=False) | |||
|
|||
def pre_fork(self): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
IIRC, post_fork is called in two places... does this need to be as well?
207f8d6
to
65e5697
Compare
…n test. (pantsbuild#4407) ### Problem Currently, on Linux the first thin client call to the daemon can deadlock just after the pantsd->fork->pantsd-runner workflow. Connecting to the process with `gdb` reveals a deadlock in the following stack in the `post_fork` `drop` of the `CpuPool`: ``` #0 0x00007f63f04c31bd in __lll_lock_wait () from /lib64/libpthread.so.0 No symbol table info available. pantsbuild#1 0x00007f63f04c0ded in pthread_cond_signal@@GLIBC_2.3.2 () from /lib64/libpthread.so.0 No symbol table info available. pantsbuild#2 0x00007f63d3cfa438 in notify_one () at /buildslave/rust-buildbot/slave/stable-dist-rustc-linux/build/src/libstd/sys/unix/condvar.rs:52 No locals. pantsbuild#3 notify_one () at /buildslave/rust-buildbot/slave/stable-dist-rustc-linux/build/src/libstd/sys_common/condvar.rs:39 No locals. pantsbuild#4 notify_one () at /buildslave/rust-buildbot/slave/stable-dist-rustc-linux/build/src/libstd/sync/condvar.rs:208 No locals. pantsbuild#5 std::thread::{{impl}}::unpark () at /buildslave/rust-buildbot/slave/stable-dist-rustc-linux/build/src/libstd/thread/mod.rs:633 No locals. pantsbuild#6 0x00007f63d3c583d1 in crossbeam::sync::ms_queue::{{impl}}::push<futures_cpupool::Message> (self=<optimized out>, t=...) at /home/kwilson/.cache/pants/rust-toolchain/registry/src/github.com-1ecc6299db9ec823/crossbeam-0.2.10/src/sync/ms_queue.rs:178 guard = <optimized out> self = <optimized out> pantsbuild#7 0x00007f63d3c588ed in futures_cpupool::{{impl}}::drop (self=<optimized out>) at /home/kwilson/.cache/pants/rust-toolchain/git/checkouts/futures-rs-a4f11d094efefb0a/f7e6bc8/futures-cpupool/src/lib.rs:236 self = 0x37547a0 pantsbuild#8 0x00007f63d3be871c in engine::fs::{{impl}}::post_fork (self=0x3754778) at /home/kwilson/dev/pants/src/rust/engine/src/fs.rs:355 self = 0x3754778 pantsbuild#9 0x00007f63d3be10e4 in engine::context::{{impl}}::post_fork (self=0x37545b0) at /home/kwilson/dev/pants/src/rust/engine/src/context.rs:93 self = 0x37545b0 pantsbuild#10 0x00007f63d3c0de5a in {{closure}} (scheduler=<optimized out>) at /home/kwilson/dev/pants/src/rust/engine/src/lib.rs:275 scheduler = 0x3740580 pantsbuild#11 with_scheduler<closure,()> (scheduler_ptr=<optimized out>, f=...) at /home/kwilson/dev/pants/src/rust/engine/src/lib.rs:584 scheduler = 0x3740580 scheduler_ptr = 0x3740580 pantsbuild#12 engine::scheduler_post_fork (scheduler_ptr=0x3740580) at /home/kwilson/dev/pants/src/rust/engine/src/lib.rs:274 scheduler_ptr = 0x3740580 pantsbuild#13 0x00007f63d3c1be8c in _cffi_f_scheduler_post_fork (self=<optimized out>, arg0=0x35798f0) at src/cffi/native_engine.c:2234 _save = 0x34a65a0 x0 = 0x3740580 datasize = <optimized out> pantsbuild#14 0x00007f63f07b5a62 in PyEval_EvalFrameEx () from /lib64/libpython2.7.so.1.0 ``` This presents as a hang in the thin client, because the pailgun socket is left open in the pantsd-runner. ### Solution Add pre-fork hooks and tear down the `CpuPool` instances prior to forking and rebuilding them. ### Result Can no longer reproduce the hang.
https://python-poetry.org/history/#118---2021-08-19 Fixed an error with repository prioritization when specifying secondary repositories. (pantsbuild#4241) Fixed the detection of the system environment when the setting virtualenvs.create is deactivated. (pantsbuild#4330, pantsbuild#4407) Fixed the evaluation of relative path dependencies. (pantsbuild#4246) Fixed environment detection for Python 3.10 environments. (pantsbuild#4387) Fixed an error in the evaluation of in/not in markers (python-poetry/poetry-core#189)
https://python-poetry.org/history/#118---2021-08-19 Fixed an error with repository prioritization when specifying secondary repositories. (#4241) Fixed the detection of the system environment when the setting virtualenvs.create is deactivated. (#4330, #4407) Fixed the evaluation of relative path dependencies. (#4246) Fixed environment detection for Python 3.10 environments. (#4387) Fixed an error in the evaluation of in/not in markers (python-poetry/poetry-core#189)
Problem
Currently, on Linux the first thin client call to the daemon can deadlock just after the pantsd->fork->pantsd-runner workflow. Connecting to the process with
gdb
reveals a deadlock in the following stack in thepost_fork
drop
of theCpuPool
:This presents as a hang in the thin client, because the pailgun socket is left open in the pantsd-runner.
Solution
Add pre-fork hooks and tear down the
CpuPool
instances prior to forking and rebuilding them.Result
Can no longer reproduce the hang.