You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@mesos.apache.org by Neil Conway <ne...@gmail.com> on 2016/04/11 05:49:00 UTC

Review Request 45999: Ensure GC is terminated last during libprocess shutdown.

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45999/
-----------------------------------------------------------

Review request for mesos and Joris Van Remoortere.


Bugs: MESOS-5144
    https://issues.apache.org/jira/browse/MESOS-5144


Repository: mesos


Description
-------

Without this change, `finalize()` terminates processes in the order
that they happen to be found when iterating over the `processes`
map. That means that if the GarbageCollector process is terminated
while any GC-managed processes are still running, those processes
will not be GC'd (i.e., they will be leaked).

Fix this by skipping the garbage collector process when iterating
over `processes` in `finalize()`, and then only terminating it after
all other processes have been terminated.


Diffs
-----

  3rdparty/libprocess/src/process.cpp 5e9dcfdc52f3a8223bc43af149b8e1f5dbdf5b0a 

Diff: https://reviews.apache.org/r/45999/diff/


Testing
-------

make check

Confirmed reduction of leaked memory via ASAN on Linux/amd64.


Thanks,

Neil Conway


Re: Review Request 45999: Ensure GC is terminated last during libprocess shutdown.

Posted by Mesos ReviewBot <re...@mesos.apache.org>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45999/#review128085
-----------------------------------------------------------



Patch looks great!

Reviews applied: [45991, 45995, 45996, 45999]

Passed command: export OS='ubuntu:14.04' CONFIGURATION='--verbose' COMPILER='gcc' ENVIRONMENT='GLOG_v=1 MESOS_VERBOSE=1'; ./support/docker_build.sh

- Mesos ReviewBot


On April 11, 2016, 3:48 a.m., Neil Conway wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/45999/
> -----------------------------------------------------------
> 
> (Updated April 11, 2016, 3:48 a.m.)
> 
> 
> Review request for mesos and Joris Van Remoortere.
> 
> 
> Bugs: MESOS-5144
>     https://issues.apache.org/jira/browse/MESOS-5144
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Without this change, `finalize()` terminates processes in the order
> that they happen to be found when iterating over the `processes`
> map. That means that if the GarbageCollector process is terminated
> while any GC-managed processes are still running, those processes
> will not be GC'd (i.e., they will be leaked).
> 
> Fix this by skipping the garbage collector process when iterating
> over `processes` in `finalize()`, and then only terminating it after
> all other processes have been terminated.
> 
> 
> Diffs
> -----
> 
>   3rdparty/libprocess/src/process.cpp 5e9dcfdc52f3a8223bc43af149b8e1f5dbdf5b0a 
> 
> Diff: https://reviews.apache.org/r/45999/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> Confirmed reduction of leaked memory via ASAN on Linux/amd64.
> 
> 
> Thanks,
> 
> Neil Conway
> 
>


Re: Review Request 45999: Ensure GC is terminated last during libprocess shutdown.

Posted by Neil Conway <ne...@gmail.com>.

> On April 11, 2016, 4:06 a.m., Kevin Klues wrote:
> > 3rdparty/libprocess/src/process.cpp, lines 2170-2195
> > <https://reviews.apache.org/r/45999/diff/1/?file=1338571#file1338571line2170>
> >
> >     Is there a reason, we can't just remove the gc process from the list as a preliminary step, and then leave the rest of the logic as is? Followed by the explicit gc terminate/delete?

That was actually the original approach that I took, but removing `gc` from `processes` manually breaks some invariants in the code -- for example, `process::wait(gc)` is a no-op if `gc` has been removed from `processes`, which makes it a bit tricky to wait for `gc` to terminate. I think the approach in this RR is safer.


- Neil


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45999/#review128061
-----------------------------------------------------------


On April 11, 2016, 3:48 a.m., Neil Conway wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/45999/
> -----------------------------------------------------------
> 
> (Updated April 11, 2016, 3:48 a.m.)
> 
> 
> Review request for mesos and Joris Van Remoortere.
> 
> 
> Bugs: MESOS-5144
>     https://issues.apache.org/jira/browse/MESOS-5144
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Without this change, `finalize()` terminates processes in the order
> that they happen to be found when iterating over the `processes`
> map. That means that if the GarbageCollector process is terminated
> while any GC-managed processes are still running, those processes
> will not be GC'd (i.e., they will be leaked).
> 
> Fix this by skipping the garbage collector process when iterating
> over `processes` in `finalize()`, and then only terminating it after
> all other processes have been terminated.
> 
> 
> Diffs
> -----
> 
>   3rdparty/libprocess/src/process.cpp 5e9dcfdc52f3a8223bc43af149b8e1f5dbdf5b0a 
> 
> Diff: https://reviews.apache.org/r/45999/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> Confirmed reduction of leaked memory via ASAN on Linux/amd64.
> 
> 
> Thanks,
> 
> Neil Conway
> 
>


Re: Review Request 45999: Ensure GC is terminated last during libprocess shutdown.

Posted by Kevin Klues <kl...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45999/#review128061
-----------------------------------------------------------




3rdparty/libprocess/src/process.cpp (lines 2167 - 2188)
<https://reviews.apache.org/r/45999/#comment191447>

    Is there a reason, we can't just remove the gc process from the list as a preliminary step, and then leave the rest of the logic as is? Followed by the explicit gc terminate/delete?


- Kevin Klues


On April 11, 2016, 3:48 a.m., Neil Conway wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/45999/
> -----------------------------------------------------------
> 
> (Updated April 11, 2016, 3:48 a.m.)
> 
> 
> Review request for mesos and Joris Van Remoortere.
> 
> 
> Bugs: MESOS-5144
>     https://issues.apache.org/jira/browse/MESOS-5144
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Without this change, `finalize()` terminates processes in the order
> that they happen to be found when iterating over the `processes`
> map. That means that if the GarbageCollector process is terminated
> while any GC-managed processes are still running, those processes
> will not be GC'd (i.e., they will be leaked).
> 
> Fix this by skipping the garbage collector process when iterating
> over `processes` in `finalize()`, and then only terminating it after
> all other processes have been terminated.
> 
> 
> Diffs
> -----
> 
>   3rdparty/libprocess/src/process.cpp 5e9dcfdc52f3a8223bc43af149b8e1f5dbdf5b0a 
> 
> Diff: https://reviews.apache.org/r/45999/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> Confirmed reduction of leaked memory via ASAN on Linux/amd64.
> 
> 
> Thanks,
> 
> Neil Conway
> 
>