mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jiang Yan Xu <...@jxu.me>
Subject Re: Review Request 47973: Updated gc to prevent early exit in case of error.
Date Mon, 13 Jun 2016 22:52:23 GMT


> On June 8, 2016, 9:54 a.m., Jiang Yan Xu wrote:
> > src/tests/gc_tests.cpp, lines 948-956
> > <https://reviews.apache.org/r/47973/diff/2/?file=1400007#file1400007line948>
> >
> >     Why do we still need `containerPaths`? We can get the sandbox path directly:
> >     
> >     ```
> >     const string sandbox = slave::paths::getExecutorLatestRunPath(
> >         flags.work_dir,
> >         slaveId,
> >         frameworkId.get(),
> >         executor_id);
> >         
> >     ASSERT_TRUE(os::exists(sandbox));
> >     ```
> >     
> >     right?
> 
> Megha Sharma wrote:
>     getExecutorLatestPath(...) gives us the executor run path for latest symlink which
gets collected by GC. So, this path can't be used for post GC assertions. Since we didn't
create a container ourselves in the test so we didn't have a container id to call the method
getExecutorRunPath(..., containerId). Here, I am getting all the run paths with getExecutorRunPaths(..)
and then elimating the one with latest symlink to get the actual path.

Chatted offline about the use of `os::realpath()` to solve this.


- Jiang Yan


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/47973/#review136592
-----------------------------------------------------------


On June 8, 2016, 9:57 a.m., Megha Sharma wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/47973/
> -----------------------------------------------------------
> 
> (Updated June 8, 2016, 9:57 a.m.)
> 
> 
> Review request for mesos and Jiang Yan Xu.
> 
> 
> Bugs: MESOS-5196
>     https://issues.apache.org/jira/browse/MESOS-5196
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> It's possible for tasks and isolators to lay down files that are not 
> deletable by GC. In the face of such errors GC needs to free up disk 
> space wherever it can because it's already re-offered to frameworks.
> 
> 
> Diffs
> -----
> 
>   src/slave/gc.cpp eedb8ca713d7f5195055bb6417f03ab70731af97 
>   src/tests/gc_tests.cpp 4cb7c2f612984f7f5a9378a7f972f2438bbf28c5 
> 
> Diff: https://reviews.apache.org/r/47973/diff/
> 
> 
> Testing
> -------
> 
> make check.
> Tested with option --gtest_also_run_disabled_tests.
> 
> 
> Thanks,
> 
> Megha Sharma
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message