mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Benjamin Mahler <bmah...@apache.org>
Subject Re: Review Request 60397: Check perf version compatibility in tests with disabled coredumps.
Date Sat, 01 Jul 2017 09:29:30 GMT


> On June 27, 2017, 5:44 a.m., Benjamin Mahler wrote:
> > src/tests/environment.cpp
> > Line 542 (original), 542-548 (patched)
> > <https://reviews.apache.org/r/60397/diff/1/?file=1761619#file1761619line542>
> >
> >     Shouldn't this just be covered by `perf::suppported`?
> >     
> >     Would also be great to include some additional context here, like a pointer
to the ticket or mentioning that this is more likely if running within a docker image?
> 
> Andrei Budnik wrote:
>     Ticket is [MESOS-7160](https://issues.apache.org/jira/browse/MESOS-7160).
>     
>     What I've got from the description of the problem is that `perf --version` segfaults
on an incompatible kernel version.
>     The solution for this problem is provided in this patch. Then, we have found that
it is more likely caused by incorrect `os::subprocess()` behavior. So, I have suspicion that
`perf --version` might never really cause segfaults. I'm going to fix `os::subprocess()` and
send a new patch. Afterwards, if the problem reproduces after the fix, then I'll continue
to work on this patch.
>     
>     `perf::suppported()` invokes `os::subprocess()` which calls `abort()` when path to
a binary is invalid. SIGABRT causes coredumps.
> 
> James Peach wrote:
>     Both of these are true :) `perf` will crash with kernel version mismatches, and I
strongly suspect there is a race in the process supervisor that can cause a similar symptom
when `perf` is not installed.

I think my comment was misunderstood, I'm asking why we wouldn't have perf::supported handle
this correctly, rather than every caller having to write logic to defend against an implementation
issue in perf::supported.


- Benjamin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/60397/#review178952
-----------------------------------------------------------


On June 30, 2017, 5:41 p.m., Andrei Budnik wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/60397/
> -----------------------------------------------------------
> 
> (Updated June 30, 2017, 5:41 p.m.)
> 
> 
> Review request for mesos, Benjamin Bannier, James Peach, and Neil Conway.
> 
> 
> Bugs: MESOS-7160
>     https://issues.apache.org/jira/browse/MESOS-7160
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> For autotools build, the docker-build script performs a 'distcheck'
> build. This type of build warns if any unexpected files are left in
> the build directory after an uninstall, mainly to detect broken
> uninstall Makefile rules. The return status of the build container is
> the result of the distcheck.
> This fixes an issue where in some dockerized configurations
> invocations of 'perf' segfaulted (producing a core file as a
> side-effect), where the failure case was already anticipated and
> handled by the caller.
> 
> 
> Diffs
> -----
> 
>   src/tests/containerizer/perf_tests.cpp d8aab08eb131f974821fb85662cbc6cc685d2f3e 
>   src/tests/environment.cpp a7262cd97361b55ff31238341657e764df6c9ea5 
> 
> 
> Diff: https://reviews.apache.org/r/60397/diff/2/
> 
> 
> Testing
> -------
> 
> 1. make check (mac os x 10.12, fedora 25)
> 2. internal CI
> 
> Needs to be confirmed by Apache CI, e.g., reviewbot.
> 
> 
> Thanks,
> 
> Andrei Budnik
> 
>


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