[
https://issues.apache.org/jira/browse/MESOS-10105?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17322254#comment-17322254
]
Charles Natali commented on MESOS-10105:
----------------------------------------
Regarding {{-fsanitize=address}} , leak detection can be disabled with
{{ASAN_OPTIONS=detect_leaks=0}}.
However a couple tests still fail:
{{
[ FAILED ] HTTPCommandExecutorTest.TerminateWithACK
[ FAILED ] PosixRLimitsIsolatorTest.UnsetLimits
[ FAILED ] MesosContainerizer/DefaultExecutorTest.KillTask/0, where
GetParam() = "mesos"
[ FAILED ]
MesosContainerizer/DefaultExecutorTest.CommitSuicideOnTaskFailure/0, where
GetParam() = "mesos"
[ FAILED ] MesosContainerizer/DefaultExecutorTest.CommitSuicideOnKillTask/0,
where GetParam() = "mesos"
[ FAILED ] MesosContainerizer/DefaultExecutorTest.MaxCompletionTime/0, where
GetParam() = "mesos"
}}
All of them except {{PosixRLimitsIsolatorTest.UnsetLimits}} fail because they
don't propagate the {{ASAN_OPTIONS}} environment variable.
> Make tests of builds with -fsanitize=address/memory/undefined/thread pass.
> ---------------------------------------------------------------------------
>
> Key: MESOS-10105
> URL: https://issues.apache.org/jira/browse/MESOS-10105
> Project: Mesos
> Issue Type: Wish
> Reporter: Andrei Sekretenko
> Priority: Critical
>
> As exemplified by various C++ projects and also by targeting specific issues
> in Mesos (for example, MESOS-10102), running code built with clang sanitizers
> helps with uncovering undefined behavior and data races.
> Sanitizer adoption usually happens as a sequence of steps which unblock each
> other:
> 1) making local tests pass under sanitizer at least once
> 2) making CI regularly run sanitizer builds (so that new sanitizable bugs are
> not introduced and more bugs not triggered deterministically are uncovered)
> 3) running high-level integration tests, betas, etc. with sanitizer builds
> --
> (3) is definitely out of scope of this wish, and it is not clear if (2) will
> fit into ASF CI, but (1) is definitely doable, and on its own can lead to
> figuring out causes of mysterious rare bugs (which might turn out to be not
> so rare under certain conditions).
> --
> State of Mesos w.r.t sanitizers:
> - as of Mar 2020, Mesos tests built with -fsanitize=address crash due to
> several locations that leak one object per thread lifetime
> - as of Nov 2019, libprocess tests were crashing thread sanitizer; IIRC, the
> issues in libprocess on Linux/amd64 are also "technical", but probably could
> result in a very real problems on a different platform
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
