[
https://issues.apache.org/jira/browse/YETUS-286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15102449#comment-15102449
]
Allen Wittenauer commented on YETUS-286:
----------------------------------------
bq. gcc processes also don't use a lot of memory when compared to maven (most
of the time)
It's one of the reasons why we started reporting the memory used by maven in
the report since people were setting ridiculous values for MAVEN_OPTS. It's
pretty rare to see maven go above 200MB.
bq. So I'm skeptical that running make -j 16 will be that harmful in practice.
A change is made to the build process and builds suddenly start failing on a
regular basis...
bq. We could certainly look into tuning this, but I don't see why it would be
the root cause of "failed to map segment from shared object" failures.
Even if it isn't, taking all of the CPUs and likely a significantly larger
chunk of memory on a box where the resources are shared amongst different teams
simultanously is not a good idea. I wouldn't be surprised if INFRA smacked
hadoop down with the hammer of justice.
Oh, forgot to answer:
bq. Is selinux enabled in any of our Docker images?
The Dockerfile used for Hadoop comes from Hadoop: dev-support/docker/Dockerfile.
> The native build randomly fails with "failed to map segment from shared
> object"
> -------------------------------------------------------------------------------
>
> Key: YETUS-286
> URL: https://issues.apache.org/jira/browse/YETUS-286
> Project: Yetus
> Issue Type: Bug
> Components: Test Patch
> Affects Versions: 0.2.0
> Reporter: Colin Patrick McCabe
> Assignee: Sean Busbey
> Priority: Critical
>
> On Jenkins, the native build seems to intermittently fail with "failed to map
> segment from shared object." For example:
> {code}
> [WARNING] /usr/bin/cmake: error while loading shared libraries: libxml2.so.2:
> failed to map segment from shared object: Permission denied
> {code}
> We have also seen this happen with {{librtmp.so.0}}, {{libcurl.so.4}}, and
> other native libraries which are dependencies of the thing being built.
> Basically, there is some random chance any particular native dependency will
> get "failed to map segment." When this happens, the native build fails and
> returns a -1 even if there was no actual problem. This requires us to re-run
> Jenkins to get a correct result. The problem does not seem to reproduce
> locally.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
