maven-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ben Caradoc-Davies (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MNG-6323) Deadlock in multithreaded dependency resolution
Date Mon, 18 Dec 2017 19:29:00 GMT

    [ https://issues.apache.org/jira/browse/MNG-6323?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16295523#comment-16295523
] 

Ben Caradoc-Davies commented on MNG-6323:
-----------------------------------------

By the way, reverting MNG-6025 (be223808939d75152d7157db33d45f230114555e) did not fix. I did
not try reverting the upgrade to Resolver 1.1.0 (MNG-6258).

> Deadlock in multithreaded dependency resolution
> -----------------------------------------------
>
>                 Key: MNG-6323
>                 URL: https://issues.apache.org/jira/browse/MNG-6323
>             Project: Maven
>          Issue Type: Bug
>    Affects Versions: 3.5.2
>            Reporter: Ben Caradoc-Davies
>         Attachments: geoserver-community-maven-hang-jstack-2.txt, geoserver-community-maven-hang-jstack.txt
>
>
> Maven 3.5.2 multithreaded builds experience a deadlock not seen with Maven 3.5.0.
> To reproduce the issue, clone GeoServer:
> {noformat}
> git clone https://github.com/geoserver/geoserver.git
> cd geoserver
> {noformat}
> Build GeoServer community modules with:
> {noformat}
> mvn -f src/community/pom.xml -B -T4 -U -Prelease -PcommunityRelease -DskipTests clean
install
> {noformat}
> Builds that normally take 2-4 minutes instead experience long hangs. {{jstack}} output
(attached) suggests a deadlock (two different hangs attached). Some of the locks are in {{TIME_WAIT}}
and eventually the build completes after 30-45 minutes, but this is enough to cause builds
on Travis to be killed for their failure to output for ten minutes. (Travis upgraded to Maven
3.5.2 a week ago.)
> I have only seen the failures with -U. The hang does not occur in single-threaded builds.
There are no "*.lock" files in the local repository during the hang so the locks are not mediated
by the filesystem. CPU utilisation is zero suggesting a deadlock not a livelock.
> See also discussion on the geoserver-devel mailing list: http://osgeo-org.1560.x6.nabble.com/Travis-failures-started-with-new-trusty-images-td5346836.html



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message