[
https://issues.apache.org/jira/browse/LUCENE-6530?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14576692#comment-14576692
]
Uwe Schindler commented on LUCENE-6530:
---------------------------------------
bq. // There is a risk that pid will be recycled, causing us to kill the wrong
process!
Very funny. I always had in mind, that process IDs are not recycled if you
don't call {{waitpid}} on the OS level :-) Maybe there is just some state
missing in UNIXProcess class! If you don't call {{waitpid}} and react to
SIGCHLD signal handler it will stay alive in the process list as "zombie"... I
assume Java *might* kill the process although it got SIGCHLD before! Maybe
Slowlaris is different and does not know Zombie processes :-)
> Use Java 7 ProcessBuilder.inheritIO() instead of own ThreadPumper
> -----------------------------------------------------------------
>
> Key: LUCENE-6530
> URL: https://issues.apache.org/jira/browse/LUCENE-6530
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Uwe Schindler
> Assignee: Uwe Schindler
> Priority: Minor
> Labels: Java7
> Fix For: Trunk, 5.3
>
> Attachments: LUCENE-6530.patch, LUCENE-6530.patch
>
>
> In some tests wie spawn separate processes (TestIndexWriterOnJRECrash and
> Solr's IPTables). To capture stdin/stdout/stderr we spawn several threads
> that pump those to stdout/stderr.
> Since Java 7 there is ProcessBuilder.inheritIO() that does this for us
> without any additional threads. We should use this instead. Fix is easy, just
> remove some stuff :-)
> I did the same already for my Codec classloader deadlock test, so this is
> just a followup for the other tests.
> Patch is attached and can be committed to trunk and 5.x.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
