Builds blocked in POST ACTION by a build launched previously
Hi,
I have a problem that arrived sometimes : some builds of a project can be blocked by a build launched previously.
example :
Job A :
14h50 -> build 1 is launched
15h16 -> build 2 is launched
15h20 -> build 3 is launched
15h21 -> build 3 finished (and result is SUCCESS) but is still running...
15h48 -> build 2 finished (and result is ERROR) but is still running...
16h31 -> build 1 finished (and result is SUCCESS)
-> automatically builds 2 and 3 finished.
Our jobs only launch perl script.
When I say that a build is finished, it is because I can see the last line of log that I print in my perl script.
It also still running because I can see the little loading image at the end of the console page (and the status of the job is "building").
When the build 1 is finished, I have some lines of logs added to the console of builds 2 and 3.
We use jenkins Text Finder Plugin to parse log after the end of a build. If a specific chain is found in log, job is marked as UNSTABLE.
This is the only post action for our jobs.
I have attached some snapshots of what looks like a job console output (job_blocked.png) when it is blocked (it is a different job which has the same problem).
I also made a snapshot (job_final.png) of what happenened when the job is released after the build 1 finished (a case when the job finish in UNSTABLE status).
We have a master on linux that launch all the builds on linux slaves. The master connect to slaves by executing this request :
ssh user@host /usr/java/jdk1.6.0_20/bin/java_utf8 -jar /path/to/slave/workspace/slave.jar -jnlpUrl http:/jenkins/computer/slave_name/slave-agent.jnlp -jnlpCredentials user:pwd -classpath "/path/to/commons-codec-1.6/commons-codec-1.6.jar" &
This problem happens sometimes and is quite problematic since it can block a lot of builds if one is too long.
In my log I did not find any clue to this problem.
I have the line :
"16 juil. 2012 15:21:09 hudson.model.Run run
INFO: Job A #3 main build action completed: SUCCESS"
saying that the job is finished but jenkins still thinks it is running...
I did find some tickets related to my problem (JENKINS-13330) but it's not exactly the same problem.
Not a core bug: It's up to the specific post-build step whether to require all previous builds be finished before it can run.
IIRC there have been fixes in Mailer/Email-ext though, so they no longer block build completion just to deliver a correct subject line ("still failing" vs. "failed").
If you want specific post-build steps to change their behavior, please file issues against them. In the case of Text Finder, this is
JENKINS-17507and was resolved in 1.10.