-
Bug
-
Resolution: Fixed
-
Major
Found a case where a sh step ceased to produce more output in the middle of a command, for no apparent reason, and the build did not respond to normal abort. The virtual thread dump said
Thread #80 at DSL.sh(completed process (code -1) in /...@tmp/durable-... on ... (pid: ...)) at ...
But there is no active CPS VM thread, and nothing visibly happening on the agent, and all Timer threads are idle. So it seems that a call to check would have caused the step to fail—but perhaps none came?
Possibly stop should do its own check for a non-null Controller.exitStatus and immediately fail in such a case (but we run the risk of delivering doubled-up events if check does run later); or synchronously call check (though this runs the risk of having two such calls run simultaneously—it is not thread safe); or somehow reschedule it (same problem).
At a minimum, the virtual thread dump should indicate what the current recurrencePeriod is. And the calls to schedule could save their ScheduledFuture results in a transient field, so we can check cancelled and done flags. Such diagnostics might make it clearer next time what actually happened.
Also a term claimed to be terminating the sh step, but the build still did not finish. Again nothing in the physical thread dumps, and virtual thread dump still claims to be inside sh. System log showed
... WARNING org.jenkinsci.plugins.workflow.cps.CpsStepContext onFailure already completed CpsStepContext[186]:Owner[...] java.lang.IllegalStateException: org.jenkinsci.plugins.workflow.steps.FlowInterruptedException at org.jenkinsci.plugins.workflow.cps.CpsStepContext.onFailure(CpsStepContext.java:325) at org.jenkinsci.plugins.workflow.job.WorkflowRun$5.onSuccess(WorkflowRun.java:300) at org.jenkinsci.plugins.workflow.job.WorkflowRun$5.onSuccess(WorkflowRun.java:296) at org.jenkinsci.plugins.workflow.support.concurrent.Futures$1.run(Futures.java:150) at com.google.common.util.concurrent.MoreExecutors$SameThreadExecutorService.execute(MoreExecutors.java:253) at com.google.common.util.concurrent.ExecutionList$RunnableExecutorPair.execute(ExecutionList.java:149) at com.google.common.util.concurrent.ExecutionList.execute(ExecutionList.java:134) at com.google.common.util.concurrent.AbstractFuture.set(AbstractFuture.java:170) at com.google.common.util.concurrent.SettableFuture.set(SettableFuture.java:53) at org.jenkinsci.plugins.workflow.cps.CpsFlowExecution$5.onSuccess(CpsFlowExecution.java:702) at org.jenkinsci.plugins.workflow.cps.CpsFlowExecution$5.onSuccess(CpsFlowExecution.java:689) at org.jenkinsci.plugins.workflow.cps.CpsFlowExecution$4$1.run(CpsFlowExecution.java:626) at org.jenkinsci.plugins.workflow.cps.CpsVmExecutorService$1.run(CpsVmExecutorService.java:32) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) at java.util.concurrent.FutureTask.run(FutureTask.java:266) at hudson.remoting.SingleLaneExecutorService$1.run(SingleLaneExecutorService.java:112) at jenkins.util.ContextResettingExecutorService$1.run(ContextResettingExecutorService.java:28) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) at java.util.concurrent.FutureTask.run(FutureTask.java:266) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) at java.lang.Thread.run(Thread.java:745) Caused by: org.jenkinsci.plugins.workflow.steps.FlowInterruptedException at org.jenkinsci.plugins.workflow.job.WorkflowRun.doTerm(WorkflowRun.java:295) at ...
So the program state seems to be somehow inconsistent as well; perhaps sh did complete (it is not shown as in progress in flowGraphTable).
Seems that the virtual thread dump needs some kind of fix TBD to better report the real state of a problematic program.
- relates to
-
-
- Resolved
-
- links to
[JENKINS-37730] DurableTaskStep.Execution hanging after process is dead
Hi, I traced one of the hangs to the problem mentioned here: JENKINS-38566
Ryan Campbell
added a comment - Marking this as critical since it results in unkillable job on the agent, making the agent unusable. Is there a workaround?
Ryan Campbell
added a comment - Marking this as critical since it results in unkillable job on the agent, making the agent unusable. Is there a workaround? The original issue did not result in an unkillable build, just a failure to respond to term (did respond to kill), so recampbell is possibly seeing something unrelated (and eeaston definitely was). This issue is not about a root cause, which remains unknown, but about a robustness aspect of the response to some other bug.
Possibly stop should do its own check for a non-null Controller.exitStatus and immediately fail in such a case
Plan to handle that differently in JENKINS-38769, by just making sure stop always stops the step, regardless of process exit status.
Jonas Schneider
added a comment - We're seeing the same behavior with a roughly similar pipeline, running Docker image `jenkinsci/blueocean:1.0.1` (Jenkins ver. 2.46.1) for master and slave. Abridged pipeline script, thread dump and process dump at:
https://gist.github.com/jonasschneider/ed81faffdd96d3e541cb6f487871029a
After the `docker run` finishes, Jenkins for some reason does not reap the `docker` processes as can be seen in the ps output. This is a blocker, since it hangs our ~10minute builds for multiple hours on end 
. We've only seen it appear under some amount of load, that is, when multiple builds of the same job are running.
Is there any way to better debug what's going on here?
Jonas Schneider
added a comment - We're seeing the same behavior with a roughly similar pipeline, running Docker image `jenkinsci/blueocean:1.0.1` (Jenkins ver. 2.46.1) for master and slave. Abridged pipeline script, thread dump and process dump at:
https://gist.github.com/jonasschneider/ed81faffdd96d3e541cb6f487871029a
After the `docker run` finishes, Jenkins for some reason does not reap the `docker` processes as can be seen in the ps output. This is a blocker, since it hangs our ~10minute builds for multiple hours on end . We've only seen it appear under some amount of load, that is, when multiple builds of the same job are running.
Is there any way to better debug what's going on here? jonasschneider your issue is probably unrelated. While I do not know offhand what would cause the build to “hang” (use the Thread Dump link to get some clues), certainly your script looks wrong since you are launching Docker containers which you never stop. Perhaps you forgot --rm?
Jonas Schneider
added a comment - Sorry, I must've accidentally removed the thread dump from the gist. Here it is: https://gist.github.com/jonasschneider/ed81faffdd96d3e541cb6f487871029a#file-thread-dump-txt – to me this looks like the docker run command has exited, but Jenkins is not proceeding with the execution.
Indeed I am using docker run without --rm. However, docker run will still exit as soon as the script inside exits. The container stops at that time. We do clean up the actual container filesystem later with a docker rm, which I omitted from the Jenkinsfile.
Jonas Schneider
added a comment - Sorry, I must've accidentally removed the thread dump from the gist. Here it is: https://gist.github.com/jonasschneider/ed81faffdd96d3e541cb6f487871029a#file-thread-dump-txt – to me this looks like the docker run command has exited, but Jenkins is not proceeding with the execution.
Indeed I am using docker run without --rm . However, docker run will still exit as soon as the script inside exits. The container stops at that time. We do clean up the actual container filesystem later with a docker rm , which I omitted from the Jenkinsfile. jonasschneider not sure offhand, something clearly went haywire there. Any interesting system log output? Without steps to reproduce I am not sure how to help.
Hi there,
Just letting you know this is hitting me in the wild, thanks for raising this! I get the exact same traceback above in the server logs.
A little background - I've built a moderately complex CPSWorkflowLib for building in-house Python projects with a DSL. The DSL specifies named test stages with simple closures to specify the stage body, and a number of workers to spread the stage over.
This has been working well until I updated to the latest pipeline plugin versions a few days ago; now whenever one of the test stages raises a non-zero exit code in a `sh` step it will hang the build.
I'm trying to come up with a minimal test case that reproduces the problem, it's tricky as there's a lot of setup code in the DSL library that might be affecting things. I'll post it when I get something workable.
Versions: Jenkins 2.7.4, workflow-cps-global-lib 2.3
I might also add this isn't 'minor' ! I can no longer use Pipeline until I find a workaround for this :/