Mark Hollingsworth Thanks for the feedback. As far as I can tell though there is no way for the "Creating placeholder flownodes because failed loading originals" message to be printed to the build log unless the Pipeline is resuming, which should only happen when Jenkins starts (although bugs in historical builds that have a broken state could also cause them to attempt to resume when they shouldn't). The full stack trace would help clarify, if "jenkins.model.Jenkins.<init>" is part of the stack trace for the exception, then Jenkins is starting up.

As a general update, my current understanding of this problem based on the data I have received is that in most cases it happens for Pipelines using the PERFORMANCE_OPTIMIZED durability level when Jenkins crashes. The PERFORMANCE_OPTIMIZED durability level makes no guarantees that Pipelines will be resumable if Jenkins crashes, so this behavior is expected in that case. There is supposed to be a more user-friendly error message explaining that the Pipeline cannot be resumed for these reasons rather than just the raw error of what exactly kept the Pipeline from resuming, but that is broken because of JENKINS-53358.

I have a draft PR up to try to improve the messaging around this case, so that something like the following would be printed in these cases instead:

Unable to resume because the Pipeline is using the PERFORMANCE_OPTIMIZED durability level but was not saved before Jenkins stopped. Did Jenkins crash? See https://www.jenkins.io/doc/book/pipeline/scaling-pipeline/ for details about Pipeline durability levels.

If anyone has seen this issue with Pipelines that are not using the PERFORMANCE_OPTIMIZED durability level, or can show Jenkins system logs and service logs that show this issue occurring with a PERFORMANCE_OPTIMIZED Pipeline even with a normal Jenkins restart, with log messages as described in this comment showing that the Pipeline was persisted before shutdown, that would be very interesting and we should create new tickets to track those things because they would be distinct issues. For some of the stack traces here, it looks like there is a problem where CpsStepContext.isReady is resulting in Pipelines being resumed, which is strange; I am not sure how to reproduce those issues and they probably need to be investigated separately.

Devin Nusbaum so we definitely see jenkins.model.Jenkins.<init> as part of the trace, but I'm almost 100% certain that Jenkins did not startup (I would have been paged very early in the morning if it had). What I can tell you is that we have been reproducing this issue by terminating machines. 

[2020-07-24T13:47:01.622Z] Cannot contact someNode: java.lang.InterruptedException
Creating placeholder flownodes because failed loading originals.
java.io.IOException: Tried to load head FlowNodes for execution Owner[someFolder/someJob/buildNumber:someFolder/someJob #buildNumber] but FlowNode was not found in storage for head id:FlowNodeId 1:1714
 at org.jenkinsci.plugins.workflow.cps.CpsFlowExecution.initializeStorage(CpsFlowExecution.java:679)
 at org.jenkinsci.plugins.workflow.cps.CpsFlowExecution.onLoad(CpsFlowExecution.java:716)
 at org.jenkinsci.plugins.workflow.job.WorkflowRun.getExecution(WorkflowRun.java:680)
 at org.jenkinsci.plugins.workflow.job.WorkflowRun.onLoad(WorkflowRun.java:539)
 at hudson.model.RunMap.retrieve(RunMap.java:225)
 at hudson.model.RunMap.retrieve(RunMap.java:57)
 at jenkins.model.lazy.AbstractLazyLoadRunMap.load(AbstractLazyLoadRunMap.java:501)
 at jenkins.model.lazy.AbstractLazyLoadRunMap.load(AbstractLazyLoadRunMap.java:483)
 at jenkins.model.lazy.AbstractLazyLoadRunMap.getByNumber(AbstractLazyLoadRunMap.java:381)
 at hudson.model.RunMap.getById(RunMap.java:205)
 at org.jenkinsci.plugins.workflow.job.WorkflowRun$Owner.run(WorkflowRun.java:929)
 at org.jenkinsci.plugins.workflow.job.WorkflowRun$Owner.get(WorkflowRun.java:940)
 at org.jenkinsci.plugins.workflow.flow.FlowExecutionList$1.computeNext(FlowExecutionList.java:65)
 at org.jenkinsci.plugins.workflow.flow.FlowExecutionList$1.computeNext(FlowExecutionList.java:57)
 at com.google.common.collect.AbstractIterator.tryToComputeNext(AbstractIterator.java:143)
 at com.google.common.collect.AbstractIterator.hasNext(AbstractIterator.java:138)
 at org.jenkinsci.plugins.workflow.flow.FlowExecutionList$ItemListenerImpl.onLoaded(FlowExecutionList.java:178)
 at jenkins.model.Jenkins.<init>(Jenkins.java:1017)
 at hudson.model.Hudson.<init>(Hudson.java:85)
 at hudson.model.Hudson.<init>(Hudson.java:81)
 at hudson.WebAppMain$3.run(WebAppMain.java:262)

What I can tell you is that we have been able to reproduce this issue by terminating agent machines too early.

In this case our infrastructure killed the agent in job1 before the job finished completely and we repro the issue.

Job2 is the following job that did not get cut off early. 

Job 1
16:23:36 [WS-CLEANUP] Deleting project workspace...
16:23:36 [WS-CLEANUP] Deferred wipeout is used...
16:23:36 [WS-CLEANUP] done
[Pipeline] }
[Pipeline] // node
[Pipeline] }
[Pipeline] // stage
[Pipeline] }
[Pipeline] // node
[Pipeline] }
----------------------------
Job 2
16:16:42 [WS-CLEANUP] Deleting project workspace...
16:16:42 [WS-CLEANUP] Deferred wipeout is used...
16:16:42 [WS-CLEANUP] done
[Pipeline] }
[Pipeline] // node
[Pipeline] }
[Pipeline] // stage
[Pipeline] }
[Pipeline] // node
[Pipeline] }
[Pipeline] // parallel
[Pipeline] }
[Pipeline] // script
[Pipeline] }
[Pipeline] // stage
[Pipeline] stage
[Pipeline] { (Declarative: Post Actions)
[Pipeline] node
16:16:42 Running on a node
[Pipeline] {
[Pipeline] echo
16:16:42 Reporting build status: UNSTABLE
[Pipeline] notifyBitbucket
 notifying stuff
[Pipeline] step
16:16:44 Notifying Bitbucket
16:16:44 Notified Bitbucket 
[Pipeline] }
[Pipeline] // node
[Pipeline] }
[Pipeline] // stage
[Pipeline] }
[Pipeline] // timestamps
[Pipeline] }
[Pipeline] // withEnv
[Pipeline] End of Pipeline

Hope that helps!

I found the reason for the FlowNode errors on my system:

Sep  6 00:29:20 ship kernel: [16183195.286559] Out of memory: Kill process 27501 (java) score 122 or sacrifice child
Sep  6 00:29:20 ship kernel: [16183195.292176] Killed process 27501 (java) total-vm:21995844kB, anon-rss:5537216kB, file-rss:0kB, shmem-rss:0kB
Sep  6 00:29:20 ship kernel: [16183196.039063] oom_reaper: reaped process 27501 (java), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB

And it is very consistent. I checked the last 20 occurrences, every time the flow node error occurs, the OOM-killer does his work. Why the java process on the build-node requires around 20GB of memory is unclear, because the build itself does not have this huge memory requirements (maybe some other issue).

Since the OOM kills the java process, there is no proper feedback from the FlowNode... maybe some sanity check if the build client process should do the trick and produce some helpful error message.

I my case, more memory helped...

I am able to consistently get this error. I have attached the jobs/yeticore/branches/master/builds/2/workflow-fallback/flowNodeStore.xml . My interpretation of this file could be wrong but it appears it it searching for flow node 49, but all the file has is flow nodes 51 and the 52.  This the file in workflow-fallback, the flow in workflow seems to have a single node with id 2.

Totally blocked at this time, so willing to investigate further 

Update -  I am running jenkins inside docker container and in Docker - Preferences - Resources - Memory , I increased memory from 2GB to 16GB (I restarted docker) and I was able to get a successful run.  Will keep monitoring.

 flowNodeStore.xml

On my side adding more memory to my docker container runtime worked as well.

It would be helpful if this exception would be catched and a meaningful message would be shown to the users (low memory).