So far i regularly detect this problem using the script and fix it manually.

What value are you seeing as the wrong host? The Jenkins master or another slave?  I presume that freestyle job is fixed to build on one slave.

Exampletaken from above:

jenkins_frinavci_FRI00040598D_spine2d.w64-vc14-qt56x.debug-coverage: FRIjenkins04

The name of the workspace is jenkins_frinavci_FRI00040598D_spine2d.w64-vc14-qt56x.debug-coverage. This is derived from the workspace name format jenkins_frinavci_${NODE_NAME}_${JOB_NAME}. So I would expect that the host entered into the client spec is FRI00040598D. But it is FRIjenkins04.

Does this explain my problem?

Both FRI00040598D and FRIjenkins04 are slave names.

If necessary: we are a commercial Perforce customer. So we could open an official ticket if this helps.

A support ticket would be great, our support team can work with you on a reproduction. It seems strange a different node is used to create the workspace.  Are you sure the Jenkins Job is restricted to one slave?

No, the job is not restricted to one slave. That is why I added NODE_NAME to the workspace name ... to have a unique client spec per node ...

Hmmm ... time to talk about the intended behaviour of that feature Pin the workspace to the build host ...

My perception is that the following activities happen (based on jenkins_${NODE_NAME}_${JOB_NAME} as the workspace name format):

1. initially the job MyJob runs on SlaveA => a new client spec is created with jenkins_SlaveA_MyJob
2. second time it runs on SlaveB => a new client spec is created with jenkins_SlaveB_MyJob
3. third time it runs on SlaveA again => the client spec jenkins_SlaveA_MyJob is used

My understanding is that for each (Slave,Job) pair a dedicated client spec is created?

Totally wrong?

Is the workspace name created only once or is that expression evaluated each time?

It's been a while since I looked at the host feature - so a quick look at the code (AbstractTask):

public Workspace setEnvironment(Run<?, ?> run, Workspace wsType, FilePath buildWorkspace)
      throws IOException, InterruptedException {

   Workspace ws = (Workspace) wsType.clone();

   // Set Node environment
   EnvVars envVars = run.getEnvironment(listener);
   String nodeName = NodeHelper.getNodeName(buildWorkspace);
   envVars.put("NODE_NAME", envVars.get("NODE_NAME", nodeName));
   String executor = ExecutorHelper.getExecutorID(buildWorkspace);
   envVars.put("EXECUTOR_NUMBER", envVars.get("EXECUTOR_NUMBER", executor));

   ws.setExpand(envVars);

   // Set workspace root (check for parallel execution)
   String root = buildWorkspace.getRemote();
   if (root.contains("@")) {
      root = root.replace("@", "%40");
   }
   ws.setRootPath(root);

   if (ws.isPinHost()) {
      String hostname = getHostName(buildWorkspace);
      ws.setHostName(hostname);
   } else {
      ws.setHostName("");
   }
   return ws;
}

The setEnvironment method is called every time the workspace is used (that could be for checkout, cleanup, publish, etc...) there is a remote task to evaluate the host 'getHostName(buildWorkspace)', perhaps Jenkins is executing this on a different host?  The simplest option to fix this is to use the 'nodeName', however the user MUST set the node name the same as the host (I know that seems obvious, but you would be amazed at what some users will configure).

Maybe one could add a check inside P4Hook.java whether a job is enabled before doing anything?

Or dou you expect to get only active jobs inside the jobs parameter of probeJobs()?

Yes it looks like an oversight on the triggered probe option.  Please can you raise that as a separate JIRA.

See JENKINS-50634 for the separate improvement issue.

heiko_nardmann thanks for you continuous valuable contribution to perforce Jenkins plugin. In referee to your question regarding raising a support ticket, since you are a paying Perforce customer, please raise a support ticket whenever you raise a Perforce related issue on this forum. While widely impacting issues raised by anyone get high priority on merit, we do unashamedly provide special priority to our paying customers

The change in Host field is being triggered by the Workspace cleanup:

   https://github.com/jenkinsci/jenkins/blob/a1cad0b839828e15d21a41c8ed83194e4c3eec9d/core/src/main/java/hudson/model/WorkspaceCleanupThread.java#L136

This calls:

https://github.com/jenkinsci/p4-plugin/blob/9c7cbf52a3a1468a7cd517a00f13e11ccb16cce3/src/main/java/org/jenkinsci/plugins/p4/PerforceScm.java#L629

Heiko proved this by comparing the spec depot updates to the time of the automated cleaning seen in '/var/lib/jenkins/logs/task/Workspace\ clean-up.log:

Started at Thu Apr 19 17:11:25 UTC 2018
Multibranch_test
Checking FilteredPipeline
Checking Freestyle_test

Moving to VERIFIED.

Proving very difficult to fix. The concept of a job having different workspaces is alien to Jenkins so the correct information is currently not (and unlikely to ever be) provided to P4Jenkins to clean up properly.

For now we are going to have to recommend that pin at build host is no longer used and {NODE_NAME} is used in workspace name instead.

Doc required.

Maybe

https://github.com/jenkinsci/p4-plugin/blob/master/docs/WORKSPACEMANUAL.md

needs to be changed here?