Uploaded image for project: 'Jenkins'
  1. Jenkins
  2. JENKINS-55308

intermittent "terminated" messages using sh in Pipelines

    Details

    • Type: Bug
    • Status: In Progress (View Workflow)
    • Priority: Minor
    • Resolution: Unresolved
    • Labels:
      None
    • Environment:
    • Similar Issues:

      Description

      Testing Jenkins 2.138.2 LTS, Jenkins pipelines that use sh intermittently throw the following message in the console log …

       

      sh: line 1:  4449 Terminated              sleep 3

       

       … and sometimes this … 

       

      sh: line 1: 13136 Terminated              { while [ ( -d /proc/$pid -o ! -d /proc/$$ ) -a -d '/home/ec2-user/workspace/admin-smoke-test@tmp/durable-523481b0' -a ! -f '/home/ec2-user/workspace/admin-smoke-test@tmp/durable-523481b0/jenkins-result.txt' ]; do    touch '/home/ec2-user/workspace/admin-smoke-test@tmp/durable-523481b0/jenkins-log.txt'; sleep 3;{{done; }}}

       

      Jenkins master runs from a Docker image based on jenkins/jenkins:2.138.2-alpine with specific plugins baked into the image by /usr/local/bin/install-plugins.sh

      The message originates in durable-task-plugin, which must be a dependency of one of the plugins.txt plugins.  

      Two important observations:

      1) The issue does not occur when starting with the base jenkins/jenkins:2.138.2-alpine image and manually installing plugins via UI. That might suggest the issue is around how install-plugins.sh installs plugins and/or dependencies. 

      2) The issue does not occur on our production image, which is also 2.138.2-alpine + plugins built 2018-10-11. Rebuilding the the same image from the same Dockerfile results in different installed plugins. Makes me think results using install-plugins.sh are not deterministic.

        Attachments

          Issue Links

            Activity

            Hide
            carroll Carroll Chiou added a comment -

            A Alvarez thanks, that's good to know

            Show
            carroll Carroll Chiou added a comment - A Alvarez thanks, that's good to know
            Hide
            aal A Alvarez added a comment -

            Hi Carroll Chiou not as far as I'm aware although this could be a default or coming from another plugin. All we did was upgrading the plugin and the messages disappeared from the console output when running shell commands within Pipeline scripts

            Show
            aal A Alvarez added a comment - Hi Carroll Chiou not as far as I'm aware although this could be a default or coming from another plugin. All we did was upgrading the plugin and the messages disappeared from the console output when running shell commands within Pipeline scripts
            Hide
            carroll Carroll Chiou added a comment -

            A Alvarez I'm curious, if you are running 1.33, are you passing in the FORCE_BINARY_WRAPPER=true system property?

            Show
            carroll Carroll Chiou added a comment - A Alvarez I'm curious, if you are running 1.33, are you passing in the FORCE_BINARY_WRAPPER=true system property?
            Hide
            aal A Alvarez added a comment -

            Can confirm that after upgrading durable-task to the latest version, the Terminated messages are gone from our jobs.

            Show
            aal A Alvarez added a comment - Can confirm that after upgrading durable-task to the latest version, the Terminated messages are gone from our jobs.
            Hide
            carroll Carroll Chiou added a comment - - edited

            The sleep 3 process was introduced when the heartbeat check feature was added in version 1.16 of durable-task, which translates to 2.18 of workflow-durable-task-step, which translates to 2.27 of workflow-job.

            Unless I missed a comment, it looks like the pipeline behaves as expected outside of "Terminated sleep 3" line.

            At least in Pablo Rodriguez’s example, this might be a race condition between the sleep 3 process and when the script completes. Like Devin Nusbaum mentioned earlier, the sleep 3 process is used to touch the output log file to show the script is still alive. However, when the script completes, it will write to a separate result file. A watcher service is checking for that result file every 100ms. Once that result file is found, results are transmitted and everything related to that specific step’s workspace is purged. It might be possible that the output file gets cleaned up right after the sleep 3 process checks if the file still exists, but before it gets touched again?

            There is a new release of durable-task (1.31) that removes the sleep 3 process so this line won't pop up anymore.

            Update: I have not been able to reproduce this issue, so I can't say for certain if this issue is resolved. Technically, it should, but it could be possible that the new version just changes the behavior of this bug.

            Show
            carroll Carroll Chiou added a comment - - edited The sleep 3 process was introduced when the heartbeat check feature was added in version 1.16 of durable-task, which translates to 2.18 of workflow-durable-task-step, which translates to 2.27 of workflow-job. Unless I missed a comment, it looks like the pipeline behaves as expected outside of "Terminated sleep 3" line. At least in Pablo Rodriguez ’s example , this might be a race condition between the sleep 3 process and when the script completes. Like Devin Nusbaum mentioned earlier , the sleep 3 process is used to touch the output log file to show the script is still alive. However, when the script completes, it will write to a separate result file. A watcher service is checking for that result file every 100ms. Once that result file is found, results are transmitted and everything related to that specific step’s workspace is purged. It might be possible that the output file gets cleaned up right after the sleep 3 process checks if the file still exists, but before it gets touched again? There is a new release of durable-task (1.31) that removes the sleep 3 process so this line won't pop up anymore. Update: I have not been able to reproduce this issue, so I can't say for certain if this issue is resolved. Technically, it should, but it could be possible that the new version just changes the behavior of this bug.

              People

              • Assignee:
                carroll Carroll Chiou
                Reporter:
                gc875 gc875
              • Votes:
                24 Vote for this issue
                Watchers:
                44 Start watching this issue

                Dates

                • Created:
                  Updated: