Uploaded image for project: 'Jenkins'
  1. Jenkins
  2. JENKINS-18539

CVS-Plugin 2.9 consumes 10x more hard disk usage than 1.6

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Minor Minor
    • cvs-plugin
    • None
    • Jenkins 1.516 on Tomcat 6, RedHat 5.6 64 Bit, CVS 1.11.22 (on both build and cvs server), jdk1.6.0_26

      Upgrading from 1.509.1 to 1.516 updated the bundled cvs-plugin from 1.6 to 2.8.
      I upgraded to 2.9 but as far as I can remember 2.8 was the same problem:

      After the upgrade the disk load on our CVS server increased by about 10 times and user startet complaining why cvs is so slow these days.
      I also noticed a few 1000MHz more CPU-Load on our Jenkins Server.

      I now did a downgrade from CVS-Plugin 2.9 to 1.6 which fixed the problem.

      It can be seen clearly on the attached graphics.
      (upgrade was on 13.6.2013, downgrade on 21.6.2013)
      (dont get confused by the last day's data, its so many measure points so its getting filled up).

      Unfortunatelly the 1.6 plugin seems not to be 100% compatible with Jenkins 1.516 because its startet generating "You have data stored in an older format and/or unreadable data." messages. So I added this bug as Critical because at some point I estimate v1.6 it will prevent me from upgrading our Jenkins server.

          [JENKINS-18539] CVS-Plugin 2.9 consumes 10x more hard disk usage than 1.6

          Reducing priority - the data format warning is caused by switching between new and old versions of the plugin and can be ignored. There are currently no API changes in Jenkins that prevent the use of 1.6 of the CVS plugin

          Michael Clarke added a comment - Reducing priority - the data format warning is caused by switching between new and old versions of the plugin and can be ignored. There are currently no API changes in Jenkins that prevent the use of 1.6 of the CVS plugin

          The change is probably being caused by polling being done using rlog rather than log, so the CVS server is doing more work. It's also worth checking what compression settings you currently have for communication. Could you give some indication of your job setup (number of jobs, polling frequency etc)?

          There's probably very little that can be done about this as we're at the mercy of how CVS operates and there isn't a nice polling API.

          Michael Clarke added a comment - The change is probably being caused by polling being done using rlog rather than log, so the CVS server is doing more work. It's also worth checking what compression settings you currently have for communication. Could you give some indication of your job setup (number of jobs, polling frequency etc)? There's probably very little that can be done about this as we're at the mercy of how CVS operates and there isn't a nice polling API.

          Reducing priority is fine if v1.6 API is still supported.

          We are running 790 jobs in total, about 95% using CVS. We have 4 slaves (139, 137, 14, 20 jobs), the rest is on master.
          Master has 4 executors, each slave 1 executor. Concurrent SCM polling is limited to 5.
          Jobs polling was "0,30 6-22 * * *" and is now H 6-22 * * * (but that didn't really change the load).
          Nightly builds run with "00 02 * * 1-5".

          The "Disable CVS compression" checkbox is not active. When I had installed 2.9 I didn't make any changes to the default value (I think it was "default: 3").

          But can the change to rlog really increase the load by 10 times?

          Florian Rosenauer added a comment - Reducing priority is fine if v1.6 API is still supported. We are running 790 jobs in total, about 95% using CVS. We have 4 slaves (139, 137, 14, 20 jobs), the rest is on master. Master has 4 executors, each slave 1 executor. Concurrent SCM polling is limited to 5. Jobs polling was "0,30 6-22 * * *" and is now H 6-22 * * * (but that didn't really change the load). Nightly builds run with "00 02 * * 1-5". The "Disable CVS compression" checkbox is not active. When I had installed 2.9 I didn't make any changes to the default value (I think it was "default: 3"). But can the change to rlog really increase the load by 10 times?

            Unassigned Unassigned
            skybird Florian Rosenauer
            Votes:
            1 Vote for this issue
            Watchers:
            3 Start watching this issue

              Created:
              Updated: