Uploaded image for project: 'OpenDJ'
  1. OpenDJ
  2. OPENDJ-3283

Cleaner threads unable to clean files, changelogDb grows until disk fills up

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: 2.6.3
    • Fix Version/s: 4.0.0
    • Component/s: backends
    • Labels:
    • Support Ticket IDs:

      Description

      There are three replicating DS+RS servers. The client read and write traffic is generally pointing at the primary and secondary servers, while the 3rd backup server is almost purely updating its changelogDb.

      There is significant write load on the "ou=rprofiles,dc=example,dc=com" domain. Each change is approx 70KB. All three servers have a 24 hour purge-delay.

      Under some unknown conditions, the changelogDb on one of the servers (most recently on the backup server) has started growing until the disk has nearly filled.

      The changelogDb directories are approximately 150GB on the primary and secondary servers, and over 1TB on the backup server.

      We dumped the rprofiles changes from all 3 servers:

                          primary         secondary       backup
                          changes size    changes size    changes size
      22098 ou=rprofiles  1893210 57.7GB  1877264 57.2GB  1572153 46.2GB
      32057 ou=rprofiles  1901216 58.0GB  1891239 57.7GB  1562067 45.8GB
      

      So it looks as though the checkpointer threads on the backup server are properly trimming old changes. The disparity in sizes is due to the backup server being run for a bit longer than the other servers.

      However the cleaner threads seem to be in some serious trouble, and they are not catching up.
      We observe the following messages frequently in the backup server's changelogDb/je.info.0 file:

      2016-08-29 03:12:52.257 UTC INFO [/prod/dsd/apps/opendj/changelogDb] Chose lowest utilized file for cleaning. fileChosen: 0xc040a lnSizeCorrectionFactor: 1.1420875 totalUtilization: 49 bestFileUtilization: 4 isProbe: false
      2016-08-29 03:12:52.684 UTC SEVERE [/prod/dsd/apps/opendj/changelogDb] Average cleaner backlog has grown from 110238.8 to 110240.4. If the cleaner continues to be unable to make progress, the JE cache size and/or number of cleaner threads are probably too small. If this is not corrected, eventually all available disk space will be used.
      

      Note the high `lnSizeCorrectionFactor` value. The "backlog" seems to be roughly the number of files to clean.

      We also see this frequently:

      2016-08-29 03:12:52.752 UTC WARNING [/prod/dsd/apps/opendj/changelogDb] Cleaner has 16274 files not deleted because of read-only processes.
      

      The cause of that second message is not clear. An online backup does *not* access the environment in this way.

      Using DbSpace -r we can see that there are a large number of files with very low occupancy. Top 10:

      52470 total files with %0 occupancy
      56309 total files with %1 occupancy
      12273 total files with %2 occupancy
      2491 total files with %3 occupancy
      1068 total files with %4 occupancy
      1161 total files with %5 occupancy
      1164 total files with %6 occupancy
      890 total files with %7 occupancy
      751 total files with %8 occupancy
      689 total files with %9 occupancy
      

      The changelogDb only uses 2 cleaner threads and this is not presently configurable. However note that the other 2 servers are able to keep up with this trim load so I don't believe the number of cleaner threads is the issue.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                matthew Matthew Swift
                Reporter:
                cjr Chris Ridd
              • Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: