Uploaded image for project: 'Jira Data Center'
  1. Jira Data Center
  2. JRASERVER-73032

Altering group membership has performance impact on large instances

XMLWordPrintable

      Issue Summary

      This issue comes from JFR data from JAC.
      Every time we change group membership, nodes in the cluster recalculate the current user count to check against the license. This process is asynchronous and deduplicated, but I believe it needs to be further improved.
      For JAC, I've seen it triggered 375 time in a timespan of 4 hours, with each call taking ~20s (that's a 2h total).
      It has been a problem in the past and is currently invisible in our synthetic testing.

      Steps to Reproduce

      1. Set up a breakpoint in com.atlassian.jira.application.DefaultApplicationRoleManager#clearUserCounts
      2. add a user to the system

      Expected Results

      1. the method will not execute the recalc right away
      2. the method will not be called on all the nodes

      Actual Results

      1. the recalc is initiated right away
      2. each node recalculates data individually

      Workaround

      Currently there is no known workaround for this behavior. A workaround will be added here when available

            [JRASERVER-73032] Altering group membership has performance impact on large instances

            Maciej Swinarski (Inactive) made changes -
            Remote Link New: This issue links to "Page (Confluence)" [ 603802 ]
            Carlos Moya (Inactive) made changes -
            Remote Link New: This issue links to "Page (Confluence)" [ 601855 ]
            Roman Kolosovskiy (Inactive) made changes -
            Remote Link New: This issue links to "Page (Confluence)" [ 601853 ]
            Jakub Cegiel made changes -
            Status Original: Needs Triage [ 10030 ] New: Gathering Impact [ 12072 ]
            Andriy Yakovlev [Atlassian] made changes -
            Link New: This issue is related to JRASERVER-71231 [ JRASERVER-71231 ]
            Andriy Yakovlev [Atlassian] made changes -
            Link New: This issue is related to JRASERVER-64384 [ JRASERVER-64384 ]
            Maciej Swinarski (Inactive) made changes -
            Labels Original: jfr New: deltaUserLicense jfr
            Michelle Chin made changes -
            Description Original: h3. Issue Summary

            This issue comes from JFR data from JAC.
            Every time we change group membership, nodes in the cluster recalculate the current user count to check against the license. This process is asynchronous and deduplicated, but I believe it needs to be further improved.
            For JAC, I've seen it triggered 375 time in a timespan of 4 hours, with each call taking ~20s (that's a 2h total).
            It has been a problem in the past and is currently invisible in our synthetic testing.

            h3. Steps to Reproduce
             # Set up a breakpoint in com.atlassian.jira.application.DefaultApplicationRoleManager#clearUserCounts
             # add a user to the system

            h3. Expected Results
            # the method will not execute the recalc right away
            # the method will not be called on all the nodes

            h3. Actual Results
            # the recalc is initiated right away
            # each node recalculates data individually
            h3. Workaround

             Currently there is no known workaround for this behavior. A workaround will be added here when available             		New: h3. Issue Summary

            This issue comes from JFR data from JAC.
            Every time we change group membership, nodes in the cluster recalculate the current user count to check against the license. This process is asynchronous and deduplicated, but I believe it needs to be further improved.
            For JAC, I've seen it triggered 375 time in a timespan of 4 hours, with each call taking ~20s (that's a 2h total).
            It has been a problem in the past and is currently invisible in our synthetic testing.

            h3. Steps to Reproduce
             # Set up a breakpoint in com.atlassian.jira.application.DefaultApplicationRoleManager#clearUserCounts
             # add a user to the system

            h3. Expected Results
            # the method will not execute the recalc right away
            # the method will not be called on all the nodes

            h3. Actual Results
            # the recalc is initiated right away
            # each node recalculates data individually

            h3. Workaround

             Currently there is no known workaround for this behavior. A workaround will be added here when available
            Bugfix Automation Bot made changes -
            Introduced in Version New: 8.13
            Przemek Bruski made changes -
            Labels New: jfr

              Unassigned Unassigned
              pbruski Przemek Bruski
              Affected customers:
              3 This affects my team
              Watchers:
              8 Start watching this issue

                Created:
                Updated: