Uploaded image for project: 'Jira Data Center'
  1. Jira Data Center
  2. JRASERVER-41241

JIRA can run out of memory due to an unbounded cache in atlassian-gadgets

XMLWordPrintable

    • 6.03
    • 5
    • Severity 1 - Critical
    • 2
    • Hide
      Atlassian Update – 2-11-2018

      Hi everyone,

      We have recently reviewed this issue and the overall interest in the problem. As the issue hasn't collect votes, watchers, comments, or support cases from many customers during its lifetime, it's very low on our priority list, and will not be fixed in the foreseeable future, i.e. in next 6 months.

      Although we're aware the issue is still important to those of you who were involved in the conversations around it, we want to be clear in managing your expectations. The Jira team is focusing on issues that have broad impact and high value, reflected by the number of comments, votes, support cases, and customers interested. Please consult the Atlassian Bugfix Policy for more details.

      We understand how disappointing this decision may be, but we hope you'll appreciate our transparent approach and communication.

      Atlassian will keep the issue open and continue to watch this issue for the further updates.

      Thank you,
      Ignat Alexeyenko
      Jira Bugmaster

      Show
      Atlassian Update – 2-11-2018 Hi everyone, We have recently reviewed this issue and the overall interest in the problem. As the issue hasn't collect votes, watchers, comments, or support cases from many customers during its lifetime, it's very low on our priority list, and will not be fixed in the foreseeable future, i.e. in next 6 months. Although we're aware the issue is still important to those of you who were involved in the conversations around it, we want to be clear in managing your expectations. The Jira team is focusing on issues that have broad impact and high value, reflected by the number of comments, votes, support cases, and customers interested. Please consult the Atlassian Bugfix Policy for more details. We understand how disappointing this decision may be, but we hope you'll appreciate our transparent approach and communication. Atlassian will keep the issue open and continue to watch this issue for the further updates. Thank you, Ignat Alexeyenko Jira Bugmaster

      NOTE: This bug report is for JIRA Server. Using JIRA Cloud? See the corresponding bug report.

      The cache created in atlassian-shindig - DefaultHttpCache can cause JIRA to leak memory and eventually get an out of memory error (OOME).

      Furthermore, it seems that all caches created via AtlassianCacheProvider are an unlimited, non expirable caches that will grow and grow and so will suffer from this same problem.

        1. image-2019-11-20-20-02-18-112.png
          image-2019-11-20-20-02-18-112.png
          2.18 MB

            [JRASERVER-41241] JIRA can run out of memory due to an unbounded cache in atlassian-gadgets

            Alex [Atlassian,PSE] added a comment - - edited

            Was observed on 7.11.2

            In this case we had 4 com.google.common.cache.LocalCache$Segment , of ~ 500MB each

            with ~21,000 cached org.apache.shindig.gadgets.http.HttpResponse per segment. The content of this included a string with the actual response from the server. This caused the instance to be pushed into long full GC time then finally OOM state

             

            Alex [Atlassian,PSE] added a comment - - edited Was observed on 7.11.2 In this case we had 4 com.google.common.cache.LocalCache$Segment , of ~ 500MB each with ~21,000 cached org.apache.shindig.gadgets.http.HttpResponse per segment. The content of this included a string with the actual response from the server. This caused the instance to be pushed into long full GC time then finally OOM state  

            metrics added a comment -

            This is further compounded by the Bamboo gadgets using a nocache=$timestamp in the query, and the entire HttpResponse object being cached which is inefficient in what it holds in memory.

            metrics added a comment - This is further compounded by the Bamboo gadgets using a nocache=$timestamp in the query, and the entire HttpResponse object being cached which is inefficient in what it holds in memory.

            MattS added a comment -

            Thanks for the info

            MattS added a comment - Thanks for the info

            Pawel Niewiadomski (Inactive) added a comment -

            matt.doar basically the bug is related to all gadgets doing REST calls, but specifically to the Bamboo related ones.

            Pawel Niewiadomski (Inactive) added a comment - matt.doar basically the bug is related to all gadgets doing REST calls, but specifically to the Bamboo related ones.

            MattS added a comment -

            Which gadgets and how?

            MattS added a comment - Which gadgets and how?

              Unassigned Unassigned
              pniewiadomski Pawel Niewiadomski (Inactive)
              Affected customers:
              3 This affects my team
              Watchers:
              14 Start watching this issue

                Created:
                Updated: