-
Bug
-
Resolution: Obsolete
-
Medium
-
None
-
5.9.5
-
7
-
Severity 2 - Major
-
9
-
Summary
The Hazelcast 5801 port traffic between the nodes is much larger (about 50~200 times) in Confluence Data Center 5.9.x when browsing a page containing JIRA Issue Macro, comparing to Confluence 5.7.5. The more access the cluster gets, the more traffic is generated between the nodes.
Steps to Reproduce
- Install Confluence Data Center 5.9.5 cluster including 2 nodes
- Install JIRA Software 7.0.5
- Generate an Application Link between the JIRA and the CDC
- Create a page with JIRA Issue Macro in Confluence
- Access to the page by using jMeter or so forth (I used jMeter with 10 parallel threads and infinite loop count)
- Measure the traffic between the nodes by dstat command (specifically "dstat -n -N <interface>" command)
- You'll find the traffic is much larger (In my case, the traffic was about 50Mbps):
- much larger than the case when accessing a page without JIRA Issue Macro (In my case, the traffic was about 250Kbps)
- much larger than the case when accessing the same page in Confluence Data Center 5.7.5, which is the version that the customer is currently using (In my case, the traffic was about 250Kbps)
- As a result, CPU usage is also increased, though the CPU % does not grow disastrously severe
I confirmed the traffic was hazelcast 5801 port, by creating a separate network interface only for the hazelcast traffic. Another network interface is used for all other traffics such as HTTP, PostgreSQL, NFS.
Expected Results
The traffic is as high as it used to be in Confluence Data Center 5.7.5
Actual Results
The traffic is about 40~200 times higher than it used to be in Confluence Data Center 5.7.5
Workaround
There is no workaround found at the moment.
[CONFSERVER-41264] JIRA Issue Macro sends much higher hazelcast traffic between the Data Center nodes
Hi jens.kasperek1, since this ticket was initially raised, we have had no further reports of any such behaviour affecting other customers.
We have also not been able to re-produce this issue.
It is our belief that irrespective of whether their is an increased Hazelcast traffic when using a JIRA issues macro, that it is not having any noticeable impact on customers.
As such, at this point we do not plan to attempt to fix this issue. If we get further reports of this issue affecting customers then we will certainly look back into it.
Is there any hint if this issue has been fixed or not? This could be a show stopper for us.
One of our customers and myself confirmed that this problem did not happen on Confluence 5.10.0 Data Center - no high traffic and no CPU usage increase. The bug might have been fixed while some other bug fixes and/or features were released
Since this ticket has been open (over 1 year) it has not been reported or reproduced by any customers on Confluence 5.10 or above. Closing this issue as Obsolete.
Please comment (with spects to reproduce) if you are experiencing this issue in a later/supported version.