要約

Jira's cluster lock health check fails often for lock com.atlassian.jira.plugins.dvcs.service.RepositoryService.syncRepositoryList.

Node '<node-name>' has been holding cluster lock, 'com.atlassian.jira.plugins.dvcs.service.RepositoryService.syncRepositoryList', for <seconds> seconds.

This may lead to further index replication health check alerts as a consequence of this.

環境

All versions of Jira Software 7.x and 8.x.

診断

1. The health check alert on syncRepositoryList in Jira's health check page;

2. The query below results in a value of 3600000 or lower than 86400000:

   select interval_millis from clusteredjob where job_runner_key='com.atlassian.jira.plugins.dvcs.scheduler.DvcsScheduler';

原因

DVCS integration has two main strategies to update Jira issues:

1. Webhooks that are triggered upon each action in DVCS and update Jira right on time
2. A scheduled "sync-everything" job that serves as a fail over to the webhook, syncing everything from time to time.

The scheduled job is prone to performance issues when there are too many repos or communication delays, lading to the following issue:

JSWSERVER-14061 - Getting issue details... STATUS

ソリューション

The scheduled DCVS "sync-everything" job is default to run every hour — which may not give it enough time to complete.

A reasonable solution is to increase this schedule interval to more than 1 hour (3600000 millis). Please refer to this article on How to change the interval or schedule of the DVCS repositories sync for instructions on that.