If a big number of Confluence pages have been generated in a short period of time, it can also create a bunch of unread notifications sitting in the DB. When the system tries to fetch those notifications for a user, it doesn't apply any pagination and can fail with OOM. This has been reproduced during load testing scenario (with 100K notifications in AO_9412A1_AONOTIFICATION table).

The problematic method is AONotificationDao#findAllUnread in myworkday plugin. Solution to the problem is to apply pagination to workday plugin in Confluence.

Workaround

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