Detected after this week's release on Confluence Cloud, whenever a customer ran pdf export, the connection pool used is more than 10 connections which made it fairly easy to run into connection pool exhaustion and the instance is not available.

Steps for replication

Run a pdf export for a space, preferable a larger space as it shows the issue better.

Expected Result

The import shows in a page similar to indexing, where it indicates the progress. And manage the resources accordingly.

Actual Result

The page seems to be loading, until the timeout at 60 seconds, the page unavailable is shown. Then checking on the connections to db, noticed that approximately 13~16 connections is being used despite the instance has no activity.

Workaround

There is no workaround at this moment. Make sure that only one export is requested at a time, across the entire instance. If it reaches time-out, do not attempt export again.

Solution

• We still acquire semaphore
• Instead of acquiring indefinitely timeout set to 30 seconds (controller via system property)