Race condition during maven-metadata.xml deployment can cause invalid blob references and 500 response for subsequent GET requests

A race condition exists when two threads simultaneously make updates to a GA level maven-metadata.xml file.

The is is the scenario:

1. A GET request triggers a GA level maven-metadata.xml rebuild due to an earlier deployment, and the GA level file is deleted and recreated.
2. Later, a new GET request for the new GA level maven-metadata.xml file comes in.
3. At the same time, a PUT request for the GA level maven-metadata.xml file occurs.

The second GET request triggers an update for the last download time.
The PUT request triggers an update for the maven-metadata.xml file.

This causes a race condition, and sometimes the wrong thread wins. When this occurs, the next GET request for the GA level maven-metadata.xml file will fail with a 500 response. The logs show that the blob referenced by the database is in soft deleted state.

Workaround

Repeating the failed build and or GET request in a way that avoids the race condition.

Expected

Implement a change that prevents a GET request for the GA level maven-metadata.xml file failing with a 500 response code for the reproduce scenario.