This bug report describes a possible deadlock that may occur during repository deletion. This is likely specific to the Oracle RDBMS which has more specific risks around locks and unindexed foreign keys[1]

The thread stacks on the system that exhibited this problem showed many threads trying to acquire a DB connection, while 77 of the 80 DB connections were blocked executing statements. The threads holding connections were:

75 in HibernateWatcherDao.findByWatchableAndUser(HibernateWatcherDao.java:69)

The above were probably victims and not involved in the deadlock. But they explain why all DB connections are exhausted.

Then the two that are likely involved in the deadlock:

1 in DriftCommentUpdateProcessor.process(DriftCommentUpdateProcessor.java:190)

1 in HibernatePullRequestDao.deleteByToRepository(HibernatePullRequestDao.java:174)

[1] https://docs.oracle.com/cd/E11882_01/server.112/e40540/consist.htm#CNCPT1339