When it All Goes Wrong: Database Incident Response Dos and Don'ts

Once at Adjust we faced a problem of impending xid wraparound in a very central database due to a long-stalled autovacuum run. Because we spotted warnings 5 hours before impending disaster we were able to minimize the customer impact. Come learn how we used the time to prepare, and what lessons this has for others facing unusual problems in large databases.