A SELECT That Stopped a Factory

I ran one unverified SELECT against production and equipment at an overseas factory site stopped.

It was a single statement that scanned about 20,000 rows. I didn't check its execution plan, I didn't bound the row count, and I didn't think about where it took locks. It took a library cache lock. The pipeline behind the floor depends on strict transactional consistency, so once the lock sat there, the equipment downstream stopped.

For 15 to 20 minutes I looked everywhere except at myself. I checked the integration server, the upstream feed, the usual suspects. Then I found that the session holding the lock was mine, and I reported it.

The corrective is plain:

A large SELECT gets checked before production — execution plan, a bound on the rows it touches, where it takes locks.

There was no such step at the time.