That is an odd design. PostgreSQL only holds the lock long enough to increment a counter in memory, and every 32th time also write a write-ahead log memory in R

That is an odd design. PostgreSQL only holds the lock long enough to increment a counter in memory, and every 32th time also write a write-ahead log memory in RAM. I can't see why one would need to lock the counter for the duration of the query.