Re: Sequences - jumped after power failure

Steve T <steve(at)retsol(dot)co(dot)uk> writes:
> I have a set of claims tables that cover the claim itself, the customer,
> contact points etc. Yesterday there was a power failure and the server
> suffered an immediate power outage. When the server came back,
> everything seemed fine, apart from the fact that the claim related
> sequences had all jumped and left a gap of 33 (last was 52 before power
> failure, next one allocated after power failure 85). This seems
> consistent across all the tables related to the claim (it may be across
> the tables in the database - I haven't checked all of them as yet).

> Does this sound feasible and if so, what is the cause?

Yeah, this is intentional behavior designed to reduce the amount of disk
write traffic generated by nextval()s. From a standing start, a
nextval() actually advances the sequence 33 times (1 + SEQ_LOG_VALS),
so that the next 32 nextval()s won't need to generate their own WAL
records. I guess you must have crashed before that first nextval()
was able to commit its result into the database ...

regards, tom lane