Re: Bug #613: Sequence values fall back to previously checkpointed

Bruce Momjian <pgman@candle.pha.pa.us> writes:
> bgrimm@zaeon.com wrote:
>> the sequence again.  Kill -9 the postmaster.  Tried to insert into the 
>> table, but to no avail... duplicate key.  currval of the sequence and 
>> it matched the value right after the checkpoint.  I've been able to 
>> duplicate that scenario several times.

> I just tested it here by doing a similar test of several nextval()
> calls, but then doing an INSERT and kill, and on restart, the sequence
> counter did have the proper value.

There have been prior reports of similar problems --- all quite
unrepeatable in my testing, and despite considerable study of the source
code I can't see how it could happen.  A reproducible test case would be
a tremendous help.

			regards, tom lane