On 10/11/11 17:48, Tatsuo Ishii wrote:
No. It's just a matter of when 'shutdown host pgpool(5432)' message
was issued. Actual shutdown (set down status in shared memory area)
was done befoere follow command was executed.
I think real problem here is, when the private backend status (copy of
shared memory area, which pcp process looks into) gets
updated. Currently at the very end of failover process, pgpool main
sends SIGUSR1 to pcp process requesting a restart. So it might be
possible for pcp process to look up outdated status if the follow
command starts too early.
Attached is a patch to solve the problem. With the patch pcp process
refesh the copy of backend status whenever it's idle, rather than
after receiving SIGUSR1. Please try.
So far in testing, this patch appears to resolve the problem. Thanks much!
Any chance the follow_master_command could get an escape code indicating what
the new primary node has become?
--
Jeff Frost<[email protected]>
CTO, PostgreSQL Experts, Inc.
Phone: 1-888-PG-EXPRT x506
FAX: 415-762-5122
http://www.pgexperts.com/
_______________________________________________
Pgpool-general mailing list
[email protected]
http://pgfoundry.org/mailman/listinfo/pgpool-general
