"Andrew Bartley" <abartley@evolvosystems.com> writes:
> 2002-03-22 16:44:28 [383] DEBUG: SIInsertDataEntry: table is 70% full,
> signaling postmaster
> 2002-03-22 16:44:28 [383] NOTICE: RegisterSharedInvalid: SI buffer
> overflow
> 2002-03-22 16:44:28 [315] NOTICE: InvalidateSharedInvalid: cache state
> reset
AFAIK these notices are completely harmless (and they have in fact been
downgraded to DEBUG level in 7.2). SI buffer overrun is an expected
condition when a large number of system catalog updates are committed
at once.
> In this circumstance it is causing a peer process job that useses
> persistant connections to receive a SIGPIPE signal.
I think the SIGPIPE has little or no direct connection to the SI buffer
overruns --- what *exactly* are you doing and what are you seeing?
regards, tom lane