A few days ago, I noticed the following in the postmaster log while
messing with a bug that led to a backend coredump:
[ report of backend sig11, then normal recovery, ending with ]
LOG: transaction ID wrap limit is 1073759685, limited by database "regression"
LOG: invalid server process ID -1
LOG: invalid server process ID -1
LOG: invalid server process ID -1
LOG: invalid server process ID -1
The "invalid server process ID" messages could only have come from
pgstat.c. Apparently something sent some pgstat messages before executing
InitBackendSharedInvalidationState, but I don't see where, and I've not
been able to reproduce it. Any ideas?
regards, tom lane