In 9.3 HEAD I am getting what seems to be spurious wrap-around shutdowns.
postgres=# SELECT datname, datfrozenxid, age(datfrozenxid) FROM pg_database;
datname | datfrozenxid | age
-----------+--------------+-----------
template1 | 2621759843 | 0
template0 | 2621759843 | 0
postgres | 2571759843 | 50000000
jjanes | 2437230921 | 184528922
postgres=# select txid_current();
ERROR: database is not accepting commands to avoid wraparound data loss in database "jjanes"
HINT: Stop the postmaster and use a standalone backend to vacuum that database.
You might also need to commit or roll back old prepared transactions.
184,528,922 is well short of 2 billion, so what is going on?
I thought maybe the ShmemVariableCache were not getting updated when vacuum finished, but if I restart the server (forcing shared memory to get rebuilt from disk) the condition continues.
I tried setting a breakpoint on SetTransactionIdLimit, but that seems to get executed on startup before the -W flag takes effect, so I can't find it.
Any tips on how to debug this? I figure the next step is running git bisect, but that is sure to be tedious.
I'm using a variant of the below to reach wraparound quicker, perhaps that is introducing a bug?
Cheers,
Jeff