Tom Lane schrieb:
> Are you sure the master is checkpointing?
Yes. I double checked using pg_controldata on the master.
On both servers checkpoint_segments is set to 16 and checkpoint_timeout
is 300 seconds default.
There were two messages in the logs of the master that checkpointing
happened too often during bulk imports last night.
It's possible that the problem occures during periods if high database
activity where checkpointing on the master happens very often.
It would be ok if the slave skipped checkpoints if its falling behind
but i'd expect it to checkpoint again if it's in sync again.
I watched the servers a while after it happend today and found that the
slave catched up waiting for new logs. But it didnt do any checkpointing.
The last checkpoint was over 2 hours and about 500 log segments ago.
I upgraded the slave to 8.2.4-1PGDG today and will upgrade the master
too as soon as I can get me a maintenance window for this.
Greetings,
Frank Wittig