The following bug has been logged on the website:
Bug reference: 8225
Logged by: Jeff Frost
Email address: jeff@pgexperts.com
PostgreSQL version: 9.1.8
Operating system: various
Description: =
I've seen this a few times on client servers but still can't seem to boil it
down to a nice reproducible test case. =
What happens is that we change various logging options in postgresql.conf,
then reload, and every so often, the settings don't seem to take effect even
though they are logged as being changed.
Here's an example from a recent logging run:
grep -hi checkpoint *.csv
2013-05-31 12:58:33.051 MDT,,,5398,,5189c940.1516,8,,2013-05-07 21:40:48
MDT,,0,LOG,00000,"parameter ""log_checkpoints"" changed to
""on""",,,,,,,,,""
2013-05-31 15:18:36.131 MDT,,,5398,,5189c940.1516,22,,2013-05-07 21:40:48
MDT,,0,LOG,00000,"parameter ""log_checkpoints"" removed from configuration
file, reset to default",,,,,,,,,""
So, note that log_checkpoints was logged as being changed, then 2hrs 20mins
later it's logged as being set back to default, but there's not a single
checkpoint logged in that time period. I checked and checkpoint_timeout is
indeed set to 300s. =
Also, in addition, we change the log_filename to datacollection-%H and while
that also shows up in the log during the reload:
2013-05-31 12:58:33.051 MDT,,,5398,,5189c940.1516,9,,2013-05-07 21:40:48
MDT,,0,LOG,00000,"parameter ""log_filename"" changed to
""datacollection-%H""",,,,,,,,,""
and the logfile is created, postgresql continues to log to the old filename
for the duration of the logging run.
I've seen this on recent 9.0, 9.1 and 9.2 versions. This particular run was
on 9.1.8.