Re: Logging checkpoints and other slowdown causes

Here's a work-in-progress updated version of this patch. It's not
completed, but I've settled on:

- a log_checkpoints GUC variable to control whether anything is printed
or not. This is in line with log_connections, log_autovacuum etc.

- when checkpoints are logged, LOG-level is used

- the following log output:

LOG: checkpoint starting (CHECKPOINT command)
LOG: checkpoint complete; buffers written=3.1 MB (9.6%) write=96.8 ms
sync=32.0 ms

where "checkpoint starting" is obviously printed when checkpoint starts,
and "checkpoint complete" when the checkpoint is fully complete.

- all information on the progress of checkpoint is collected to a global
struct. A global struct sounds a bit ugly, but it avoids the ugliness of
littering all functions with multiple return values to pass back
information to the caller that's only used to print it out. This makes
it easy to add more data points. MoveOfflineLogs should probably be
changed to use that as well for the sake of consistency, instead of
returning nsegsremoved and nsegsrecycled through pointers.

- RequestCheckpoint and CreateCheckpoint was changed to take a
CheckpointReason enum as argument, instead of boolean flags. This allows
printing a different message depending on the cause of checkpoint, and
is also cleaner IMHO.

I'm posting now to allow others, and Greg Smith in particular, to review
the changes this far, and take over from here.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com