On Tue, Nov 30, 2010 at 8:38 AM, Greg Smith <greg(at)2ndquadrant(dot)com> wrote:
Thanks for the update.
> This might be ready for some proper review now. I know there's at least one
> blatant bug still in here I haven't found yet, related to how the averages
> are computed.
Yes, the blatant bug:
average_sync_time = CheckpointStats.ckpt_longest_sync /
That should clearly be ckpt_agg_sync_time, not ckpt_longest_sync.
> I saw this once:
> LOG: checkpoint complete: wrote 0 buffers (0.0%); 0 transaction log file(s)
> added, 0 removed, 1 recycled; write=0.000 s, sync=0.000 s, total=0.001 s;
> sync files=0, longest=0.000 s, average=-9223372036854775808.-2147483 s
> After an immediate checkpoint, so at least one path not quite right yet.
Not clear what the right thing to do here is. I guess we should
special case the div by zero to yield zero for the average?
The patch is in unified diff rather than context diff. Doesn't bother
me, but the wiki on doing a review says...
In response to
pgsql-hackers by date
|Next:||From: Andres Freund||Date: 2010-11-30 20:16:57|
|Subject: Re: DELETE with LIMIT (or my first hack)|
|Previous:||From: Josh Berkus||Date: 2010-11-30 20:09:31|
|Subject: Where are we on Standby Promotion?|