On Fri, Jan 25, 2013 at 2:01 PM, Peter Eisentraut <peter_e@gmx.net> wrote:
> On 1/25/13 1:35 AM, Kevin Grittner wrote:
>> Peter, do you have a version that works with 9.3?
>
> I don't, but it shouldn't be hard for someone more up to date with the
> internal WAL addressing changes.
I've been thinking about that. Should pg_clearxlogtail for 9.3 only
work for logs from 9.3, or should it be arranged to work on logs from
older versions as well (which probably means parts of the 9.2 headers
would need to be copied into the body of pg_clearxlogtail)?
>> Given that we're past the start of the last CF, do we still want to
>> consider including it? If you tilt your head at just the right
>> angle, I guess it could look like a bug fix -- but it is a stretch.
>
> Given that clearxlog doesn't actually work in 9.3 right now (see above)
> and doesn't have any documentation, I think the stretch is a bit too big.
>
> I do like the idea behind it, but the first step would be to address
> these two points and let people try it out. Note also that in our
> discussion we had some questions about how exactly to write the archive
> command to use it.
This is what I use:
archive_command = 'set -C -o pipefail;
(/usr/local/pgsql9_2/bin/pg_clearxlogtail < %p || cat %p) | gzip >
/mnt/foo/archivelog9_2/%f.gz'
The "|| cat %p" business is because pg_clearxlogtail will fail on
.history files and so needs an alternative path.
I think it would be better to have pg_clearxlogtail instead just
silently pass any unrecognized files through without manipulation,
rather than erroring out.
Most people would probably want to use a more explicit shell script
with notifications and such, like one Kevin posted to general a few
months ago under "checkpoint_timeout and archive_timeout".
Cheers,
Jeff
