Re: pgsql: Avoid archiving XLOG_RUNNING_XACTS on idle server

On 4 April 2016 at 07:49, Andres Freund <andres(at)anarazel(dot)de> wrote:

> > It doesn't?
>

Nope, clearly in the code 2 lines above.

For that matter, it's also important for hot standby; to deal with
> FATALed transactions which didn't write an abort record. It's kinda
> important to call StandbyReleaseOldLocks for those. And if a standby is
> in STANDBY_SNAPSHOT_READY it's also important to see this kind of
> record.
>

Those objections apply to all solutions to the problem so far proposed,
likely any solution. My understanding was that those issues are considered
the lesser of the two evils. I'm happy to revert this patch, as long as we
agree that it also blocks all further "bug fixes" so far proposed based
upon those arguments.

> Additionally, we're now logging more WAL if archive timeout isn't
> configured, which doesn't strike me as a good idea.
>

That's not true either...

Thanks for your comments, but you seem to be misreading the patch with
respect to the if clause and new brackets.

I suggest we give this some hours of thought before discussing it further.

--
Simon Riggs http://www.2ndQuadrant.com/
<http://www.2ndquadrant.com/>
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services