On Fri, Jun 23, 2017 at 3:25 PM, Michael Paquier
<michael(dot)paquier(at)gmail(dot)com> wrote:
> On Thu, Jun 22, 2017 at 6:10 AM, Andres Freund <andres(at)anarazel(dot)de> wrote:
>> We've not heard any complaints about this afaik, but it's not something
>> that's easily diagnosable as being a problem. Therefore I suspect we
>> should fix and backpatch this?
>
> Agreed. I have just poked at this problem and have finished with the
> attached. Logically it is not complicated at the argument values used
> by the callers of RemoveXlogFile() are never updated when scanning
> pg_wal. Surely this patch needs an extra pair of eyes.
Andres has pointed me out offline that a bad copy-paste re-introduced
fb886c15. So it is important to properly track the segment name of the
switch point as well as the the segment from where the recycling
begins. There was as well a second mistake in RemoveOldXlogFiles()
causing unnecessary segments to be kept caused by the same copy-pasto.
--
Michael