On 2 February 2018 at 02:17, Michael Paquier <michael.paquier@gmail.com> wrote:
> On Fri, Feb 02, 2018 at 12:21:49AM +0000, Simon Riggs wrote:
>> Yes, it would be about 99% of the time.
>
> When it comes to recovery, I don't think that 99% is a guarantee
> sufficient. (Wondering about the maths behind such a number as well.)
>
>> But you have it backwards - we are not assuming that case. That is the
>> only case that has risk - the one where an old WAL record starts at
>> exactly the place the latest one stops. Otherwise the rest of the WAL
>> record will certainly fail the CRC check, since it will effectively
>> have random data in it, as you say.
>
> Your patch assumes that a single WAL segment recycling is fine to
> escape based on the file name, but you need to think beyond that. It
> seems to me that your assumption is wrong if the tail of a segment gets
> reused after more cycles than a single one, which could happen when
> doing recovery from an archive, where segments used could have junk in
> them. So you actually *increase* the odds of problems if a segment is
> forcibly switched and archived, then reused in recovery after being
> fetched from an archive.
This seems to be a pivotal point in your argument, yet it is just an assertion.
Please explain for the archive why you think the odds increase in the
way you describe.
Thanks
--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services