Re: pg_xlogdump fails to handle WAL file with multi-page XLP_FIRST_IS_CONTRECORD data

On Thu, Mar 31, 2016 at 6:27 AM, Craig Ringer <craig(at)2ndquadrant(dot)com> wrote:
>
>
> Can you describe the process used to generate the sample WAL segment?
>
>
Shame that I can't find the sql file used to create the problematic WAL
segment. But this is what I did.

I wrote a plpgsql function which inserts rows in a loop, every time
checking if the remaining space in the WAL segment has fallen to less than
couple of hundred bytes. Of course, I used pg_settings to get the WAL
segment size, WAL page size and pg_current_xlog_insert_location() to
correctly compute remaining bytes in the WAL segment. At this point, do a
non-HOT update, preferably to table which is already fully vacuumed and
CHECKPOINTed to avoid getting any other WAL records in between. Assuming
FPW is turned ON, the UPDATE should generate enough WAL to cross over the
first page in the new WAL segment.

Let me know if you would like to me to put together a sql based on this
description.

Thanks,
Pavan

--
Pavan Deolasee http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services