Re: Write Ahead Logging for Hash Indexes

On Fri, Sep 9, 2016 at 12:39 AM, Jeff Janes <jeff.janes@gmail.com> wrote:
>
> I plan to do testing using my own testing harness after changing it to
> insert a lot of dummy tuples (ones with negative values in the pseudo-pk
> column, which are never queried by the core part of the harness) and
> deleting them at random intervals.  I think that none of pgbench's built in
> tests are likely to give the bucket splitting and squeezing code very much
> exercise.
>

Hash index tests [1] written by Mithun does cover part of that code
and we have done testing by extending those tests to cover splitting
and squeezing part of code.

> Is there a way to gather statistics on how many of each type of WAL record
> are actually getting sent over the replication link?  The only way I can
> think of is to turn on wal archving as well as replication, then using
> pg_xlogdump to gather the stats.
>

Sounds sensible, but what do you want to know by getting the number of
each type of WAL records?  I understand it is important to cover all
the WAL records for hash index (and I think Ashutosh has done that
during his tests [2]), but may be sending multiple times same record
could further strengthen the validation.

> I've run my original test for a while now and have not seen any problems.
> But I realized I forgot to compile with enable-casserts, to I will have to
> redo it to make sure the assertion failures have been fixed.  In my original
> testing I did very rarely get a deadlock (or some kind of hang), and I
> haven't seen that again so far.  It was probably the same source as the one
> Mark observed, and so the same fix.
>

Thanks for the verification.


[1] - https://commitfest.postgresql.org/10/716/
[2] - https://www.postgresql.org/message-id/CAE9k0PkPumi4iWFuD%2BjHHkpcxn531%3DDJ8uH0dctsvF%2BdaZY6yQ%40mail.gmail.com
-- 
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com