Re: SSI 2PC coverage

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: "Heikki Linnakangas" <heikki(dot)linnakangas(at)enterprisedb(dot)com>
Cc: "Dan Ports" <drkp(at)csail(dot)mit(dot)edu>,<pgsql-hackers(at)postgresql(dot)org>
Subject: Re: SSI 2PC coverage
Date: 2011-07-07 15:43:59
Message-ID: 4E158DEF020000250003F05D@gw.wicourts.gov
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Heikki Linnakangas <heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
> On 05.07.2011 20:06, Kevin Grittner wrote:

>> In reviewing the recent fix to 2PC coverage in SSI, I found some
>> cases which didn't seem to be covered. Dan bit the bullet and
>> came up with an additional isolation test to rigorously cover all
>> the permutations, to find *all* 2PC statement orderings which
>> weren't working right. Because it was so big, he pared out tests
>> which were redundant, in that they exercised the same code paths
>> and pointed at the same issues. A patch to add this test is
>> attached. Run against HEAD it shows the errors. It's kinda big,
>> but I think it's worth having.
>
> I agree it'd be very nice to have this test, but 2.3 MB of
> expected output is a bit excessive. Let's try to cut that down
> into something more palatable.

OK

> There's two expected output files for this, one for
> max_prepared_xacts=0 and another for the "normal" case. The
> max_prepared_xacts=0 case isn't very interesting, since all the
> PREPARE TRANSACTION commands fail. I think we should adjust the
> test harness to not run these tests at all if
> max_prepared_xacts=0. It would be better to skip the test and
> print out a notice pointing out that it was not run, it'll just
> give a false sense of security to run the test and report success,
> when it didn't test anything useful.
>
> That alone cuts the size of the expected output to about 1 MB.

OK. I'll work on this tonight unless Dan jumps in to claim it.

> That's much better, although it's still a lot of weight just for
> expected output. However, it compresses extremely well, to about
> 16 KB, so this isn't an issue for the size of distribution
> tarballs and such, only for git checkouts and on-disk size of
> extracted tarballs. I think that would be acceptable, although we
> could easily cut it a bit further if we want to. For example
> leaving out the word "step" from all the lines of executed test
> steps would cut it by about 80 KB.

That seems simple enough. Any other ideas?

>> Attached is also a patch to fix those, so that all permutations
>> work.
>
> Thanks, committed the bug fix with some additional comments.

Thanks!

-Kevin

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Heikki Linnakangas 2011-07-07 15:45:14 Re: SSI atomic commit
Previous Message Robert Haas 2011-07-07 15:43:30 Re: Make relation_openrv atomic wrt DDL