Re: [COMMITTERS] pgsql: Bloom index contrib module

From: Noah Misch <noah(at)leadboat(dot)com>
To: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: Teodor Sigaev <teodor(at)sigaev(dot)ru>, pgsql-hackers(at)postgresql(dot)org
Subject: Re: [COMMITTERS] pgsql: Bloom index contrib module
Date: 2016-04-10 06:01:02
Message-ID: 20160410060102.GC1741844@tornado.leadboat.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-committers pgsql-hackers

On Sat, Apr 09, 2016 at 10:08:01PM -0400, Tom Lane wrote:
> I wrote:
> > I was depressed, though not entirely surprised, to find that you get
> > exactly that same line-count coverage if the table size is cut back
> > to ONE row.
>
> Oh, I found the flaw in my testing: there are two INSERTs in the test
> script and I was changing only one of them. After correcting that,
> the results behave a little more sanely:
>
> Line Coverage Functions
> 1 row: 70.4 % 349 / 496 93.1 % 27 / 29
> 10 row: 73.6 % 365 / 496 93.1 % 27 / 29
> 100 rows: 73.6 % 365 / 496 93.1 % 27 / 29
> 1000 rows: 75.4 % 374 / 496 93.1 % 27 / 29
>
> Still, we've reached the most coverage this test can give us at 1000
> rows, which still means it's wasting the last 99% of its runtime.

If dropping the row count to 1000 shaves >500ms on your primary machine, +1
for committing such a row count change. This is exactly what I meant by
"someone identifies a way to realize similar coverage with lower duration."
Thanks for contributing this study.

In response to

Responses

Browse pgsql-committers by date

  From Date Subject
Next Message Alexander Korotkov 2016-04-10 06:08:30 Re: [COMMITTERS] pgsql: Bloom index contrib module
Previous Message Tom Lane 2016-04-10 04:24:35 pgsql: Further minor improvement in generic_xlog.c: always say REGBUF_S

Browse pgsql-hackers by date

  From Date Subject
Next Message Alexander Korotkov 2016-04-10 06:03:37 Re: Move PinBuffer and UnpinBuffer to atomics
Previous Message Alexander Korotkov 2016-04-10 05:40:38 Re: Move PinBuffer and UnpinBuffer to atomics