Re: Reduce pinning in btree indexes

Kevin Grittner <kgrittn(at)ymail(dot)com> wrote:

> It doesn't seem worth posting to the list for the small changes
> since the last version; I'll wait until I update the comments and
> README files. If you want review or test the latest, you can
> peek at:
>
> https://github.com/kgrittn/postgres/tree/btnopin

Here is v3, with the promised README and code comment changes.

In the process of documenting the mark/restore area I noticed a
subtlety that I had missed (in the case that there was a mark,
advance to the next page, restore, advance within the page, and
restore). I fixed that, and in the process gave the patched code
an additional direct performance edge over unpatched code. For the
1000k marks, average timings are now:

master: 970.999 ms, stdev: 4.043
patched: 940.460 ms, stdev: 4.874

So, in the micro-benchmark showing the biggest benefit the direct
improvement is now just over 3%. It remains the case that the
primary motivation for the patch is to reduce blocking of vacuum
processes; but that's a nice side-benefit.

--
Kevin Grittner
EDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company