Horton

Re: Cutting initdb's runtime (Perl question embedded)

Andreas Karlsson <andreas@proxel.se>

View thread View on postgr.es
From: Andreas Karlsson <andreas@proxel.se>
To: Tom Lane <tgl@sss.pgh.pa.us>, Andres Freund <andres@anarazel.de>
Cc: pgsql-hackers@postgresql.org
Date: 2017-04-15T01:44:18Z
Lists: pgsql-hackers 
On 04/14/2017 11:54 PM, Tom Lane wrote:
> I failed to resist the temptation to poke at this, and found that
> indeed nothing seems to break if we just use one transaction for the
> whole processing of postgres.bki.  So I've pushed a patch that does
> that.  We're definitely down to the point where worrying about the
> speed of bootstrap mode, per se, is useless; the other steps in
> initdb visibly take a lot more time.

Looked some at this and what take time now for me seems to mainly be 
these four things (out of a total runtime of 560 ms).

1. setup_conversion:        140 ms
2. select_default_timezone:  90 ms
3. bootstrap_template1:      80 ms
4. setup_schema:             65 ms

These four take up about two thirds of the total runtime, so it seems 
likely that we may still have relatively low hanging fruit (but not 
worth committing for PostgreSQL 10).

I have not done profiling of these functions yet, so am not sure how 
they best would be fixed but maybe setup_conversion could be converted 
into bki entries to speed it up.

Andreas

Commits

  1. Attempt to identify system timezone by reading /etc/localtime symlink.

  2. Make sure to run one initdb TAP test with no TZ set

  3. Use one transaction while reading postgres.bki, not one per line.

  4. Move bootstrap-time lookup of regproc OIDs into genbki.pl.