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>, pgsql-hackers@postgreSQL.org, Andres Freund <andres@anarazel.de>
Date: 2017-04-12T19:15:53Z
Lists: pgsql-hackers

Attachments

On 04/12/2017 05:00 PM, Andreas Karlsson wrote:
> Looked at this an option 1 seems simple enough if I am not missing
> something. I might hack something up later tonight. Either way I think
> this improvement can be done separately from the proposed replacement of
> the catalog header files. Trying to fix everything at once often leads
> to nothing being fixed at all.

Here is my proof of concept patch. It does basically the same thing as 
Andres's patch except that it handles quoted values a bit better and 
does not try to support anything other than the regproc type.

The patch speeds up initdb without fsync from 0.80 seconds to 0.55 
seconds, which is a nice speedup, while adding a negligible amount of 
extra work on compilation.

Two things which could be improved in this patch if people deem it 
important:

- Refactor the code to be more generic, like Andres patch, so it is 
easier to add lookups for other data types.

- Write something closer to a real .bki parser to actually understand 
quoted values and _null_ when parsing the data. Right now this patch 
only splits each row into its fields (while being aware of quotes) but 
does not attempt to remove quotes. The PoC patch treats "foo" and foo as 
different.

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.