Re: Add support for logging the current role

* Itagaki Takahiro (itagaki(dot)takahiro(at)gmail(dot)com) wrote:
> On Mon, Feb 7, 2011 at 04:10, Stephen Frost <sfrost(at)snowman(dot)net> wrote:
> > Yeah, doesn't seem to work for me (missing '/bin/collateindex.pl',
> > apparently..).
>
> You might need "yum install openjade stylesheets" or similar packages
> and re-"configure".

I've got openjade, etc, installed, but I'm on Debian and it doesn't
appear to include that collateindex.pl anywhere..

> For implementation, write_csvlog() has many following lines:
> if (curr_field != num_fields) appendStringInfoChar(&buf, ',');
> It will be cleaner if we add first_col flag and move it out of
> the switch statement.

Done.

> Other questions I raised before might be matters of preference.
> I'd like to here about them form third person.
> * name: log_csv_fields vs. csvlog_fields

Done.

> * when to assign: PGC_POSTMASTER vs. PGC_SIGHUP

I'm still in the PGC_POSTMASTER camp on this and I really think it's a
more complicated change than just changing that value in the code, even
if we all agreed it should be allowed on SIGHUP (which certainly isn't
the case anyway..). In the end, if we really want that, we can always
add it in the future.

Updated patch attached, full git log below.

Thanks,

Stephen

commit 6bd2b9f1d2bc3b166a3e5598ee590e25159c61a5
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Fri Feb 11 11:16:17 2011 -0500

Rename log_csv_fields GUC to csvlog_fields

This patch renames the log_csv_fileds GUC to csvlog_fields, to better
match the other csvlog_* options.

Also cleaned up the CSV generation code a bit by moving the comma-adding
code out of the switch() statement.

commit a281ca611e6181339e92b488c815e0cb8c1298d2
Merge: d8dddd1 183d3cf
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Fri Feb 11 08:37:27 2011 -0500

Merge branch 'master' of git://git.postgresql.org/git/postgresql into log_csv_options

commit d8dddd1c425a4c320540769084ceeb7d23bc3662
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Sun Feb 6 14:02:05 2011 -0500

Change log_csv_options listing to a table

This patch changes the listing of field options available to
log_csv_options into a table, which will hopefully both look
better and be clearer.

commit f9851cdfaeb931f01c015f5651b72d16957c7114
Merge: 3e71e33 5ed45ac
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Sun Feb 6 13:26:17 2011 -0500

Merge branch 'master' of git://git.postgresql.org/git/postgresql into log_csv_options

commit 3e71e338a2b9352d730f59a989027e33d99bea50
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Fri Jan 28 22:44:33 2011 -0500

Cleanup log_csv_options patch

Clean up of various function declarations to hopefully be correct
and clean and matching PG conventions. Also move TopMemoryContext
usage to later, since the local variables don't need to be in
TopMemoryContext. Lastly, ensure that a comma is not produced
after the last CSV field, and that one is produced if
application_name is not the last field.

Review by Itagaki Takahiro, thanks!

commit 1825def11badd661d219fa4c516f06e0ad423443
Merge: ff249ae 847e8c7
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Wed Jan 19 06:50:03 2011 -0500

Merge branch 'master' of git://git.postgresql.org/git/postgresql into log_csv_options

commit ff249aeac7216da623bf77840380d5e767f681fc
Author: Stephen Frost <sfrost(at)snowman(dot)net>
Date: Wed Jan 19 00:26:52 2011 -0500

Add log_csv_fields GUC for CSV output & curr_role

This patch adds a new GUC called 'log_csv_fields'. This GUC allows
the user to control the set of fields written to the CSV output as
well as the order in which they are written. The default set of
fields remains those that were included in 9.0, to avoid breaking
existing user configurations.

In passing, update 'user_name' for log_line_prefix and log_csv_fields
to mean 'session user' (which could be reset by a superuser with
set session authorization), and add a 'role_name' option (%U) to
log_line_prefix and log_csv_fields, to allow users to log the
current role (as set by SET ROLE- not impacted by SECURITY DEFINER
functions).