WAL logging of hash indexes

"John Smith" <sodgodofall@gmail.com> writes:

Can someone clarify if WAL-logging of hash indexes will be supported
in the 8.4 release?

There is absolutely 0 value in tackling that until someone can fix
hash's performance problems. If there is no real-world scenario for
using it ... which there really isn't ... then adding WAL support
still leaves you with no real-world scenario for using it.

This is not to suggest that I wouldn't like to see all of that fixed;
I would. But let's concentrate on the showstoppers first, rather than
expending effort that might ultimately be a waste.

regards, tom lane

There is absolutely 0 value in tackling that until someone can fix
hash's performance problems. If there is no real-world scenario for
using it ... which there really isn't ... then adding WAL support
still leaves you with no real-world scenario for using it.

This is not to suggest that I wouldn't like to see all of that fixed;
I would. But let's concentrate on the showstoppers first, rather than
expending effort that might ultimately be a waste.

I agree on that. I think working on Hash clusters would add more use-cases

than working on hash-indexes.

Thanks,
Gokul.

On Tue, Jan 15, 2008 at 07:18:17PM -0800, John Smith wrote:

http://archives.postgresql.org/pgsql-hackers/2007-10/msg01468.php,

"... I very much want to encourage authors of new Resource Managers and it
looks like we may be getting at least 3 new RMs that produce WAL
records: hash indexes (currently not WAL-logged), bitmap indexes and
clustered indexes for 8.4. We should be realistic that new bugs probably
will occur in recovery code for existing and new RMs."

Can someone clarify if WAL-logging of hash indexes will be supported
in the 8.4 release?

- John

John,

You can check the mailing list archives to follow recent discussions
about addressing the performance problems of hash indexes in PostgreSQL.
I think that we have some good ideas that may pan out. If the performance
problems can be fixed, it is my goal to have support for WAL logged,
unique hash indexes ready for 8.4. But as the thread describes, we need
the performance first or doing the rest makes no sense if btree indexes
work better.

Cheers,
Ken