On Fri, Jul 18, 2008 at 1:00 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> While we are griping about readability: failure to update the comments
> to match the code is NOT, NOT, NOT acceptable. I had barely started
> to read the patch before encountering this insult to the reader:
> ...
As this is Xiao's first patch to the community, I'd appreciate it if
you guys would chill a little. I'm fully aware of what needs to be
done with it clean-up wise, but given that we're under some time
constraints, I asked that he submit his preliminary patch for those
who wanted to preview/test it.
Again, this patch is nowhere near acceptance quality, nor was it
intended to be. I'll make sure he gets the next one cleaned up prior
to submitting it.
The real question now has been listed above; why are hash index
queries, including this patch, no better than b-tree even for straight
equality comparisons? Because hash is lossy, we could easily be
performing multiple I/Os for recheck, and that may be causing some of
the performance problems. I haven't checked I/O for that yet, but was
wondering if you (Tom) knew of any major design/implementation
deficiency we should be taking into consideration regarding PG's hash
index implementation?
--
Jonah H. Harris, Sr. Software Architect | phone: 732.331.1324
EnterpriseDB Corporation | fax: 732.331.1301
499 Thornall Street, 2nd Floor | jonah.harris@enterprisedb.com
Edison, NJ 08837 | http://www.enterprisedb.com/