From: | Matthew Wakeling <matthew(at)flymine(dot)org> |
---|---|
To: | Sergei Politov <spolitov(at)gmail(dot)com> |
Cc: | pgsql-performance(at)postgresql(dot)org |
Subject: | Re: Most effective insert or replace |
Date: | 2009-07-03 11:20:50 |
Message-ID: | alpine.DEB.2.00.0907031209540.16183@aragorn.flymine.org |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
On Fri, 3 Jul 2009, Sergei Politov wrote:
> As far as I recall postgres does not have built-in support for "insert or replace" feature.
> Please comment these ways and propose effective ways to simulate "insert or replace" behavior.
> Also in may case I'm making a lot of inserts in a batch.
A few years ago I researched this, and came up with the following method
as seeming the fastest:
BEGIN;
DELETE FROM table WHERE id IN (big long list);
COPY table FROM STDIN BINARY;
COMMIT;
However, our circumstances may not be the same as yours for the following
reasons:
1. We are updating whole rows indexed by primary key, not just a single
field in each row.
2. We are able to use the COPY command - indeed we wrote a fair amount of
Java to enable batching, background writing, and COPY support. See
http://www.flymine.org/api/index.html?org/intermine/sql/writebatch/Batch.html
and http://www.intermine.org/
3. HOT has been invented since then, and it won't play well with this
method.
Matthew
--
Trying to write a program that can't be written is... well, it can be an
enormous amount of fun! -- Computer Science Lecturer
From | Date | Subject | |
---|---|---|---|
Next Message | Віталій Тимчишин | 2009-07-03 11:22:35 | Re: - Slow Query |
Previous Message | Sergei Politov | 2009-07-03 11:06:13 | Most effective insert or replace |