Re: Using multi-row technique with COPY

From: Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us>
To: Simon Riggs <simon(at)2ndquadrant(dot)com>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: Using multi-row technique with COPY
Date: 2005-11-29 19:17:53
Message-ID: 200511291917.jATJHrM08049@candle.pha.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Simon Riggs wrote:
> As a further enhancement, I would also return to the NOLOGGING option
> for COPY. Previously we had said that COPY LOCK was the way to go -
> taking a full table lock to prevent concurrent inserts to a block from a
> COPY that didn't write WAL and another backend which wanted to write WAL
> about that block. With the above suggested all-inserts-at-once
> optimization, it would no longer be a requirement to lock the table.
> That means we can continue to take advantage of the ability to run
> multiple COPY loads into the same table. Avoiding writing WAL will
> further reduce CPU by about 15% and I/O by about 50%.
>
> I would also suggest that pgdump be changed to use the NOLOGGING option
> by default, with an option to work as previously.

For those who have been around, they know I dislike having options that
95% of our users desire not be the default behavior. I think the COPY
NOLOGGING idea falls in that category. I would like to explore if there
is a way to have COPY automatically do no logging where possible by
default.

First, I think NOLOGGING is probably the wrong keyword. I am thinking
SHARE/EXCLUSIVE is best because they are already keywords, and they
explain the effect of the flag on other applications, rather than the
LOGGING capability, which is invisible to applications.

I am thinking we would have COPY WITH [ [ EXCLUSIVE | SHARE ] [ LOCK ]] ...
EXCLUSIVE lock would be NOLOGGING, SHARE would do logging because other
applications could insert into the table at the same time (and do
UPDATES/DELETES of the inserted rows).

One idea for default behavior would be to use EXCLUSIVE when the table
is zero size. I think that would do pg_dump and most of the user cases,
and of course users could override the default by using a keyword. We
could emit a NOTICE if an an exclusive lock is used without an EXCLUSIVE
keyword. One problem I see is that there is no way to insure zero size
without a lock that blocks other writers. Is that reliable?

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tom Lane 2005-11-29 19:30:28 Re: Using multi-row technique with COPY
Previous Message Bruce Momjian 2005-11-29 18:55:01 Open Source management resource