Comments on: The value of multi insert values

By: Olly

Olly — Fri, 03 Jul 2009 02:11:00 +0000

I made extensive use of these bulk inserts, and combined them with “ON DUPLICATE KEY UPDATE `columnname` = VALUES(columnname)”.
You get really huge performance improvements from that.

By: Shlomi Noach

Shlomi Noach — Wed, 17 Jun 2009 04:36:22 +0000

Combined with INSERT IGNORE, the multi-insert feature allows for a really nice peoperty:
You can insert 100 rows together, and those that fail a constraint (FOREIGN, UNIQUE), don’t get inserted, now errors thrown, while all the rest get inserted just fine.

By: Brooks Johnson

Brooks Johnson — Wed, 17 Jun 2009 00:57:58 +0000

The test conditions aren’t exactly clear to me. Were you sending the single inserts as a batch or one by one? If as a batch, was autocommit on or were you committing all the inserts as one?

By: Anders Karlsson

Anders Karlsson — Tue, 16 Jun 2009 23:02:27 +0000

Rereading what I wrote before, I realize my typing was worse than usual. Sorry anout that.

By: Ryan

Ryan — Tue, 16 Jun 2009 22:43:35 +0000

If you’re using InnoDB in autocommit mode, you will also see a very large improvement from the multi/batched insert statement due to fewer commits. Even moreso if you use innodb flush with setting 1.

By: Anders Karlsson

Anders Karlsson — Tue, 16 Jun 2009 22:28:29 +0000

On the other hand, the ay Oracle does this with the array intercafe is even more performant (although requires a bit more programming). The Oracle C API (Pro*C or OCI) is not the easiest to get along with, but once you know them, they can kick butt. I know, I used to teach people how to use them.

By: Roland Bouman

Roland Bouman — Tue, 16 Jun 2009 22:25:36 +0000

Yes – very useful feature.

There are a few popular myths around this feature – some people still think this is MySQL-specific (it is not – Postgres supports it in v >= 8.2; MS SQL Server supports it in v 2008), and some people think it is non-standard (it is in the 2003 version of the SQL standard)

Anyway, for Oracle, I sometimes generate scripts like this:

INSERT INTO t (col1,…,colX)
SELECT v1,…,vX FROM DUAL UNION ALL
SELECT v1,…,vX FROM DUAL UNION ALL
…
SELECT v1,…,vX FROM DUAL
;

clunky, but it does allow you to throw the data in there in one statement.

By: Mark Callaghan

Mark Callaghan — Tue, 16 Jun 2009 21:49:41 +0000

MySQL has a few things that seem simple at first glance but make it very efficient. This is one. Having the client-server protocol push many rows at a time back to the client (rather than the client pull 1 row at a time) is another.