Home/ Questions/Q 8556577
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-06-11T15:23:04+00:00

I have insert query of this form: on INSERT into abc values …. on

  • 0

I have insert query of this form:

on INSERT into abc values .... on DUPLICATE KEY UPDATE ....

In my original table without partitions, the unique key was on (subject_id, object_id) but now I am partitioning by ts and I am forced to have the unique key as (subject_id, object_id, ts) because of mysql requirement of having the partition key in all the unique keys.

This means my insert query will fail coz an insert with same subject_id and object_id (from some previous insert) but a different timestamp will be treated as a unique new row and will get inserted instead of the desired update.

Is there any workaround?

CREATE TABLE `abc` (
  `id` bigint(20) NOT NULL AUTO_INCREMENT,
  `subject_id` varchar(40) COLLATE utf8_unicode_ci NOT NULL,
  `object_id` varchar(36) COLLATE utf8_unicode_ci NOT NULL,
  `ts` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
  PRIMARY KEY (`id`,`ts`),
  UNIQUE KEY `userint_sub_type_obj` (`subject_id`,`object_id`, `ts`)
) ENGINE=InnoDB AUTO_INCREMENT=2 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci
/*!50100 PARTITION BY RANGE (unix_timestamp(ts))
(PARTITION p2012_08_27 VALUES LESS THAN (1346025600) ENGINE = InnoDB,
 PARTITION p2012_08_28 VALUES LESS THAN (1346112000) ENGINE = InnoDB,
 PARTITION p2012_08_29 VALUES LESS THAN (1346198400) ENGINE = InnoDB,
 PARTITION p2012_08_30 VALUES LESS THAN (1346284800) ENGINE = InnoDB,
 PARTITION p2012_08_31 VALUES LESS THAN (1346371200) ENGINE = InnoDB) */
Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    To put it shortly, there is no way that I know of to add a partition without making the partition key a part of all your unique (and therefore also primary) keys.

    http://dev.mysql.com/doc/refman/5.5/en/partitioning-limitations-partitioning-keys-unique-keys.html

    I actually had this same problem last week trying to drop a partition on an enabled/disabled column. There’s two solutions that came to mind.

    1. You can move the timestamp into a meta_info table and partition that. Your look ups that have a time condition can then utilize the partition without messing with your (subject_id, object_id) key. The overhead is of course you’ll need to do a join. If your table is huge, this may not be viable.
    2. Do a check before you update. Instead of using the ON DUPLICATE KEY option, you split the action into a SELECT to see if the record exists, then either an INSERT or an UPDATE depending on the answer to that.

    Needless to say I’ve found this restriction of partitioning to be a huge pain in the rear, but it does appear to be a limitation in MySQL and after some Google Fu and asking around, it seems like there is simple way around it.

    • 0