MySQL ALTER TABLE with subpartition fails - mysql

Question

I'm trying to alter an existing table to add year and week subpartitions, like so:
CREATE TABLE test_table(
id INT NOT NULL AUTO_INCREMENT PRIMARY KEY,
dtime DATETIME NOT NULL);
ALTER TABLE test_table
PARTITION BY RANGE ( YEAR(dtime) )
SUBPARTITION BY RANGE( WEEK(dtime) ) (
PARTITION y0 VALUES LESS THAN (2013) (
SUBPARTITION w0 VALUES LESS THAN (2),
...
SUBPARTITION w52 VALUES LESS THAN (54)
),
PARTITION y1 VALUES LESS THAN (2014) (
SUBPARTITION w0 VALUES LESS THAN (2),
...
SUBPARTITION w52 VALUES LESS THAN (54)
),
PARTITION y2 VALUES LESS THAN (2015) (
SUBPARTITION w0 VALUES LESS THAN (2),
...
SUBPARTITION w52 VALUES LESS THAN (54)
),
PARTITION y3 VALUES LESS THAN (2016) (
SUBPARTITION w0 VALUES LESS THAN (2),
...
SUBPARTITION w52 VALUES LESS THAN (54)
)
);
However, this gives me the vague and unhelpful response of:
ERROR 1064 (42000): You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'RANGE( WEEK(DTIME) ) (
PARTITION y0 VALUES LESS THAN (2013) (
SUBPARTITION ' at line 3
I've checked the docs: MySQL ALTER TABLE Partition operations and MySQL RANGE and LIST Partitions. However, neither of these describe how to alter a table to create subpartitions.
The second part of my question is for feedback on this partitioning scheme. The data that will go into this is sensor readings that are recorded every minute, and the most common query operation is for data in the last week. I think this should greatly speed up my queries, since a "WHERE dtime > date" is very common, without having to manually move data out of the table periodically into archive tables.

Answer 1

If you want to add a partition BY LIST to an already existing table, drop the primary key and create a composite primary key:
alter table test_table drop primary key, add primary key (id,<some other key>);
alter table orders partition by list(<some other key>) (
partition p0 values IN (1),
partition p1 values IN (2),
partition p2 values IN (3),
partition p3 values IN (4),
partition p4 values IN (5),
partition p5 values IN (6),
partition p6 values IN (7),
partition p7 values IN (8),
partition p8 values IN (9),
partition p9 values IN (10)
);

Answer 2

After further investigation, I have discovered several problems with this approach.
It is impossible to range partition on a DATETIME value (which dtime in the example is). http://dev.mysql.com/doc/refman/5.1/en/partitioning-limitations-functions.html
The table I was partitioning had a primary key on an auto increment id column, and you cannot partition on an index if there is a different primary key.
ERROR 1503 (HY000): A PRIMARY KEY must include all columns in the table's partitioning function
See also http://blog.mclaughlinsoftware.com/2011/05/09/mysqls-real-partition-key/
http://dev.mysql.com/doc/refman/5.1/en/partitioning-limitations-partitioning-keys-unique-keys.html
WEEK() is not allowed as a partitioning function. http://dev.mysql.com/doc/refman/5.1/en/partitioning-limitations-functions.html
From what I now know, if you have a UNIQUE AUTO_INCREMENT id as the primary key, it is impossible to partition on anything except that value.
My queries all use the dtime column in the WHERE conditions, so it seems that unless I can partition somehow on dtime still, there is no benefit to partitioning this table (from a performance perspective).