Mysql – Why is MySQL order by performance poor within a single partition of a partitioned table

MySQLorder-bypartitioning

I have to store some sequence numbered data in MySQL. I have about 300,000 data items per day for about a 10 year span. Let's say the table structure is just sequence number (a big int) and data (a varchar). One very common query i'll be making is for all data for a single day, ordered by sequence number.

Storing the data in one table per day works really well. Querying all data from a table takes 0.8 seconds (which is acceptable), and adding order by sequence number takes it up to 1.0 seconds (still acceptable), but this does result in a massive number of tables that i'd rather have less of.

Storing in one table per month (and adding an indexed day number field), the time to get a day's data goes up to 1.6 seconds (not great), but adding partitioning by day number brings it right back down to 0.8.

However… when I add the order by sequence number to the partitioned table query, the time goes up to 2.5 seconds! Does this make sense? I would have thought that since my where clause (day number = X) instantly limits to all the data in a single partition, it should then be about the same performance as one of my original day tables (which it is, until I add the order by clause, then everything goes to hell).

It's like it is doing the order by before checking the partitions or something. Anyone got any ideas? (either "yes, that'll happen, and here's why", or "here's what you need to do to speed it up").

Thanks.

Best Answer

Assuming that you range partition your table by the (auto incrementing) sequence and you limit your query by another value (day number), you might see, that a lot of partitions will be touched when you use:

EXPLAIN PARTITIONS SELECT * FROM table_name WHERE day_number=X ORDER BY...

The reason for that is, that there is no linkage between the partitioning attribute and the data filter. Since MySQL uses a different index for each partition, it needs to open all partitions to find all data matching day_number.

If you know the partition name, maybe by querying INFORMATION_SCHEMA.PARTITIONS before the real select, and you use MySQL-5.6+, you can use partition selection like

SELECT * FROM table_name PARTITION(partition_name) WHERE day_number=X ORDER BY...

But this is just chitchat based on a lot of assumtions. To give you a real answer, you need to post the query plan and more information about table structure - like Phil said before.

Related Solutions

Mysql – How to optimize this theSQL table that will need to hold 5+ million rows

Since the table is InnoDB

Bulk Insert Buffer (bulk_insert_buffer_size) is no good for you because it handles bulk inserts for MyISAM tables only
DISABLE KEYS / ENABLE KEYS only works for MyISAM. InnoDB handles secondary index processing in the system tablespace ibdata1.
- InnoDB will throw a warning about this
- See my post : https://stackoverflow.com/a/9525780/491757
- See the bug report : http://bugs.mysql.com/bug.php?id=5187

You may have to resort to altering the InnoDB buffer protocol to handle only INSERTs : Mysql load from infile stuck waiting on hard drive

Unless you plan to have store values in MLSNUMBER bigger than 2147483647, you may want to consider making MLSNUMBER an INT (4 bytes) instead of BIGINT (8 bytes) to save space on secondary index creation. If your values for MLSNUMBER are less than 4294967296, maybe MLSNUMBER should be INT UNSIGNED.

Mysql – How to partition a table by timestamp, where data from every i’th day goes to partition i

You can't use FROM_UNIXTIME() because hash partitions must be based on an integer expression. But assuming your timestamp is stored as an integer, you can use DIV to return an integer.

Here's a demo of partitioning in the way you describe:

mysql> create table table_to_partition (
  my_timestamp int unsigned primary key
) partition by hash(my_timestamp DIV (60*60*24)) partitions 3;

mysql> insert into table_to_partition values (unix_timestamp(now()));
mysql> insert into table_to_partition values (unix_timestamp(now()-interval 1 day));
mysql> insert into table_to_partition values (unix_timestamp(now()-interval 2 day));
mysql> insert into table_to_partition values (unix_timestamp(now()-interval 3 day));
mysql> insert into table_to_partition values (unix_timestamp(now()-interval 4 day));
mysql> insert into table_to_partition values (unix_timestamp(now()-interval 5 day));

mysql> select table_name, partition_name, table_rows 
from information_schema.partitions where table_name='table_to_partition';

+--------------------+----------------+------------+
| table_name         | partition_name | table_rows |
+--------------------+----------------+------------+
| table_to_partition | p0             |          2 |
| table_to_partition | p1             |          2 |
| table_to_partition | p2             |          2 |
+--------------------+----------------+------------+

Best Answer

Related Solutions

Mysql – How to optimize this theSQL table that will need to hold 5+ million rows

Mysql – How to partition a table by timestamp, where data from every i’th day goes to partition i

Related Question