Re: How to handle secondary keys in partitioning?

[Mattias Jonsson]

Rick: the priority queue is used for index reads that requires sorted results, and the 'additional sorting cost' is the cost of that priority queue.

The calculation of which partition id a row belongs to is in RANGE/LIST a binary search O(log(num_partitions/list_values)), and in KEY/HASH a simple modulo operation O(1) and the partitioning expression, and is not related to the index. The index management is done in the storage engine used (MyISAM, InnoDB etc).

John: you are correct that it will still need to read from 5 indexes. My thought was if you want to partition by 25 partitions, maybe it would make sense to partition on two different columns, if the select queries is a mix of either of the two columns (recipient_id or sender_id). Also these matching subpartitions will be only a 1/5 of the size of the table. It all depends on what the usage are and where you want the gain.

So my idea to add subpartitioning was that you would make use of pruning for both recipient_id and sender_id, as a sort of compromise if you query a lot on sender_id too.