Re: HOw is in partitioning by range implemented the UPDATE statement ...

[Mattias Jonsson]

If you want to have `ID` unique (and enforced by the server) then you should have it in a primary key, or use a unique key (which will force you to include it in the partitioning function).

If the update is within the same partition, then there is no real overhead.
If the update moves the row from one partition to another, then it is the same as a INSERT + DELETE.
(for the details, look into sql/ha_partition.cc, ha_partition::update_row)

The record is only marked for deletion (it is really up to the storage engine).

It might cause fragmentation, if the 'deleted' hole is filled again and the row format is dynamic.