Percona Server with XtraDB

Adaptive hash index memory is incorrectly calculated in SHOW ENGINE INNODB STATUS and I_S

Reported by Alexey Kopytov on 2013-08-29
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Percona Server
Status tracked in 5.6
5.1
Undecided
Unassigned
5.5
Medium
Alexey Kopytov
5.6
Medium
Alexey Kopytov

Bug Description

XtraDB exposes the amount of memory allocated by the adaptive hash index structures in SHOW ENGINE INNODB STATUS (PS 5.5), the status variable Innodb_mem_adaptive_hash (PS 5.5 and 5.6), or the INFORMATION_SCHEMA.XTRADB_INTERNAL_HASH_TABLES table (PS 5.6).

In all cases the value is calculated as follows (with simplifications):

1. Get the number of cells in the first hash table and multiply it by the size of a cell.
2. Multiple value #1 by the number of adaptive hash index partitions.
3. Get the amount of heap memory allocated in the first hash table.
4. Multiple value #3 by the number of adaptive hash index partitions.

SHOW ENGINE INNODB STATUS in 5.5 and I_S table in 5.6 then show values #2 and #4 separately. And the status variable is always the sum of values #2 and #4.

The problem with #4 is that assumes all hash tables are uniformly filled. Which is not necessarily the case. Trivial example: multiple AHI partitions, but just one index using AHI. So depending on whether the hash table for that index happens to be in partition #0, we may get either much higher AHI memory than actually allocated, or much lower value than actually allocated.

Alexey Kopytov (akopytov) wrote :

Does not technically apply to 5.6, but will be soon once AHI partitioning is ported there.

tags: added: ahi-partitions
To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers