Bug #56438 Make it possible for handler to affect the hash value for partition
Submitted: 1 Sep 2010 8:56 Modified: 27 Jan 2011 16:55
Reporter: Magnus Blåudd Email Updates:
Status: Closed Impact on me:
None 
Category:MySQL Server: Partitions Severity:S2 (Serious)
Version:5.5.7-m3 OS:Any
Assigned to: Mattias Jonsson CPU Architecture:Any
Tags: SEAGULL
Triage: Triaged: D5 (Feature request) / R2 (Low) / E2 (Low)

[1 Sep 2010 8:56] Magnus Blåudd
Description:
It should be possible for a handler/engine to influence the hash value used for partitioning.

How to repeat:
.

Suggested fix:
Extend handler with with a new function 'calculate_key_hash_value' which takes an array of the fields which the partition key consist of and returns a uint32.

This function would then be called instead of 'calculate_key_value' in the partitioning functions in sql_partition.cc

The default 'calculate_key_value' implementation would be moved to ha_partition::calculate_key_hash_value.
 
=== modified file 'sql/handler.h'
--- sql/handler.h	2010-08-18 11:55:37 +0000
+++ sql/handler.h	2010-09-01 08:50:18 +0000
@@ -1562,6 +1562,8 @@ public:
   virtual int info(uint)=0; // see my_base.h for full description
   virtual void get_dynamic_partition_info(PARTITION_STATS *stat_info,
                                           uint part_id);
+  virtual uint32 calculate_key_hash_value(Field **field_array)
+  { DBUG_ASSERT(0); return 0; }
   virtual int extra(enum ha_extra_function operation)
   { return 0; }
   virtual int extra_opt(enum ha_extra_function operation, ulong cache_size)
[1 Sep 2010 9:15] Magnus Blåudd
Function can actually be marked 'const', there should be no need for the handler to update any of its member data as part if this call.
[8 Sep 2010 11:43] Mattias Jonsson
[Mikael] This is a necessary fix to the partitioning to enable handlers to take care of the hash value calculation. This is a must for the NDB engine. Fix should have been done already on 5.1 but this wasn't done for some reason.
[28 Oct 2010 15:22] Bugs System
A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/122220

3545 Mattias Jonsson	2010-10-28
      Bug#56438: Make it possible for handler to affect the hash value for partition
      
      This is a necessary fix to the partitioning to enable handlers to take care of
      the hash value calculation. This is a must for the NDB engine. Fix should have been done
      already on 5.1 but this wasn't done for some reason.
     @ sql/ha_partition.cc
        Moved calculate_key_value from sql_partition.cc to here as
        ha_partition::calculate_key_hash_value.
     @ sql/ha_partition.h
        Extended set_part_info and added calculate_key_hash_value
     @ sql/handler.h
        Extended set_part_info and added calculate_key_hash_value
     @ sql/partition_info.h
        Added a reference to the table instance.
     @ sql/sql_partition.cc
        Using the handlers calculate_key_hash_value (the old calculate_key_value
        was moved to ha_partition.cc) and setting the table reference on part_info
[2 Nov 2010 13:21] Magnus Blåudd
Patch is OK.
[2 Nov 2010 13:56] Mattias Jonsson
Related to bug#29550.
[26 Jan 2011 21:15] Bugs System
A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/129705

3557 Mattias Jonsson	2011-01-26 [merge]
      merge of bug#56438 into mysql-trunk
[26 Jan 2011 21:16] Bugs System
Pushed into mysql-trunk 5.6.2 (revid:mattias.jonsson@oracle.com-20110126211256-7netk3waleausbsw) (version source revid:mattias.jonsson@oracle.com-20110126211256-7netk3waleausbsw) (merge vers: 5.6.2) (pib:24)
[27 Jan 2011 16:41] Jon Stephens
Mattias, Magnus -- I don't see how this impacts end users. Can you help me out? Thanks!
[27 Jan 2011 16:50] Mattias Jonsson
This does not impact any end user. Only internal for engine developers.
[27 Jan 2011 16:55] Jon Stephens
That's what I thought. :)

No end-user changes to document per developer comment above; closed without further action.