MySQL Bugs: #45759: feature request: allow storage engines to be notified of impending index scan

Bug #45759	feature request: allow storage engines to be notified of impending index scan
Submitted:	25 Jun 2009 16:25	Modified:	3 Jul 2009 16:43
Reporter:	Zardosht Kasheff (OCA)	Email Updates:
Status:	Verified	Impact on me:	None
Category:	MySQL Server: Storage Engines	Severity:	S4 (Feature request)
Version:	5.1.35	OS:	Any
Assigned to:		CPU Architecture:	Any
Tags:	Contribution, INDEX, index scan

Description:
Storage engines are sometimes notified of impending query operations. They are notified of range query operations by the functions read_range_first and read_range_next. They know of a full table scan by rnd_init.

Storage engines are NOT notified when a full index scan about to take place. This can be helpful to know. In such cases, storage engines can do smart things.

Having a handler function be called before an index scan is about to happen is very helpful.

Looking in the code, there seem to be two places where index scans are initiated: join_read_first and join_read_last.

It is here where we would like to see the calls added

How to repeat:
n/a

Suggested fix:
will contribute a patch

patch to implement feature

Attachment: prepare_index_scan.txt (text/plain), 1.31 KiB.

adding tag "contribution".

Thank you for the feature request and code contributed.

Monty was kind enough to implement this feature, with test cases, in MariaDB. Here what he stated on the internals mailing list:

I have now pushed all your proposed changes + more to the current
MariaDB 5.1 tree. (changeset 2713)

You can get them by either getting the lastest MariaDB code from
launchpad or looking at the patch
https://lists.launchpad.net/maria-developers/msg00463.html

What about index_first() ?

I personally like prepare_index_scan, because it lets storage engine developers know what is coming, whereas one can mistake index_init with index_first. However, a name is a name, so I do not have very strong feelings. I care more about the functionality getting in :)

I mean, there is handler::index_first() method - it's used to find the first/minimal entry in the index. Can you use it instead of introducing a new one ?

(true, it does not *always* mean an index scan, but often it does)

Oops, sorry. I do not think it can be used, because I believe index_first is used for queries such as "select min(a) from foo use index(a);" In that case, it is not an index scan.

okay, thanks