Comments on: Duplicate index checker version 1.9 released

By: Xaprb

Xaprb — Thu, 05 Oct 2006 22:43:21 +0000

Hello my brother, I love you too :-)

[Note: "john" above really is my brother, whose name isn't John.]

By: john

john — Thu, 05 Oct 2006 22:09:51 +0000

Yeah, well this whole idea isn't very helpful to me. I need the thing to make it not do that stuff. If you could fix that it will help my employees be much more efficient. See, I run a large enterprise company that deals in interactive software that uses databases. We have a large Access database that we administrate. Let me know if you want to work for us, we could make a good offer.

By: Xaprb

Xaprb — Wed, 04 Oct 2006 23:46:10 +0000

Yes, prefixes made it break silently. I fixed that, and in the process made it ignore column prefixes entirely. Now these key definitions end up being exactly the same, so they will be reported as duplicates:

  PRIMARY KEY  (`day`,`account`),
  KEY `day` (`day`(12),`account`),
  KEY `day` (`day`,`account`(5)),

Of course, they're not really duplicates, or might not be anyway, but my whole philosophy with this tool is "catch enough problems to be helpful, then report them and let the DBA figure it out."

By: Xaprb

Xaprb — Mon, 02 Oct 2006 11:37:36 +0000

I thought about prefixes but decided not to go there. I could, though. I guess it wouldn't be too hard to take out the parenthesized prefix notation and leave just the column names.

Now that I think about it, I bet my regular expression breaks and doesn't get all the column names when there's a parenthesized prefix. I should look into this.

By: Roland Bouman

Roland Bouman — Mon, 02 Oct 2006 00:58:24 +0000

Hi Baron,

great to hear you got round the bug...let's hope it gets fixed anyway :)

By the way, what's your take on the column prefixes? I bumped into that when I was well into writing my article and at that point I just felt like "...oh no, not another complicating factor...". For example, we could have:

table (col1 varchar(1000), col2 char(1))

index1 (col1(100))
index2 (col1(1),col2)

And if it get's as insane as this it is likely that index1 is more selective than index2. In the end, that's what determines the performance, and the 'left-prefix' criterion is just a simple and mostly effective measure for the selectivity.