Mailing-List: contact derby-dev-help@db.apache.org; run by ezmlm
Precedence: bulk
Reply-To: <derby-dev@db.apache.org>
Received-SPF: neutral (asf.osuosl.org: local policy)
Message-ID: <44B95B44.6060702@amberpoint.com>
Date: Sat, 15 Jul 2006 14:16:52 -0700
From: Bryan Pendleton <bpendleton@amberpoint.com>
User-Agent: Thunderbird 1.5.0.4 (Windows/20060516)
MIME-Version: 1.0
To: derby-dev@db.apache.org
Subject: Re: [jira] Commented: (DERBY-1357) Short-circuit logic in optimizer
 appears to be incorrect...
References: <7826241.1152900434754.JavaMail.jira@brutus>
 <44B7EF3A.8080100@gmail.com>
In-Reply-To: <44B7EF3A.8080100@gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

>  So adding a test case with 
> timeout disabled won't show the symptom, and adding it with timeout 
> enabled will likely lead to intermittent failures on different machines. 

I agree. Unreliable tests can be worse than no tests at all. I'm comfortable
with this decision.

> release note would have to be something generic like "optimizer 
> fixes/enhancements could cause the optimizer to choose different query 
> plans, so performance of existing queries may change". 

I agree that this is a hard case, and I can see both sides. But I do think
that we can craft an umbrella release note which will be valuable to users
as they move to the new release, covering the multiple changes to the
optimizer as a single logical unit. It seems to me that the release note
should observe that:

  - there have been some optimizer changes in this release (with pointers
    to the specific bugs fixed)
  - the optimizer will now consider additional query plans
  - the optimizer's timeout logic has changed
  - we believe that the optimizer will now choose better plans, and choose
    them more quickly.
  - if you think you are seeing an optimizer-related problem as a result of
    upgrading to this release, please capture the actual query plan that is
    being chosen by the old release and by the new release, and send them
    to us so we can help with the analysis, and here's how to do that.

Perhaps that last bullet point should be a page on the wiki, and the release
note could just point to http://wiki.apache.org/db-derby/OptimizerDiagnosisTips
(or whatever the actual page turns out to be).

I'd be glad to help by reviewing the release note, etc.

thanks,

bryan