Mailing-List: contact derby-dev-help@db.apache.org; run by ezmlm
Precedence: bulk
Reply-To: <derby-dev@db.apache.org>
Received-SPF: neutral (athena.apache.org: local policy)
Message-ID: <4852F15F.6080607@sbcglobal.net>
Date: Fri, 13 Jun 2008 15:14:55 -0700
From: Kathey Marsden <kmarsdenderby@sbcglobal.net>
User-Agent: Thunderbird 1.5.0.14 (Windows/20071210)
MIME-Version: 1.0
To: derby-dev@db.apache.org
Subject: Re: Advice in debugging plan selection
References: <4852B309.6050102@sbcglobal.net> <4852C450.8060405@gmail.com>
In-Reply-To: <4852C450.8060405@gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Army wrote:
>
> If you're in a situation where you have to use printlns, you may want 
> to start by printing out the contents of the bestJoinOrder array in 
> OptimizerImpl in two places: 1) at the end of rememberBestCost(), and 
> 2) at the end of modifyAccessPaths().  The former tells you every time 
> the OptimizerImpl thinks it has found a new "best join order" thus far 
> in processing, the latter tells you what the OptimizerImpl's _final_ 
> choice of "best join order" was.  If you print out the cost of the 
> join orders (bestCost) as well, and then compare the results from a 
> "fail" plan and a "pass" plan, that might give you an indication of 
> where to start.  But as I said, there's no silver bullet in tracking 
> that kind of thing down...
>
Well I finally got the printlns with the failing and passing runs. It 
seemed much less likely to happen once I put them in, but eventually I 
got a hit, but I don't think I am understanding what I am seeing. I 
expected to see large costs with the failing run.  But I think I messed 
up and printed the current cost instead of the best cost, so I will try 
to print that out too and see what happens.

Kathey