couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Göran Krampe <go...@krampe.se>
Subject Re: Possible bug in indexer... (really)
Date Sun, 05 Jul 2009 00:06:44 GMT
Noah Slater wrote:
> On Sat, Jul 04, 2009 at 02:57:20AM +0200, Chris Anderson wrote:
>> We go our of our way to say "don't do it like that", and we shouldn't
>> optimize for cases that aren't supported. I think a performance fix
>> for this experience would actually be a bad thing, as it would make
>> the symptoms of a bad reduce more subtle. If you write a reduce that
>> doesn't reduce, it should blow up in your face as soon as possible.
> 
> +1

Some feedback:

- We knew we were doing a "largish" reduce but is a *bounded* one.

- Yes, we got the warning, I disabled it. So no worries there, it works.

- I did try out this idea on Paul Davis on IRC. He thought it looked 
fine as long as we made sure about the boundaries. So don't be too harsh 
on me, we did ask :) :)

- group=true sucks performance wise. We used that in the beginning but 
realized quickly we couldn't use it. I don't recall the details now.

Finally, we would be SUPER HAPPY to get a lesson in how to produce these 
"period sums" the Right Way. So we have dated transactions on accounts.

We want fast lookup of precalculated sums for each account per month and 
also typically for the full year.

Also, regarding "memory". The processes don't seem to take much memory, 
I mean there is plenty of RAM to go here. Not sure why Couch would take 
harsh measures before RAM is scarce?

regards, Göran


Mime
View raw message