On 23.2.2010, at 22.49, Stan Hoeppner wrote:
Well, v1.1+ does case-insensitive searches by converting all mails to UTF-8 first. But it really shouldn't make it that much slower. What OS and CPU is this with?
Debian 5.0.4 Linux kernel 2.6.31.1 rolled by me from kernel.org source Dovecot, Debian backport Version: 1:1.2.10-1~bpo50+1 Intel 440BX Abit BP6 mobo, Dual Mendocino Celerons @550 MHz, 384MB PC100
So not really the fastest CPU :)
You could also try if it happens to work faster by compiling Dovecot without iconv support.
Was iconv support in dovecot 1.0.15? How can I check to see if my dovecot 1.2.10 already has iconv support? Is there a dovecot command similar to postconf -m which shows what database types were compiled into the postfix binaries?
Actually now that I think of it, iconv was used in v1.0 just the same. The difference is just the extra step of converting all text to titlecase before comparing it. That should be a pretty fast operation too.
I guess there could also be some kind of a bug that you're triggering.
Is there any setting in dovecot.conf, that if set incorrectly, could cause this?
No.
If not, and assuming the Debian team didn't screw up the package, I'd lean toward this bug explanation, except that no one else is reporting this problem, are they?
I haven't heard. But then again most people don't seem to be using server-side SEARCH anyway, especially on older machines.
These searches used to take about 10 seconds on 1.0.15. They're taking 3 minutes or more on 1.2.10. Also, on 1.0.15, if I ran the same search back-to-back, the second search completed _instantly_. On 1.2.10, the second search takes as long as the first. This leads me to believe the indexes and caches aren't being used at all during the searches, or something along these lines. Shouldn't they be?
No, body searches aren't indexed (without fts plugins).
This happened for some spam mailbox? Could you send it to me and I'll try if I can reproduce it?