Re: spamd keeps running at 99% CPU until i kill the process

From: Richard Hobbs <richard.hobbs(at)crl.toshiba.co.uk>
Date: Wed Aug 29 2007 - 04:29:04 EDT

Hello,

Anthony Peacock wrote:

> Hi,
> 
> Richard Hobbs wrote:

>> Hello,
>>
>> John D. Hardin wrote:
>>> On Tue, 28 Aug 2007, Richard Hobbs wrote:
>>>
>>>> Could the size of "bayes_seen" and "bayes_toks" be causing this
>>>> timeout?
>>> Yes.
>>>
>>>> If so, what can i do about this?
>>> Disable automatic Bayes expiry and do a manual expiration run, and
>>> allow it to complete.
>>
>> So what you (and Michael Parker) are saying is that it's not the
>> checking of spam against the tokens that is causing the timeout, it's
>> the automatic expiration of old tokens that "conveniently" gets tagged
>> onto the same operation, right?
>>
>> If so, let me get this straight - an email comes in and goes off to
>> spamd. spamd then checks the message against the tokens to determine
>> whether it's spam or not, then runs an expiry of old tokens (or perhaps
>> it happens the other way around), and only then returns the mail to
>> exim. The expiration of old tokens takes a lot longer than the spam
>> checking and as a result it's timing out.
>>
>> So, if i disable automatic expiration, spamd will only attempt one
>> operation at a time, (checking of spam against the tokens) and should
>> therefore not timeout, correct?
>>
>> But, if i do disable automatic expiration, i will have to remember to do
>> it manually, or via cron.
>>
>> Is this all correct?

>
> Yes. Set up expiration in a cron job, once per day is usually fine.

OK, will do - i assume "sa-learn --force-expire" is the command to run via cron, right?

Also, how do i disable automatic expiration?

Thanks again,
Richard.

>> Would a suitable alternative be to delete "bayes_seen" and "bayes_toks",
>> then restart spamd? I know i would be deleting everything that had been
>> learned over the last period of time, but starting afresh may not be a
>> bad thing, seeing as the rules in the database are probably 6 months to
>> a year old now (we've not been using spamd for a year or so, because i
>> broke it and had no time to fix it!).
>>
>> Please let me know your thoughts, and also let me know whether deleting
>> both of those files is a good way to go.

> 
> No.
> 
> 

-- 
Richard Hobbs (Systems Administrator)
Toshiba Research Europe Ltd. - Speech Technology Group
Email: richard.hobbs@crl.toshiba.co.uk
Web: 
http://www.toshiba-europe.com/research/
Tel: +44 1223 376964        Mobile: +44 7811 803377

Received on Wed Aug 29 04:29:55 2007