Optimization for the expireover procedure.
Bill Davidsen
davidsen at tmr.com
Tue Oct 16 22:10:28 UTC 2007
Julien ÉLIE wrote:
> Hi Kirill,
>
>
>
>> But in the case of
>> expiry the main keys are time of the arrival and group number.
>>
>
> So expireover will not remove articles which no longer exist in
> the news spool if it is based upon such information.
> And articles stored in CNFS buffers will be expired from the overview
> even though there are still available from the spool.
>
> But well, these behaviours could be tweaked.
>
>
>
>> For example we can use a list of pointers to overview
>> records sorted according to arrival date or even expiration date ( this
>> is a little bit tricky). I believe this will be much more faster.
>>
>
> I think that it is not the arrival date which matters but the Date: header
> (correct me if I am wrong).
> As for scheduled expiration date, it cannot be determined since expire.ctl
> can change after the moment it is computed. And there is also the
> Expires: header to take into account.
>
>
> There is also expire to optimize (not only expireover).
>
>
We discussed this long ago, making a list of articles to expire and the
time at which that should happen. All based on the rules and policies in
place at the moment the article arrives. This can save a great deal of
time depending on how the information is stored for both articles and
overview. And with CNFS stores it depends on policy, do you expire when
and article wraps in the store (keep as long as possible) or when your
policy says to remove it? Actually there's a very good reason to expire
from policy, it keeps your policy from visibly changing under the users
depending on article size and volume.
But the most desirable change would be to spread the expire load over
all uptime, so that the server avoids a load spike when expire runs. To
that end the ability to set expire time at the moment the article
arrives is a benefit, and frankly most sites don't change their policy
all that often, and even if a recalculation of all the expire times were
needed when the policy changed, it would still probably be a win.
--
bill davidsen <davidsen at tmr.com>
CTO TMR Associates, Inc
Doing interesting things with small computers since 1979
More information about the inn-workers
mailing list