Cost of newgroup messages
rra at stanford.edu
Fri Mar 9 22:30:23 UTC 2001
Marco d'Itri <md at Linux.IT> writes:
> On Mar 08, bill davidsen <davidsen at tmr.com> wrote:
>> Processing a newgroup message seems to have quite an impact on the
>> system if you get a bunch of them. It seems that controlchan prevents
>> getting a multiprocess load spike, but the active is reloaded, the
>> innfeeds are disrupted, and other moderately high cost things happen.
>> These are totally valid messages, not crap.
> Looks like it should queue all ctlinnd commands and execute them maybe
> once per hour from crontab after throttling innd.
I think we can do even better than that; INN already stashes a copy of the
newsfeeds file. If the newsfeeds file hasn't changed, there should be no
real cost to an rmgroup beyond removing the group from the hash table and
a newgroup should just involve running that group through all the
individual feed wildmat patterns to build its subscription list. No need
to reload newsfeeds, no need to reparse anything, and no need to restart
any outgoing feeds.
Russ Allbery (rra at stanford.edu) <http://www.eyrie.org/~eagle/>
More information about the inn-workers