Re: Catching up Mail Groups

[Jason L Tibbitts III <tibbs@uh.edu>]

>>>>> "AS" == Aharon (Al) Schkolnik <aharon@healdb.matat.health.gov.il> writes:

AS> What I, and I believe many others, want is for articles in mail groups
AS> which are not marked to be saved (u command), and which are read or
AS> deleted, to be deleted after the expiration period.  I guess what this
AS> all means is that we want deletion (or whatever is done by catching-up,
AS> the d command, etc.) to be treated as expiration.

This is exactly, positively, 100% what total-expire does.  Every message
that is killed, deleted or caught-up is deleted after the delay you set
(default 28 days).  Now, if I'm not mistaken, Gnus has to stat the files
(nnml) to see if they should go away.  This could, perhaps, be made more
efficient by storing some kind of age of each article somewhere so that the
stat doesn't get done.  I don't know what Gnus has to do to expire things
in an nnfolder group; I suspect it's a pile of region deletions which
causes a bunch of buffer gap movement and VM thrashing.  But then, nnfolder
isn't very good for big folders.  Myself, I'm drooling over the prospect of
nngdbm.

BTW, on a large nnml group (~1000 articles) with 28 day expiry it takes
about three seconds to do the expiry if the disk is local to the machine.
This goes up dramatically if I'm on an NFS mounted volume, even when the
interconnect is our 1+Gbit network.  NFS sucks for this kind of thing.

AS> Of course, this must be efficient.

That's just the problem, isn't it?  It's kind of difficult to say where the
bottleneck is since you don't provide any information at all about your
environment, how many articles you're looking at or what backends you're
using.  Perhaps with a little more information we could give you some
optimization tips.

 - J<