httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Niklas Edmundsson <>
Subject Re: svn commit: r450105 - in /httpd/httpd/trunk: CHANGES modules/cache/mod_disk_cache.c modules/cache/mod_disk_cache.h
Date Wed, 27 Sep 2006 11:19:51 GMT
On Wed, 27 Sep 2006, Joe Orton wrote:

> I don't get it - as discussed, this approach is completely unsound.
> There is no reason to assume it's possible to copy the entire content
> into the cache before sending anything to the client just because it
> happens to be a FILE bucket (think slow NFS servers).  That is something
> which needs to be *fixed*, not explicitly hard-coded.

Yes, it has to be fixed eventually. Until then we're better off with 
gradual improvements than just saying "the solution isn't perfect".

mod_disk_cache isn't exactly the best code out there, so it'll take a 
while to get it decent, and it'll take more than a single patch to do 
it. I fully agree that the "copy everything into cache and then reply" 
method is utterly stupid, but that's the way mod_disk_cache currently 
works, and despite that it got tagged as "stable"...

In an effort to improve things, I'll start taking more stuff from 
our jumbo patch and building smaller incremental patches that will 
eventually mean that mod_disk_cache will have read-while-caching.

When that code is in there will be a plethora of options on how to 
solve the "client that does the caching request hangs" problem which 
we have kludged to work in the nfs-backend case. As said, we have 
kludged it to work in our setup (which has a slow NFS backend), but 
the "perfect solution" will have to come from people who knows all the 
deep magic in httpd and I know I'm not that person.

  Niklas Edmundsson, Admin @ {acc,hpc2n}      |
  "Only together can we turn him to the dark side of the Force." - Emperor

View raw message