[OpenAFS] Re: Client Cache Question

Andrew Deason adeason@sinenomine.net
Tue, 25 Jun 2013 15:59:05 -0500

On Tue, 25 Jun 2013 13:49:06 -0700
Timothy Balcer <timothy@telmate.com> wrote:

> So, this means that there is no networkly or I/O reason on the
> destination or source that could be the culprit particularly.

That's not necessarily true; even discarding the issues with Rx (the
transport protocol AFS uses), some networking issues can be UDP-only,
which would affect AFS but not straight rsync. You may find that very
unlikely, and I certainly try not to jump straight to that conclusion
right away, but it has happened before.

> Im running an fstrace right now, but as you said, its very cryptic.

Yeah, I meant that for an fstrace you'd need to give it to an openafs
developer (give it to me, or to the list, or whatever). That would
contain filenames and volume names, but not file data.

> Are you suggesting I run an strace on the rsync process? Or that I
> attach a strace to all the afs background daemons?

Just strace the rsync process. Just to see if we're slow on, say,
open/write/close, or something you may not expect, like lstat or
getdents or something. That would also tell you if a specific file is
slower than all of the others. Before we can look at what it is within
AFS that's making things so slow, it would help to know what operation
is actually being slow.

Obviously if you share the raw strace output with someone, that would
contain path names and a little bit of the file data. I was intending
for you to look at that and see if anything is unusual, or just report
what seems "slow".

Andrew Deason