Re: Strange behavior on the 2.4.18 XFS tree?

[Michael Sinz <msinz@xxxxxxxxx>]

Michael Sinz wrote:
> I have only been able to see this behavior on my system running the
> kernel from the 2.4 XFS CVS tree.  It is somewhat reproduceable.
>
> The problem is that I started up a number of "rxvt" sessions and they
> took well over a minute to start up.  During that time, the cpu
> utilization was almost nil.  (Less than 1%)
>
> There was a background find operation happening, and this seems to be
> the key item.
>
> I could not run top while waiting for the sessions to start.  I could
> not run "ps" either.  Both blocked until the sessions finally came up
> and thus they produced no useful information.
>
> Is there some BKL that is being held longer than it should be in the
> XFS code?  The 2.4.18 non-XFS system does not show this behavior.
> (Well, at least not to this degree.  The find in the background does
> slow things up a bit but only by a second or two, not over 60 seconds)

Well, I must be unlucky or something - I have been trying to reproduce
this for some time and just happened to get it to happen again.  There
seems to be something strange going on in the VFS layer as I was also
able to get this to happen in a non-XFS system.  So, it looks like it
really is not an XFS problem afterall.  I really wish it were easier
to reproduce...

--
Michael Sinz ---- Worldgate Communications ---- msinz@xxxxxxxxx
A master's secrets are only as good as
        the master's ability to explain them to others.