[OpenAFS-devel] Re: freezes acessing /afs/.git

[chas williams - CONTRACTOR]

On Thu, 14 Aug 2014 11:24:41 -0500
Andrew Deason <adeason@sinenomine.net> wrote:

> You need to bound the size anyway. Whether it's kept in afsd or the
> kernel I don't want it taking up all the memory in my machine. Caching
> it in the kernel means we really don't need to worry about the afsd
> upcall speed at all, since that path almost never happens.

Yes, but the bounds for the kernel would be more strict.  My thought is
"If it doesn't need to be there, it shouldn't be there."

> But yeah, the speed penalty of the afsd upcall may not be bad; I just
> thought the 'struct cell' was an intuitive place for it to go. We don't
> need a whole separate structure and framework for the negative cache
> entries specifically.

Take a look at http://gerrit.openafs.org/#change,11390

I am not sure one needs to cache positive results.  In fact, I don't
know that the kernel module ever forgets a cell it has learned via the
afsdb lookup (unless you intervene with fs)