On Fri, Oct 15, 1999 at 09:35:23AM -0500, Bill Gribble wrote:
> I'm observing that after a few days of little use, or several hours 
> of heavy use (compiling etc) I have to stop and restart venus 
> (5.3.1) because the cache gets plugged up and I start getting 
> ENOSPC all over the place.  

I haven't been able to reliably reproduce this behaviour yet. There is
definitely a cache leak. I'm hoping to find some quick and reliable way
to trigger this leak, and then turn up debugging and find out where it
forgets to account for released blocks (or accounts for used blocks
twice).

Do you observe frequent disconnections?
What is the ratio of the cache size compared to the working set?
Are you using hoard to actively pull objects into the cache?

> The output of 'cfs listcache' seems to be relevant... I see the names
> of files that I haven't touched for days in there.  Some are marked
> with '*'.  Does the '*' mean that venus thinks someone has the file
> open for writing, or does it mean that the object is discardable,
> or something else? 

The '*' implies that only the object's status and not the data is
cached. As your client is confused about how much data actually occupies
cache-space, it purges the data part of the files. But as there is no
pressure on the number of files, it hangs on to files with only
attributes and acls.

Jan