Re: [fuse-devel] Suspecting memory leak in fuse in Linux 2.6.26

[Goswin v. B. <gos...@we...>]

Miklos Szeredi <mi...@sz...> writes:

> On Thu, 11 Dec 2008, Goswin von Brederlow wrote:
>> No. The fuse filesystems never stay mounted long, just long enough for
>> me to find a bug or test that I fixed it. The cound above is without
>> any fuse filesystem mounted. I guess I have to make some test to see
>> how I can make i go up specifically. Just thought to ask if that is a
>> known issue.
>
> It might be perfectly normal: the linux kernel will use free memory
> pretty agressively for caching.  Then when something else needs the
> memory it will shrink the caches.

I just see a lot more ram and swap used in 2.6.26 than in my old
kernel. Previously I always had a fair amount of buffer/cache but now
that shrinks to about 100MB unless I do a ton of I/O. I now have 288MB
swap used where I had 1MB used after 100+ days with the same
applications running all the time. In fact I have one memmory hugger
(sanchos) not running anymore now. But I changed a few things: kernel
version, activate xen, experimental XOrg for xen support and extensive
fuse usage. So who there are too many variables.

> Fuse itself is unlikely to be leaking dentries: that would result in
> nasty warnings about inodes still being used at umount.

I guess that rules out fuse then. Although ...

What if the filesystem segfaults or quits in an odd way that leaves
the connection endpoint disconnected but still mounted? What if I
restart/remount the FS from such a state? Any chance that could
confuse the kernel part?

MfG
        Goswin