On Fri, Jul 17, 2015 at 03:12:45PM +0200, Etienne Vogt wrote:
> Hi there,
> I am currently testing the lastest
> linux-image-vserver-3.18-beng vserver kernel on Debian
> jessie
> (that's linux-image-3.18.11-vs2.3.7.4-beng 0.1+jessie1).
> It seems there is a problem with inodes accounting : on a
> vserver guest with dlimits enabled, inodes_used inside the
> guest goes up over time and finally hits the inodes_total
> limit.
> Then of course things start to fail with "no space left on
> device" error messages.
Yes, thanks for reporting, "we" are aware of this issue.
> At the same time, df -i on the host shows no such heavy
> inode usage on the /vservers partition.
> I can see such thing as :
> <host> ~# vserver guest exec df -i
> Filesystem Inodes IUsed IFree IUse% Mounted on
> /dev/hdv1 1000000 279341 720659 28% /
> <host> ~# df -i
> Filesystem Inodes IUsed IFree IUse% Mounted on
> /dev/sda8 10993664 216732 10776932 2% /vservers
> So more inodes appearing in use on a guest than are really
> in use on the whole /vservers partition.
> Stopping the guest and clearing /var/cache/vservers/guest/
> restores the usage to normal values but then it starts
> to increase anomalously again.
> It looks like the inodes count is not decreased properly
> in some circonstances.
The diagnosis is correct, it seems that when a file is
overwritten, a "new" inode is accounted, but the old one
is not freed, which ultimately accumulates and hits the
limit.
It would be nice to know what version before 3.18.x was
unaffected of this issue, as I presume either something
in the kernel changed or we "lost" a hunk during porting.
Thanks in advance,
Herbert
> --
> Etienne Vogt (Etienne.Vogt@obspm.fr)
> Observatoire de Paris-Meudon
> Service Informatique
Received on Fri Jul 17 15:04:14 2015