"Andrew M.A. Cater" <amaca...@galactic.demon.co.uk> writes:
> On Sun, Sep 21, 2014 at 02:48:59AM +0200, lee wrote: >> Hi, >> >> any idea why an NFS volume is being unmounted when a VM runs out of >> memory and kills some processes? These processes use files on the NFS >> volume, but that's no reason to unmount it. >> > > The OOM process killer is not necessarily aware: it will kill processes until > memory usage > works again - and going OOM is, itself, a sign that something is fairly > drastically wrong. The VM needs more memory which the server doesn't have. For now, I assigned more swap space to the VM. The process killed was seamonkey. Killing seamonkey frees at least 1GB. > Suppose: > > There's heavy NFS usage and stuff is queueing to get on and off the mount / > high disk I/O NFS usage isn't heavy. > and the kernel hits a stuck "something" - memory usage may spike and the OOM > will eventually > kill processes. Even NFS processes, after freeing 1GB+ by killing seamonkey? > If something is using the NFS and there are stale mounts, that prevents > clean unmounting ... further problems. There's only one mount, and it isn't stale. > Under high I/O I've occasionally seen information > messages where the kernel is backing off for 120s and the note that you can > disable further > displays of this message by cat'ing something into proc. [Hey, it's a Sunday > morning and > I can't remember everything :) There wasn't much I/O going on. Seamonkey was actually idling while I was doing something else that doesn't touch the VM seamonkey was running in in any way. > If a disk is being left unclean / marked as such, it won't be automatically > mounted, > of course. If there's a problem with a remote mount / dead file handles that > will also > clobber it. The VM seamonkey was running in doesn't export anything via NFS. It only mounts a volume exported by another VM. The export was fine or I would have noticed on my desktop because it has the same volume mounted the same way (/home) and would have frozen up. >> Also annoying: The volume doesn't get mounted when booting despite it's >> in /etc/fstab. I have to log in into the VM and mount it manually. The >> entry in fstab is correct: I can just say 'mount /mountpoint' and it's >> mounted. Is that some Debian-specific bug? >> > > What's the entry? [Too many times, I've seen a typo or something anomalous > only > once I've called a colleague in to have a look :) ] jupiter:/srv/data_home /home nfs defaults 0 0 -- Knowledge is volatile and fluid. Software is power. -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/87zjdsdftf....@yun.yagibdah.de
