This is a silly chicken-and-egg problem. I've found when I issue a reboot on a server that's mounted its own gluster volume via NFS (with the 'hard' option set), it stops gluster before unmounting volumes. This causes the NFS client to hang waiting for a response from the gluster server (which will never happen because it's stopped), blocking the reboot, leaving the server hanging indefinitely. The only way out seems to be to power-cycle the server. How is this supposed to work? Is there some way of scheduling unmounts before killing gluster? Marcus
On Thu, 25 Jul 2013 14:52:19 +0200 Marcus Bointon <marcus at synchromedia.co.uk> wrote:> This is a silly chicken-and-egg problem. I've found when I issue a reboot on a server that's mounted its own gluster volume via NFS (with the 'hard' option set), it stops gluster before unmounting volumes. This causes the NFS client to hang waiting for a response from the gluster server (which will never happen because it's stopped), blocking the reboot, leaving the server hanging indefinitely. The only way out seems to be to power-cycle the server. > > How is this supposed to work? Is there some way of scheduling unmounts before killing gluster? > > Marcus > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://supercolony.gluster.org/mailman/listinfo/gluster-usersIt's interesting. My experience, if you mount nfs on other computer, poweroff gluster node, your nfs client will hang too. It looks like bug of gluster native nfs server. Maybe some masters can explain the reason.
On Thu, Jul 25, 2013 at 5:52 AM, Marcus Bointon <marcus at synchromedia.co.uk>wrote:> This is a silly chicken-and-egg problem. I've found when I issue a reboot > on a server that's mounted its own gluster volume via NFS (with the 'hard' > option set),Mounting NFS export from localhost is a recipe for disaster for many other reasons (including deadlocks under heavy IO).> it stops gluster before unmounting volumes. This causes the NFS client to > hang waiting for a response from the gluster server (which will never > happen because it's stopped), blocking the reboot, leaving the server > hanging indefinitely. The only way out seems to be to power-cycle the > server. > > How is this supposed to work? Is there some way of scheduling unmounts > before killing gluster? >Not sure about this - should the init.d shutdown script be de-prioritized for your runlevel? But then, we are talking about a use case which is highly non-recommended. Avati -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130728/18fe6187/attachment.html>