Search results

Also not that CephFS is considered 'experimental' by Ceph, and no guarantees that it won't crash or eat your data for breakfast. (At least last time I've checked.) *update* jewel announcement says "This is the first release in which CephFS is declared stable! Several features are disabled by...

Is there a simple commandline way to list the containers an vms which is not added to the HA?

What is the preferred and safe method to prevent a node from rebooting due to (observed) loss of quorum? Or in other words: how to temporarily disable fencing on a node?

bond mode 802.3ad is there any known problem related to multicast and bonding?

Great. Set the title to [resolved] ;-)

Oh okay, you're using 3.4. My 4.xx is jessie. Sorry. Then you should have the info in /var/log/kern.log with timestamps. But I believe without systemd even the syslog is able to contain all the relevant info, so if you didn't see anything relevant I'm out of [basic] ideas and leave others to...

Try journalctl -xn 500 or similar, I don't know the optimal command line since it's not really my favourite tool, to put it nicely. Should show all kinds of logs with timestamps, unless it decides it's not in the mood. ha-manager comes to my mind first, but others may come up with better ideas...

Look for the reasons before that. Who's initiating it, HA manager, or something else? Is there some related stuff in dmesg (or in the case of this bloody damned systemd, the journalctl)? Maybe OOM, or any lxc or kernel errors?

So far so good, no reboots after a few restarts. I consider this resolved. (I only wish someone would really comment on the linked corosync problem... *sigh*)

Could you tell us exactly when the machine goes down and when comes it up? (Since it's a CT clocks should be in sync.)

We throw all kinds of stuff into graphite (using collectd mostly) and visualise it with grafana (so this feature is just a "nice to have" to me), but as a general note: we always prefer to have historical data (graph / rrd) to gauges, usually much more useful than current values. When I display...

Maybe anything relevant in your syslog?

I don't think containers would ever support anything other than kernel-supported filesystem or block device based "disks". If there was a qcow2 kernel driver block device driver then the answer would be probably yes, but I don't remember such one existing. If you really insist you can possibly...

This keeps happening on various clusters, and only on some member nodes. Right now I have a 3 node cluster (a different one from the above) where node1 and node2 observes this extreme slowdown behaviour (CT/VM are okay but anything related to cluster management, from corosync-* command line...

@dietmar additional info: current version (Environment 4.3-1/e7cdc165 on this cluster I am reporting from) happily reboots, when: - doing systemctl restart corosync or - doing systemctl restart pve-cluster with no apparent reason (apart from obvious watchdog timeout), possibly (and hopefully)...

Some people call this system "FAQ". :-D Regards from the Stating The Obvious™ Department. ;-)

Great. If it was I wasn't able to spot it by mere searching. Doesn't really matter, though, I'm sure you know what you mean, so we expect to get jewel in PVE soon after 10.2.4 is out, if I understand correctly.

Wasn't jewel (stable) released in April, 2016?

You mean by ceph, not by proxmox, I reckon. My question was rather about when do we see the new ceph in proxmox, as @wolfgang mentioned, since hammer's EOL is right at the corner. (Which I don't worry about, unlike lots of bugfixes and features in jewel.)

What's the state of the matters? Are we there yet? :rolleyes: