23 Mar
2020
23 Mar
'20
5:50 p.m.
I haven't seen any MGR hangs so far since I disabled the prometheus
module. It seems like the module is not only slow, but kills the whole
MGR when the cluster is sufficiently large, so these two issues are most
likely connected. The issue has become much, much worse with 14.2.8.
On 23/03/2020 09:00, Janek Bevendorff wrote:
> I am running the very latest version of Nautilus. I will try setting up
> an external exporter today and see if that fixes anything. Our cluster
> is somewhat large-ish with 1248 OSDs, so I expect stat collection to
> take "some" time, but it definitely shouldn't crush the MGRs all the
time.
>
> On 21/03/2020 02:33, Paul Choi wrote:
>> Hi Janek,
>>
>> What version of Ceph are you using?
>> We also have a much smaller cluster running Nautilus, with no MDS. No
>> Prometheus issues there.
>> I won't speculate further than this but perhaps Nautilus doesn't have
>> the same issue as Mimic?
>>
>> On Fri, Mar 20, 2020 at 12:23 PM Janek Bevendorff
>> <janek.bevendorff(a)uni-weimar.de
>> <mailto:janek.bevendorff@uni-weimar.de>> wrote:
>>
>> I think this is related to my previous post to this list about MGRs
>> failing regularly and being overall quite slow to respond. The problem
>> has existed before, but the new version has made it way worse. My MGRs
>> keep dyring every few hours and need to be restarted. the Promtheus
>> plugin works, but it's pretty slow and so is the dashboard.
>> Unfortunately, nobody seems to have a solution for this and I
>> wonder why
>> not more people are complaining about this problem.
>>
>>
>> On 20/03/2020 19:30, Paul Choi wrote:
>> > If I "curl http://localhost:9283/metrics" and wait
sufficiently long
>> > enough, I get this - says "No MON connection". But the mons
are
>> health and
>> > the cluster is functioning fine.
>> > That said, the mons' rocksdb sizes are fairly big because
>> there's lots of
>> > rebalancing going on. The Prometheus endpoint hanging seems to
>> happen
>> > regardless of the mon size anyhow.
>> >
>> > mon.woodenbox0 is 41 GiB >= mon_data_size_warn (15 GiB)
>> > mon.woodenbox2 is 26 GiB >= mon_data_size_warn (15 GiB)
>> > mon.woodenbox4 is 42 GiB >= mon_data_size_warn (15 GiB)
>> > mon.woodenbox3 is 43 GiB >= mon_data_size_warn (15 GiB)
>> > mon.woodenbox1 is 38 GiB >= mon_data_size_warn (15 GiB)
>> >
>> > # fg
>> > curl -H "Connection: close" http://localhost:9283/metrics
>> > <!DOCTYPE html PUBLIC
>> > "-//W3C//DTD XHTML 1.0 Transitional//EN"
>> > "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
>> > <html>
>> > <head>
>> > <meta http-equiv="Content-Type"
content="text/html;
>> > charset=utf-8"></meta>
>> > <title>503 Service Unavailable</title>
>> > <style type="text/css">
>> > #powered_by {
>> > margin-top: 20px;
>> > border-top: 2px solid black;
>> > font-style: italic;
>> > }
>> >
>> > #traceback {
>> > color: red;
>> > }
>> > </style>
>> > </head>
>> > <body>
>> > <h2>503 Service Unavailable</h2>
>> > <p>No MON connection</p>
>> > <pre id="traceback">Traceback (most recent call
last):
>> > File
>> "/usr/lib/python2.7/dist-packages/cherrypy/_cprequest.py", line
670,
>> > in respond
>> > response.body = self.handler()
>> > File
>> "/usr/lib/python2.7/dist-packages/cherrypy/lib/encoding.py", line
>> > 217, in __call__
>> > self.body = self.oldhandler(*args, **kwargs)
>> > File
>> "/usr/lib/python2.7/dist-packages/cherrypy/_cpdispatch.py", line
61,
>> > in __call__
>> > return self.callable(*self.args, **self.kwargs)
>> > File "/usr/lib/ceph/mgr/prometheus/module.py", line 704, in
>> metrics
>> > return self._metrics(instance)
>> > File "/usr/lib/ceph/mgr/prometheus/module.py", line 721, in
>> _metrics
>> > raise cherrypy.HTTPError(503, 'No MON connection')
>> > HTTPError: (503, 'No MON connection')
>> > </pre>
>> > <div id="powered_by">
>> > <span>
>> > Powered by <a
href="http://www.cherrypy.org">CherryPy
>> 3.5.0</a>
>> > </span>
>> > </div>
>> > </body>
>> > </html>
>> >
>> > On Fri, Mar 20, 2020 at 6:33 AM Paul Choi <pchoi(a)nuro.ai
>> <mailto:pchoi@nuro.ai>> wrote:
>> >
>> >> Hello,
>> >>
>> >> We are running Mimic 13.2.8 with our cluster, and since
>> upgrading to
>> >> 13.2.8 the Prometheus plugin seems to hang a lot. It used to
>> respond under
>> >> 10s but now it often hangs. Restarting the mgr processes helps
>> temporarily
>> >> but within minutes it gets stuck again.
>> >>
>> >> The active mgr doesn't exit when doing `systemctl stop
>> ceph-mgr.target"
>> >> and needs to
>> >> be kill -9'ed.
>> >>
>> >> Is there anything I can do to address this issue, or at least
>> get better
>> >> visibility into the issue?
>> >>
>> >> We only have a few plugins enabled:
>> >> $ ceph mgr module ls
>> >> {
>> >> "enabled_modules": [
>> >> "balancer",
>> >> "prometheus",
>> >> "zabbix"
>> >> ],
>> >>
>> >> 3 mgr processes, but it's a pretty large cluster (near 4000
>> OSDs) and it's
>> >> a busy one with lots of rebalancing. (I don't know if a busy
>> cluster would
>> >> seriously affect the mgr's performance, but just throwing it
>> out there)
>> >>
>> >> services:
>> >> mon: 5 daemons, quorum
>> >> woodenbox0,woodenbox2,woodenbox4,woodenbox3,woodenbox1
>> >> mgr: woodenbox2(active), standbys: woodenbox0, woodenbox1
>> >> mds: cephfs-1/1/1 up {0=woodenbox6=up:active}, 1
>> up:standby-replay
>> >> osd: 3964 osds: 3928 up, 3928 in; 831 remapped pgs
>> >> rgw: 4 daemons active
>> >>
>> >> Thanks in advance for your help,
>> >>
>> >> -Paul Choi
>> >>
>> > _______________________________________________
>> > ceph-users mailing list -- ceph-users(a)ceph.io
>> <mailto:ceph-users@ceph.io>
>> > To unsubscribe send an email to ceph-users-leave(a)ceph.io
>> <mailto:ceph-users-leave@ceph.io>
>>
> _______________________________________________
> ceph-users mailing list -- ceph-users(a)ceph.io
> To unsubscribe send an email to ceph-users-leave(a)ceph.io