If there is actually a connection, then it's no wonder our MDS kept crashing. Our Ceph has 9.2PiB of available space at the moment. On 26/03/2020 17:32, Paul Choi wrote: > I can't quite explain what happened, but the Prometheus endpoint > became stable after the free disk space for the largest pool went > substantially lower than 1PB. > I wonder if there's some metric that exceeds the maximum size for some > int, double, etc? > > -Paul > > On Mon, Mar 23, 2020 at 9:50 AM Janek Bevendorff > <janek.bevendorff@xxxxxxxxxxxxx > <mailto:janek.bevendorff@xxxxxxxxxxxxx>> wrote: > > 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@xxxxxxxxxxxxx > <mailto:janek.bevendorff@xxxxxxxxxxxxx> > >> <mailto:janek.bevendorff@xxxxxxxxxxxxx > <mailto:janek.bevendorff@xxxxxxxxxxxxx>>> 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@xxxxxxx > <mailto:pchoi@xxxxxxx> > >> <mailto:pchoi@xxxxxxx <mailto:pchoi@xxxxxxx>>> 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@xxxxxxx > <mailto:ceph-users@xxxxxxx> > >> <mailto:ceph-users@xxxxxxx <mailto:ceph-users@xxxxxxx>> > >> > To unsubscribe send an email to ceph-users-leave@xxxxxxx > <mailto:ceph-users-leave@xxxxxxx> > >> <mailto:ceph-users-leave@xxxxxxx > <mailto:ceph-users-leave@xxxxxxx>> > >> > > _______________________________________________ > > ceph-users mailing list -- ceph-users@xxxxxxx > <mailto:ceph-users@xxxxxxx> > > To unsubscribe send an email to ceph-users-leave@xxxxxxx > <mailto:ceph-users-leave@xxxxxxx> > _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |