Hi, Actually down/slow OSDs are detected by other OSDs. OSDs exchange heartbeats every few seconds. When an OSD haven't received a reply from its neighbor within the grace period (default: 20 seconds), the OSD (which has sent the heartbeat) considers its neighbor down and reports this to a monitor. After receiving 3 such messages in a row the monitor marks the OSD in question as down. Also OSDs report their status directly to monitors every few (~2) minutes (to avoid flooding monitors). If the monitor haven't received any status reports from an OSD within a grace period (~15 minutes) the OSD in question is considered down. If all OSDs are down there is no OSD which could have reported its neighbors are down. Thus monitor will consider all OSDs as up until the OSD report grace period (~15 minutes) expires. See http://docs.ceph.com/docs/giant/rados/configuration/mon-osd-interaction/#configuring-monitor-osd-interaction for more details. Best regards, Alexey On Sat, Jan 14, 2017 at 7:59 PM, Jin Cai <caijin.laurence@xxxxxxxxx> wrote: > Hi all, > I deployed a ceph cluster with jewel in four physical machines. > Three physical machines were used for OSD and each of them had eight > OSDs. The left one was used to serve as a monitor. > At first, everything worked well. > By the reason of test, I stopped all the OSD daemon and double checked > no OSD process running. > After that, I executed ceph -s and got the following output: > osdmap e164: 24 osds: 7 up, 7 in > No matter how much time elapsed, the output didn't change. > > The expected output should be: > osdmap e164: 24 osds: 0 up, 0 in > > I think it is the matter of synchronisation between OSDs and monitor > Would you like you explain this strange phenomenon for me. > Thanks a bunch in advance. > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
