[ceph-users] Re: Not all OSDs in rack marked as down when the rack fails

On 30/10/2020 11:28, Wido den Hollander wrote: See: https://tracker.ceph.com/issues/48274 2020-11-18T02:33:21.498-0500 7f849ae8c700 0 log_channel(cluster) log [DBG] : osd.58 reported failed by osd.105 2020-11-18T02:33:21.498-0500 7f849ae8c700 10 mon.CEPH2-MON1-206-U39(a)0(leader).osd e108196 osd.58 has 1 reporters, 82.847365 grace (20.000000 + 62.8474 + 5.67647e-167), max_failed_since 2020-11-18T02:32:59.498216-0500 2020-11-18T02:33:21.498-0500 7f849ae8c700 10 mon.CEPH2-MON1-206-U39(a)0(leader).log v9218943 logging 2020-11-18T02:33:21.499338-0500 mon.CEPH2-MON1-206-U39 (mon.0) 165 : cluster [DBG] osd.58 reported failed by osd.105 The MONs kept adding time to the grace period. In the end setting mon_osd_adjust_heartbeat_grace to 'false' solved it. Why? I'm not sure yet. root default rack 206 host A rack 207 host B rack 208 host C The names of the racks are 'integers' so I tried renaming them to 'r206' for example, but that didn't work either. We achieved our goal, but I'm not sure why this setting is preventing the OSDs from being marked as down. Wido > Wido > >> >> Cheers, >> >> Dan >> >> >> On Thu, 29 Oct 2020, 18:05 Wido den Hollander, &lt;wido(a)42on.com >> <mailto:wido@42on.com>> wrote: >> >>     Hi, >> >>     I'm investigating an issue where 4 to 5 OSDs in a rack aren't >> marked as >>     down when the network is cut to that rack. >> >>     Situation: >> >>     - Nautilus cluster >>     - 3 racks >>     - 120 OSDs, 40 per rack >> >>     We performed a test where we turned off the network Top-of-Rack for >>     each >>     rack. This worked as expected with two racks, but with the third >>     something weird happened. >> >>       From the 40 OSDs which were supposed to be marked as down only 36 >>     were >>     marked as down. >> >>     In the end it took 15 minutes for all 40 OSDs to be marked as down. >> >>     $ ceph config set mon mon_osd_reporter_subtree_level rack >> >>     That setting is set to make sure that we only accept reports from >> other >>     racks. >> >>     What we saw in the logs for example: >> >>     2020-10-29T03:49:44.409-0400 7fbda185e700 10 >>     mon.CEPH2-MON1-206-U39(a)0(leader).osd e107102  osd.51 has 54 >> reporters, >>     239.856038 grace (20.000000 + 219.856 + 7.43801e-23), >> max_failed_since >>     2020-10-29T03:47:22.374857-0400 >> >>     But osd.51 was still not marked as down after 54 reporters have >>     reported >>     that it is actually down. >> >>     I checked, no ping or other traffic possible to osd.51. Host is >>     unreachable. >> >>     Another osd was marked as down, but it took a couple of minutes as >> well: >> >>     2020-10-29T03:50:54.455-0400 7fbda185e700 10 >>     mon.CEPH2-MON1-206-U39(a)0(leader).osd e107102  osd.37 has 48 >> reporters, >>     221.378970 grace (20.000000 + 201.379 + 6.34437e-23), >> max_failed_since >>     2020-10-29T03:47:12.761584-0400 >>     2020-10-29T03:50:54.455-0400 7fbda185e700  1 >>     mon.CEPH2-MON1-206-U39(a)0(leader).osd e107102  we have enough >> reporters >>     to mark osd.37 down >> >>     In the end osd.51 was marked as down, but only after the MON >> decided to >>     do so: >> >>     2020-10-29T03:53:44.631-0400 7fbda185e700  0 log_channel(cluster) log >>     [INF] : osd.51 marked down after no beacon for 903.943390 seconds >>     2020-10-29T03:53:44.631-0400 7fbda185e700 -1 >>     mon.CEPH2-MON1-206-U39(a)0(leader).osd e107104 no beacon from osd.51 >>     since >>     2020-10-29T03:38:40.689062-0400, 903.943390 seconds ago.  marking >> down >> >>     I haven't seen this happen before in any cluster. It's also strange >>     that >>     this only happens in this rack, the other two racks work fine. >> >>     ID    CLASS  WEIGHT      TYPE NAME >>         -1         1545.35999  root default >> >>     -206          515.12000      rack 206 >> >>         -7           27.94499          host CEPH2-206-U16 >>     ... >>     -207          515.12000      rack 207 >> >>        -17           27.94499          host CEPH2-207-U16 >>     ... >>     -208          515.12000      rack 208 >> >>        -31           27.94499          host CEPH2-208-U16 >>     ... >> >>     That's how the CRUSHMap looks like. Straight forward and 3x >> replication >>     over 3 racks. >> >>     This issue only occurs in rack *207*. >> >>     Has anybody seen this before or knows where to start? >> >>     Wido >>     _______________________________________________ >>     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