Ceph » CephFS

If the leader handles two up:boot beacons from a new MDS, it may fail the new MDS if the two beacon updates are batched in the same PAXOS transaction. For example:

2023-04-05T00:58:13.435+0000 7fc56d485700  7 mon.a@0(leader).mds e286 prepare_update mdsbeacon(105603/d up:boot seq=2 v286) v8
2023-04-05T00:58:13.435+0000 7fc56d485700 12 mon.a@0(leader).mds e286 prepare_beacon mdsbeacon(105603/d up:boot seq=2 v286) v8 from mds.? [v2:127.0.0.1:6832/3118878188,v1:127.0.0.1:6833/3118878188]                                                                                                                                                                                         
2023-04-05T00:58:13.435+0000 7fc56d485700 15 mon.a@0(leader).mds e286 prepare_beacon got health from gid 105603 with 0 metrics.
2023-04-05T00:58:13.435+0000 7fc56d485700  0 log_channel(cluster) log [INF] : daemon mds.d restarted
2023-04-05T00:58:13.435+0000 7fc56d485700  1 -- [v2:127.0.0.1:40314/0,v1:127.0.0.1:40315/0] --> [v2:127.0.0.1:40314/0,v1:127.0.0.1:40315/0] -- log(1 entries from seq 734 at 2023-04-05T00:58:13.436351+0000) v1 -- 0x55f4fd3cd880 con 0x55f4f1849000                                                                                                                                         
2023-04-05T00:58:13.435+0000 7fc56d485700  1 mon.a@0(leader).mds e286 fail_mds_gid 105603 mds.d role -1
2023-04-05T00:58:13.435+0000 7fc56d485700 10 mon.a@0(leader).paxosservice(osdmap 1..608) propose_pending
2023-04-05T00:58:13.435+0000 7fc56d485700 10 mon.a@0(leader).osd e608 encode_pending e 609