Hi, it's been years since I last upgraded a non-cephadm cluster, but what I remember for MDS is: as soon as you have an upgraded MDS, it will take over as active. So your last Reef MDS will immediately become a standby. At least that's how it was until Nautilus, after that we migrated all clusters to cephadm. It might work differently, but I wanted to mention it anyway, just so you're aware of that possibility. Regards, Eugen Zitat von Christopher Durham <caduceus42@aol.com>:
Hi, The non-cephadm update procedure to update reef to squid (for rpm-based clusters) here: https://docs.ceph.com/en/latest/releases/squid/#upgrading-non-cephadm-cluste... suggests that your monitors/mds/radosgw/osds are all on separate servers. While perhaps theyshould be that is not possible at current.
If for example, I have a mon and an mds on a single server, I really don't want to updatejust a mon as the mds and mon may use common libraries. As such, I am thinking that to do an update in this scenario I do the following set noout set max_mds to for my filesystem, and note the remaining active mds disable standby_replay Then, for each of the servers running an mds (but not the remaining active mds): 1. stop all ceph daemons (mon,mds,radosgw,osd) and do the update, reboot and/or restart the mon and mds 2. Proceed to all the other mds servers except the last one that is on reef3. Finish all other remaining mon updates as in #1
4. Then when the last server with a reef mds is left stop it and let a standby squid mds take over and then do the update and either restart all the daemons or reboot 5. if there are other osd-only nodes, update them and restart the osds Then, afterwards, allow standby replay, reset max_mds to what it was before, unset noout, require squid for osds.
Does this make sense? Am I forgetting something? I know setting max_mds 1 is important but want to be sure I have not forgotten anything. Thanks. -Chris _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io