Ceph Cluster Upgrade¶

0. Terminology¶

Trivial Update: * each Ceph component can be updated in random order

Non-trivial upgrade: * each Ceph component must be updated in fixed order, usually (if not specified otherwise in release notes): monitors always first, rest after that, mds always last. * sometimes specific config settings are required prior or during updates

1. Prerequisites¶

• read ceph release notes: https://docs.ceph.com/docs/master/release-notes/
• cluster must be health ok
• set noout on cluster
• set backfil max to minimal values (if needed)
• check that all MDS (including standby!) are up and running
• keep a terminal with ‘ceph -w’ open during upgrades to monitor cluster messages

2. Host System Updates¶

• upgrade all admin*.ceph.bfh.ch hosts first
• then upgrade all storage*.ceph.bfh.ch hosts (while ceph-osds are not running in containers yet, install all system updates except ceph)
• finally upgrade all gateway*.ceph.bfh.ch hosts

For every host upgrade:

• install all updates
• shutdown all containers
• reboot host system
• remove obsolete kernels, if any

check ‘ceph -s’ if everything is well before proceeding to the next.

3. Container System Updates¶

for every ceph container:

• upgrade all packages (ceph package updates do not restart ceph services)
• restart all ceph services

check ‘ceph -s’ if everything is well before proceeding to the next.