By default, Ceph uses host as failure domain (which means it forces placement groups to replicate across multiple hosts).
Our temporary setup has a single host (until the hypervisor3 ssds are emptied and repurposed). We need to set up a temporary CRUSH rule allowing replication on a single host.
| Status | Assigned | Task | ||
|---|---|---|---|---|
| Migrated | gitlab-migration | T2501 Proxmox reliability improvements (Summer 2020) | ||
| Migrated | gitlab-migration | T2502 Migrate VM storage to ceph | ||
| Migrated | gitlab-migration | T2503 Set up hyperconverged ceph cluster | ||
| Migrated | gitlab-migration | T2508 Update Ceph pool to use the default CRUSH replication rule | ||
| Migrated | gitlab-migration | T2504 Add hypervisor3 ssds as Ceph OSDs | ||
| Migrated | gitlab-migration | T2505 Migrate hypervisor3 vms to (temporary) ceph pool | ||
| Migrated | gitlab-migration | T2507 Set up temporary (single-host) ceph pool | ||
| Migrated | gitlab-migration | T2506 Add new beaubourg SSDs as Ceph OSDs |
Set up CRUSH rule using OSD as failure domain (instead of host)
sudo ceph osd crush rule create-replicated single_host_rule default osd
Move pool to this CRUSH rule
sudo ceph osd pool set proxmox crush_rule single_host_rule
Check osd health
sudo ceph status
check that cluster health is HEALTH_OK
check that pgs are all active+clean.