Instance RecoveryΒΆ
Each compute node runs a Consul agent which periodically performs a TCP health check against the network interface card (NIC). Consul server on the control node detects the unreachable consul agent periodically. If the NIC is down, the following actions are taken on the affected node:
Instances are shut down gracefully.
The
nova-computeservice is disabled.
Masakari checks the status from the Consul server and triggers the recovery. Refer to the recovery matrix in the Instance Recovery how-to for when recovery is triggered and whether instances needs to be evacuated.
For VMs with encrypted disks, the recovery workflow may need to retrieve
Barbican secret payloads. Canonical OpenStack keeps the normal Barbican decrypt
paths for project administrators, project members, secret owners, non-private
secrets, and ACLs. It also allows trusted service users in the services
project to decrypt secrets when they have the secret-decrypter role. Masakari
uses this service role by default so it can rebuild encrypted VMs during
instance recovery.