Skip to content

Latest commit

 

History

History
37 lines (22 loc) · 2.47 KB

leader_election.md

File metadata and controls

37 lines (22 loc) · 2.47 KB
Raw

Leading ETCD main container’s sidecar is the backup leader

  • The backup-restore sidecar poll its corresponding etcd main container to see if it is the leading member in the etcd cluster. This information is used by the backup-restore sidecars to decide that sidecar of the leading etcd main container is the backup leader.

  • Only backup leader sidecar among the members have the responsibility to take/upload the snapshots(full as well as incremental) for a given Etcd cluster as well as to trigger the defragmentation for each Etcd cluster member.

Work flow

Backup-restore can be in following 3 states:

  1. Follower:
    • When the corresponding etcd main container is a Follower etcd.
    • The default state for every backup-restore member is the Follower State.
  2. Leader:
    • When the corresponding etcd main container becomes the leader then sidecar will also become leading sidecar.
  3. Unknown:
    • When there is no etcd leader present in the cluster or in the case of a quorum loss.
    • When corresponding etcd main container is down and endpoint status api call fails.

Backup

  • Only backup leader among the backup-restore members have the responsibility to take/upload the snapshots(full as well as incremental) for a given Etcd cluster.
  • The backup leader also has the responsibility to garbage-collect the backups from the object storage bucket according to a configured garbage collection policy.

Member-lease

Each backup-restore member has the responsibility to renew its member lease periodically. This is intended to simulate as heartbeat which indicate that the backup-restore cluster members are in Healthy State.

Defragmentation

Defragmentation for all etcd cluster members is triggered by the leading backup-restore sidecar. The defragmentation is performed only when etcd cluster is in full health and it is done in a rolling manner for each member to avoid disruption.At first, leading backup-restore sidecar triggers defragmentation on all etcd follower members one by one and at last, on itself.

Complete work flow leader-election state diagram.

leader-election