Skip to content

Latest commit

 

History

History
19 lines (11 loc) · 1.12 KB

high-availability.md

File metadata and controls

19 lines (11 loc) · 1.12 KB
Raw

High Availability

freno offers two alternate methods for high availability:

  1. raft consensus, where freno nodes are all raft nodes and coordinate state with each other.
  2. MySQL based, where freno uses a MySQL backend for state and leadership resolution. In this setup freno assumes MySQL is available.

MySQL backend adds MySQL as a dependency, and also requires the user to maintain high availability for MySQL itself. It makes sense in environments where MySQL HA is a solved problem.

raft has less dependencies, but is also more difficult to deploy on some setups, namely Kubernetes, because of the strict need for nodes to explicitly know each other by name/IP.

Force leadership

It is possible to instruct a freno daemon to assume it is the leader, no matter what consensus says.

  • Provide the --force-leadership flag.
  • This does not affect other nodes. Another node may also believe its the leader, either because of consensus or because of similar configuration.

This flag can be used in emergency cases where consensus cannot be established, due to hardware/network issues.