Skip to content

Lazy failover

John Cai requested to merge jc-lazy-failover into master

This is a WIP for discussion purposes to discuss our failover strategy. The approach in this MR is to:

  • Detect when we cannot reach a server
  • Lock the coordinator so no new requests can go through
  • Update the datastore by finding all repositories with the failed connection/storage as its primary by promoting one of its replicas to the primary
  • Retry getting a new backend connection

Merge request reports