vault-operator icon indicating copy to clipboard operation
vault-operator copied to clipboard

vault-operator erroneously "updates" (kills) active node if it can't be reached/is unhealthy

Open cpick opened this issue 6 years ago • 0 comments

If Vaults.updateLocalVaultCRStatus() can't query a node or determine that it's healthy, Vaults.syncUpgrade() will:

  1. Assume an update is in progress.
  2. Erroneously determine that the active node is the only non-updated node.
  3. Kill it to "complete" the update.

This causes disruption while a standby node takes over and (in installations without auto-unsealing) reduces resiliency by eliminating one of the unsealed, standby nodes.

cpick avatar Sep 14 '18 16:09 cpick