# Auto Replenish Replicas when Servers Down
+ Replenish process
  + `PD` senses `TiKV` failure based on `heartbeat`
  + Mark all the `Regions` of the failed node as missing replica state
  + Find a new node to add replica based on label constraints
+ To avoid unnecessary waste of resources, there is a delay in replenishing replica (30 minutes by default)
+ Best practice when `TiKV` is down
  + Try to bring it back as soon as possible, adjust the **`max-downtime`** config if necessary
  + When recovery is NOT possible, set the store to **`Offline`** state
  + Reserve storage space for failover

------------------------------------------------------------