Akka Supervision Strategies

Speaker: Twitter:

Background

Gossip: Can get information about a node being down
Leader determines membership
- What happens on routing of messages?

Make and order

Send order to restaruant
- Returns estimations
Stateful akka actor

Recovery Points

Persist the state in storage
- Inlcude time of task, status, meta, and host of node
Leader: Gets notified about nodes go down
- Must query database
Recovery

Recovery Node Goes Down

Leader initiates a recovery
- Work is redistributed
Leader is killed
Functionality to find orphaned

Leader Node Goes Down

Recovery timer && Leader goes down
- Restart recovery time started on all active nodes
New Leader
- Reinitiate recovery of "recovery node"

All should be handled by cluster sharding https://doc.akka.io/docs/akka/2.5/cluster-sharding.html

https://manuel.bernhardt.io/2018/02/26/tour-akka-cluster-cluster-sharding/

Questions

Actors per nodes (creating new actors per order)
If an actor is marked as being down and sends a message later .. 2. how is that message handled?
How do you determine failure or just takes too long?
Persistent storage: Does the collapse/compat orders?
Does this work with the akka finite state machine?