- Modern Applications receive a lot of different signals but must remain responsive to Users while also not losing track of everything that is going on.
- Applications must also deal with real limits on Resources.
- Everything can't run at Maximum CPU/RAM and unlimited Replicas.
- Applications must stay up and maintain consistency even when parts may be over stressed or go down.
- Not only do you not want to lose data but you also don't want to duplicate data if writes are retried or systems go in and out of sync.
- First in/First out
- Allows Applications to decouple Submission of a Task from Handling the Completion of a Job
- Can be used to Distribute Work across Multiple Workers
- Can be Transactional
- Read At Most Once vs At Least Once
- Routing Notifications
- Triggering Tasks
- Recalculation Aggregations when New Data is loaded
- Buffer when one system can send tasks in batches faster than workers can handle.
- Retry or buffer when a dependent system is down.
- Kubernetes Event-driven Autoscaling
- Scale from Zero to N
- Plugins for Scalers
- https://keda.sh/docs/2.8/scalers/
- Like Scalers, KEDA includes several plugins for Authenticating Scalers
- https://keda.sh/docs/2.8/authentication-providers/
