I am trying to understand how decision tables (dmn) can be deployed at runtime on a multi node/camunda instance cluster (with autoscaling and application load balancing) without restarting the instances. Can you please provide some references on it if that is possible?
This is pretty straightfroward - Luckily deployment never requires any kind of restart.
If you have a clustered setup - i.e. All nodes pointing to the same database you can deploy a DMN table via the REST API to any node and ALL nodes will have access to the DMN table immediately.
Is it applicable for both Container managed Engine and Embedded Engine (GreenField)?
Can you point me reference documentation on this?
Yes, storing deployment artifacts in the database is core behavior, not depending on operation mode. The moment the transaction commits, deployed artifacts are available to all nodes.
This section of the docs might help: https://docs.camunda.org/manual/7.14/user-guide/process-engine/deployments/
Isn’t there some kind of cache that must be invalidated after a deployment? Or is a real DB access made every time a process must be startet (a DMN is to be evaluated)?
Is there any api/mechanism which can be used to refresh the cache after new/modified decision tables deployed on Camunda decision engine cluster.
The engine itself takes care of clearing caches after deployment.
There shouldn’t be an issue that needs to be addressed by the user.