Terminating ongoing invocations

[tillrohrmann]

Based on the updated failure model, it will be required to be able to terminate an invocation that is no longer making progress.

The best form of terminating an ongoing invocation is to cancel the whole call hierarchy starting with the leafs. This will ensure that once a call A is being canceled it knows that all its children calls it has issued are canceled as well. That way the call A only needs to run compensations for successfully completed calls.

In certain cases, it won't be possible to run service code again (e.g. to run the user-code compensations). For example, the permanent decommissioning of a service endpoint is such a case. In such a case it must be possible to forcefully kill an invocation w/o running any user-code. This might leave the service instance state in an inconsistent state and is therefore only the last escape hatch. The way it could work is to complete the service invocation with an exceptional result and aborting any ongoing invocations.

Tasks

Preview Give feedback

1. Implement forceful killing of ongoing invocations #401
   7 of 7
   +0
   

[slinkydeveloper]

We still need to tackle the #402 but we don't need this issue anymore now.