Zero-downtime Deployment with Long-Running Orchestrations, Entities and MSSQL storage provider

[conreaux]

Looking for guidance on the best approach for zero-downtime deployment for an application with long-running orchestrations, entities with application-critical state, and using the MSSQL storage provider. Current deployment is on Azure App Service Environment.

It seems the current recommendation is side-by-side deployment using deployment slots and unique TaskHubs. However, this requires the client to have knowledge of which TaskHub the orchestration is assigned to in order to track status. Also, this doesn't account for duplicated entity state, which is not acceptable.

Discussion topic #1781 mentions some options such as using a separate TaskHub for entities (not ideal), external storage provider for entity state (not implemented), and host.json properties (don't appear to apply to the MSSQL provider). These and other recommendations, including those documented here, don't appear to solve the issue.

Temporal has a workflow versioning capability but requires adding conditional logic to the application code, and doesn't appear to be an equivalent in DF.

Any suggestions on a path forward? Also, it may be interesting to use the RideSharing code sample as a concrete example. How could a live deployment of this application be effectively versioned without downtime?