acoustic-carpenter-78188
04/21/2026, 9:56 PMRunService.GetActionData (runs/service/run_service.go:741) and dataproxy.Service.GetActionData (dataproxy/service/dataproxy_service.go:444) both do the same job: resolve the input/output URIs for an action and read the protobuf payloads back from blob storage.
The canonical home for blob-backed data access is now the DataProxy service. The RunService.GetActionData RPC is only kept for backward compatibility with existing clients.
Why
Today both services:
• Hold a storage.DataStore and implement parallel reads of inputs.pb / outputs via errgroup.
• Resolve storage URIs from the action record (DataProxy goes through `RunService.GetActionDataURIs`; RunService reads them directly from the DB).
• Must stay in lock-step whenever we evolve the storage layout, artifact types, or error semantics (not found, partial results, trace vs task actions, etc.).
This is duplicate surface area with no good reason now that DataProxy owns data access. It also makes the storage-layout contract implicit in two places.
Proposed change
Refactor RunService.GetActionData into a thin shim that forwards to `DataProxyServiceClient.GetActionData`:
• Inject a dataproxyconnect.DataProxyServiceClient into RunService (likely the in-process one for the all-in-one binary).
• In `GetActionData`:
1. Validate the request as today.
2. Call DataProxyService.GetActionData with the same action_id.
3. Convert the dataproxy.GetActionDataResponse to workflow.GetActionDataResponse (schemas need to be checked for 1:1 compatibility — they currently both use task.Inputs / task.Outputs so this should be trivial).
• Remove the direct dataStore / repo reads from this RPC once delegation is in place. The URI resolution logic already lives behind GetActionDataURIs, which DataProxy calls.
• Keep the runs RPC registered so existing clients keep working.
Open questions
• Response compatibility: confirm dataproxy.GetActionDataResponse ⇄ workflow.GetActionDataResponse field compatibility (both appear to use task.Inputs / `task.Outputs`; double-check trace-action output handling, which is special-cased in the run-service implementation around lines 794-818).
• Whether trace-action output handling (reading from RunInfo.OutputsUri) is already covered on the DataProxy side — if not, that logic needs to move to DataProxy as part of this work, not be left in the run-service shim.
References
• runs/service/run_service.go:741 — current RunService.GetActionData
• dataproxy/service/dataproxy_service.go:444 — DataProxy GetActionData (target delegate)
• #7252 — sibling issue doing the same delegation for TailLogs
flyteorg/flyteacoustic-carpenter-78188
05/01/2026, 5:29 AM