Skip to content

fix(dataproxy): GetActionData returns empty inputs for sub-actions#7337

Merged
pingsutw merged 1 commit intomainfrom
fix/dataproxy-subaction-inputs
May 4, 2026
Merged

fix(dataproxy): GetActionData returns empty inputs for sub-actions#7337
pingsutw merged 1 commit intomainfrom
fix/dataproxy-subaction-inputs

Conversation

@pingsutw
Copy link
Copy Markdown
Member

@pingsutw pingsutw commented May 4, 2026

Tracking issue

n/a

Why are the changes needed?

GetActionData works for root actions but returns empty inputs for sub-actions. Verified against a live cluster:

  • Root action stored RunInfo.InputsUri = s3://flyte-data/uploads/.../offloaded-inputs/<hash> (prefix only).
  • Sub-action stored RunInfo.InputsUri = s3://flyte-data/.../<sub>/inputs.pb (full path).

dataproxy.GetActionData unconditionally appended inputs.pb to the stored URI:

baseRef := storage.DataReference(urisResp.Msg.GetInputsUri())
inputRef, _ := s.dataStore.ConstructReference(groupCtx, baseRef, "inputs.pb")

For root that resolved to the real file; for sub-actions it produced .../inputs.pb/inputs.pb and 404'd silently.

What changes were proposed in this pull request?

Normalize on storing the full file path in RunInfo.InputsUri and let the dataproxy read it verbatim:

  • runs/service/run_service.go: persistRunModel now stores inputPrefix + "/inputs.pb", matching what sub-actions already store via recordSingleAction.
  • dataproxy/service/dataproxy_service.go: drop the ConstructReference(..., "inputs.pb") and read InputsUri directly.

How was this patch tested?

  • Updated unit tests in dataproxy/service/dataproxy_service_test.go and runs/service/run_service_test.go to reflect the new contract.
  • go test ./dataproxy/service/... ./runs/service/... passes.
  • Verified the bug end-to-end against a local cluster (DB + RustFS) before the fix.

Labels

fixed

Check all the applicable boxes

  • I updated the documentation accordingly. (n/a)
  • All new and existing tests passed.
  • All commits are signed-off.

GetActionData returned empty inputs for sub-actions because the dataproxy
unconditionally appended "inputs.pb" to the stored InputsUri. Sub-actions
already stored the full path (.../inputs.pb), so the dataproxy ended up
reading .../inputs.pb/inputs.pb (404). Root actions worked only because
they stored a prefix without the suffix.

Normalize on the full file path: store inputs.pb in the root's
RunInfo.InputsUri and have the dataproxy read the URI verbatim.

Signed-off-by: Kevin Su <pingsutw@apache.org>
@github-actions
Copy link
Copy Markdown

github-actions Bot commented May 4, 2026

🐳 Docker CI Image Built

The CI Docker image has been built and pushed for this PR!

Image: ghcr.io/flyteorg/flyte/ci:pr-7337

This image will be automatically used by CI workflows in this PR.

To test locally:

make gen DOCKER_CI_IMAGE=ghcr.io/flyteorg/flyte/ci:pr-7337

@pingsutw pingsutw changed the base branch from v2 to main May 4, 2026 19:14
@pingsutw pingsutw self-assigned this May 4, 2026
@pingsutw pingsutw added the flyte2 label May 4, 2026
@pingsutw pingsutw added this to the V2 GA milestone May 4, 2026
@pingsutw pingsutw closed this May 4, 2026
@pingsutw pingsutw reopened this May 4, 2026
@pingsutw pingsutw merged commit 94f9cd3 into main May 4, 2026
27 checks passed
@pingsutw pingsutw deleted the fix/dataproxy-subaction-inputs branch May 4, 2026 21:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants