Skip to content

feat(web): add a run pipeline view for node status, timing, tokens and failures - #1198

Open
linus-liu-web wants to merge 1 commit into
Tencent:mainfrom
linus-liu-web:feat/run-pipeline-view
Open

linus-liu-web wants to merge 1 commit into
Tencent:mainfrom
linus-liu-web:feat/run-pipeline-view

Conversation

@linus-liu-web

Copy link
Copy Markdown

What changed

On a run detail page, "what did this run actually do" is spread across tabs and interactions: node status lives on the graph, timing/tokens only show for the selected node, artifacts need a node click, a failed node shows a red status but the reason is buried, and events are mixed into the live trace. On a 7-node SOP you check four places before you can answer "which step failed, how long did each take, how many tokens did it cost".

Add a 流水 (pipeline) tab to the run detail view that aggregates data the API already returns:

  • Summary strip: run status, completed/failed counts, wall time, total tokens, tool calls, retries/rework, artifact count and size.
  • Node rows in workflow order: status, duration, attempts/rework, tokens, tool calls, artifact count + size, and the current activity or the failed node's real error. Nodes whose workflow definition changed are appended rather than dropped.
  • Expandable per-node detail: full error, current activity, artifacts (opens the existing artifact preview) and a truncated node output.
  • Run event timeline: the last 40 run events in one list.

Frontend-only: reuses GET /api/runs/:id (nodes with timing/usage/error/activity, artifacts metadata, events, usage), so no API, schema or engine change is needed. #1191 (capturing the real script/gate error into the node error) is still a separate backend fix; once it lands this view will show that error without further changes.

Related: implements the proposal in #1197; complements #1101 / #1102 (run-detail truncation and artifact card layout) without changing those areas.

Scope

  • Platform core / API / runner
  • Public adapter / deployment
  • Classic
  • Portable
  • Shared behavior contract
  • Installer / CLI
  • Documentation only

Compatibility

Frontend-only change: no API or data-store change, and no existing tab or action is modified. The new tab is read-only; runs without node state render an empty-state message. observability.css and the web/index.html asset version are bumped so the new styles/scripts load, matching the repository's cache-busting convention.

Validation

  • go vet ./web passes.
  • New node test go test ./web -run TestRunPipelinePane -count=1 passes: duration/byte formatting, workflow-order row ordering (including a run node missing from the workflow), summary math (nodes/tokens/tool calls/retries/artifacts), failed-node error rendering, running-node progress state, and timeline/expanded-detail markup.
  • go test ./web -count=1 (full package) passes.
  • python3 scripts/check-public.py passes.
  • Not run: the browser acceptance suite (e2e/public) and a live platform deployment; this is a read-only view covered by the node test above.

Safety

  • I did not include credentials, internal endpoints, production data, or organization-specific infrastructure.
  • I preserved user files and unrelated changes.
  • I updated generated Classic host bundles when applicable. (n/a — no Classic change)
  • I documented third-party content and licensing changes when applicable. (n/a)

…d failures

On a run detail page the information about "what did this run actually do" is
spread across tabs and interactions: node status lives on the graph, timing and
tokens only show for the selected node, artifacts need a node click, failures
show as a red status without the reason, and events are mixed into the live
trace. On a seven-node SOP this is four places to check before you can answer
"which step failed, how long did each take, how many tokens did it cost".

Add a 流水 (pipeline) tab that aggregates the data already returned by
GET /api/runs/:id:

- a summary strip: run status, completed/failed node counts, wall time, total
  tokens, tool calls, retries/rework and artifact count + size;
- one row per node in workflow order: status, duration, attempts/rework,
  tokens, tool calls, artifact count + size, plus the current activity or the
  failed node's real error;
- expandable per-node detail: full error, current activity, artifacts (opening
  the existing artifact preview) and a truncated node output;
- a run event timeline (last 40 events) rendered in one place.

Frontend-only: no API, schema or engine change; it uses the node timing/usage/
error/activity fields, artifact metadata, event list and run usage that the run
endpoint already returns.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant