Traces carry no task, workflow or step attributes #1803

Closed
opened 2026-11-08 05:30:53 +00:00 by sam · 1 comment
Owner

Archie's OTLP traces carry only gRPC, HTTP and gen_ai attributes. No span names the task, workflow, step or attempt it belongs to, so the trace store can't answer which workflow or step is slow or failing, which is what the traces are being kept for.

Done when the spans for a task's run and its steps carry task id, workflow, step and attempt attributes (and the model and token counts on agent calls), so a trace search or span-metrics dimension can group by them.

Archie's OTLP traces carry only gRPC, HTTP and gen_ai attributes. No span names the task, workflow, step or attempt it belongs to, so the trace store can't answer which workflow or step is slow or failing, which is what the traces are being kept for. Done when the spans for a task's run and its steps carry task id, workflow, step and attempt attributes (and the model and token counts on agent calls), so a trace search or span-metrics dimension can group by them.
Author
Owner

Workflow runs now emit spans (internal/domain/workflow/trace.go, commit 98fb9dfd). workflow.run: archie.task.id, archie.workflow, archie.attempt, archie.outcome (task status; parked is recorded as an error, empty means interrupted). workflow.step per top-level step: same plus archie.step, outcome succeeded/failed/skipped, errors recorded. agent.run per agent call: same plus gen_ai.request.model, gen_ai.usage.input_tokens, gen_ai.usage.output_tokens. Spans use the existing global provider and ctx propagation; they are produced in the worker and reach the backend through the existing container exporter.

Workflow runs now emit spans (internal/domain/workflow/trace.go, commit 98fb9dfd). workflow.run: archie.task.id, archie.workflow, archie.attempt, archie.outcome (task status; parked is recorded as an error, empty means interrupted). workflow.step per top-level step: same plus archie.step, outcome succeeded/failed/skipped, errors recorded. agent.run per agent call: same plus gen_ai.request.model, gen_ai.usage.input_tokens, gen_ai.usage.output_tokens. Spans use the existing global provider and ctx propagation; they are produced in the worker and reach the backend through the existing container exporter.
sam closed this issue 2026-11-08 06:02:25 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: sam/archie#1803