Skip to content

docs: update sdlc-pipeline consensus wrapper for #2806 [doc-updater]#2828

Merged
jwbron merged 1 commit into
mainfrom
egg/doc-update-producer-death-hardstop
May 27, 2026
Merged

docs: update sdlc-pipeline consensus wrapper for #2806 [doc-updater]#2828
jwbron merged 1 commit into
mainfrom
egg/doc-update-producer-death-hardstop

Conversation

@james-in-a-box
Copy link
Copy Markdown
Contributor

Update docs/guides/sdlc-pipeline.md item 6 in the concurrent phase
consensus flow to reflect changes introduced in PR #2811 (fix #2806):

  • MAX_CONSENSUS_RESTARTS bumped from 2 → 3
  • Each wrapper restart now publishes a medium-priority OVERSEER_ALERT
    (anomaly agent-restart) so operators see recovery attempts in real
    time instead of only at exhaustion
  • Producer permanent death now hard-fails the pipeline (transitions to
    FAILED + emits high-priority OVERSEER_ALERT with anomaly
    producer-permanent-death) rather than the generic single-agent HITL
    path; reviewer deaths still use the HITL path

docs/guides/concurrent-execution.md was already updated as part of
#2811.

Triggered by: 352378e Fix #2806: hard-fail pipeline on producer permanent death (#2811)

Authored-by: egg

Update consensus wrapper description (item 6) to reflect changes from
issue #2806 / PR #2811:
- MAX_CONSENSUS_RESTARTS bumped from 2 to 3
- Each restart now emits a medium-priority OVERSEER_ALERT
- Producer permanent death now hard-fails the pipeline (FAILED state)
  with a high-priority OVERSEER_ALERT rather than the generic single-
  agent HITL path; reviewer deaths still use the HITL path
Copy link
Copy Markdown
Contributor

@egg-reviewer egg-reviewer Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No agent-mode design concerns. Documentation-only update synchronizing the consensus wrapper description with the behavior implemented in #2811.

— Authored by egg

@james-in-a-box

This comment has been minimized.

Copy link
Copy Markdown
Contributor

@egg-reviewer egg-reviewer Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Faithful, accurate doc update. Verified each claim against the code from #2811:

  • MAX_CONSENSUS_RESTARTS default 3 — matches orchestrator/consensus_wrapper.py:38.
  • Per-restart medium-priority OVERSEER_ALERT (anomaly agent-restart) — matches orchestrator/consensus_wrapper.py:510-511 (--anomaly agent-restart --priority medium).
  • Producer hard-fail with single re-query (race-window guard) — matches orchestrator/routes/pipelines.py:18068-18099: filtered_graph.is_producer(role_value) gate, executor.check_consensus() recheck, continue on is_complete, otherwise fall through to _emit_producer_death_alert and return 1.
  • FAILED transition + high-priority producer-permanent-death alert with restart_phase / cancel_task guidance — matches _emit_producer_death_alert at orchestrator/routes/pipelines.py:15500-15580 (subject producer-permanent-death: <role> exit=<N>[ slice=<id>] [high], metadata.anomaly_type == "producer-permanent-death", body names both recovery actions).
  • Reviewer-only deaths still flow through the existing single-agent HITL path — matches orchestrator/routes/pipelines.py:18128-18132: handle_agent_failure is only called on the else branch when the role is not a producer.
  • Cross-link target concurrent-execution.md#consensus-wrapper exists at docs/guides/concurrent-execution.md:72, and the sister doc already describes the same behavior at lines 84-85, so the two guides are now consistent.

No correctness, robustness, security, or pattern concerns. Docs-only, single paragraph, no other files touched.

— Authored by egg

@james-in-a-box
Copy link
Copy Markdown
Contributor Author

egg review completed. View run logs

1 previous review(s) hidden.

@jwbron jwbron merged commit 23b65a4 into main May 27, 2026
31 checks passed
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.

Pipeline should auto-mark FAILED when any producer agent dies permanently (retry budget exhausted)

1 participant