release: workflow runner v0.4.53 for CLI rc.30 - #40
Merged
Conversation
…via exec_session (REQ-052) When a workflow's resolved `environment` is a FULL standalone animus that advertises `environment/exec_session`, hand it the ENTIRE workflow in ONE `EnvironmentClient::exec_session` call instead of running phases locally or routing them per-phase (REQUIREMENT-048). The node runs every phase on its own animus and streams `environment/journal` events home; the runner forwards them through its `WorkflowEventEmitter` and drives the run to terminal from the session's terminal `ExecSessionResponse`. New `src/workflow_session.rs`: - capabilities_advertise_exec_session / environment_is_session_capable — detect a session-capable environment from its plugin manifest `capabilities` list. - map_journal_event_kind / session_status_to_workflow_status — pure, unit-tested mappings from node journal events + terminal status to the home runtime types. - delegate_workflow_via_session — prepare bare node, exec_session with a journal-forwarding closure, teardown, synthesize WorkflowExecuteInternalResult. Branch point added in execute_workflow_with_hub before the REQUIREMENT-048 per-phase environment resolution; only full runs delegate (a `--phase` run keeps the per-phase/local path). Non-session-capable / unset environments fall through UNCHANGED. Gated behind the `remote-animus-session` cargo feature (default OFF): the parent-side exec_session / EnvironmentJournalEvent surface (orchestrator-core REQ-052, commit bcce5012, one commit past the pinned rev d03cd174) and the environment/exec_session protocol types (animus-environment-protocol, not in the pinned tag v0.7.0-rc.6) are not in the current pins. With the feature off the module is inert; the pure detection/mapping logic compiles + is unit-tested unconditionally. needs node/CI build+test — not built locally per policy.
…ion by default - all animus-protocol crates v0.7.0-rc.6 -> v0.7.0-rc.8 (rc.8 = rc.6 + the environment/exec_session commit only — clean, minimal bump) - orchestrator-* animus-cli rev d03cd174 -> 748f684d (env-protocol repinned to rc.8, local exec_session [patch] dropped) - default = [remote-animus-session]: flip the whole-workflow exec_session delegation ON. Needs node/CI build+test (per ao-cli policy).
…on) ON Validated green on GH Actions (all 3 targets). Pins: animus-protocol v0.7.0-rc.8, orchestrator-* animus-cli rev 748f684d.
…on delegation (v0.4.47) REQ-052 exact-once: when the parent runner delegates a whole workflow via exec_session and it completes, delegate_workflow_via_session emitted a terminal journal event but never drove the parent's persisted workflow state machine to terminal. The journal_runs row stayed status=running, so the daemon's journal-resume sweep (resumable_orphans_for_redispatch, past the 90s grace) re-dispatched it as a resumable orphan until a re-dispatch happened to terminalize it -- ~3 runs per dispatch instead of exactly 1. Thread the ServiceHub into delegate_workflow_via_session and, after the session returns, drive the parent state machine terminal best-effort: bounded complete_current_phase_with_decision loop for Completed, mark_completed_failed for Failed/Escalated, cancel for Cancelled. Pure state-machine transitions (no agents/post-success); the terminal event + synthesized result are unchanged.
Whole-workflow exec_session delegation forwarded only lifecycle journal events; the agent-run transcript (output_chunk/tool_call/...) was dropped by map_journal_event_kind and never reached the parent log store, so the portal showed 'No output captured for this run' for remote runs. on_journal now mirrors transcript events into the PARENT run dir first (before the lifecycle map), re-keying the node run id onto the parent workflow id (rekey_transcript_run_id) so the portal's /logs groups them, and persists via ipc::persist_run_event -> the daemon log_storage supervisor ships them to the store the portal reads. Lifecycle stays journal-only (reaches the parent via the upstream proxy). Paired with animus-env-transport bridge change (streams journal/event). Bump v0.4.47 -> v0.4.48.
…TASK-723) Builds on the TASK-722 transcript mirror (f04306b) — combined v0.4.48. - workflow_session passes the parent's workflow_id into exec_session so the node RESUMES the parent's row instead of minting its own (one journal row per delegated dispatch; the node's transcript lands on the id the portal reads). - Terminal-drive Failed/Escalated now calls the guarded force_failed op (was mark_completed_failed, which no-ops on a still-Running row → the backstop was dead); Cancelled uses cancel; both log errors; dropped the racy caller-side get() gate (force_failed self-guards inside the store op). - Repin protocol v0.7.0-rc.8 -> rc.9 and orchestrator-* -> animus-cli ba9b0307; drop the local-build [patch] block.
…t phases commit_implementation_changes no longer bails when the working dir is not a git repo — portal-harness / orchestration agent phases run at a non-repo project root (/app) and edit no files, so requiring a repo there was wrong. No repo (or no pending changes) is now a no-op, not an error. Coding phases still commit + push via code-open-pr. Fixes "implementation phase requires a git repository for commit at /app" that blocked all portal-local agent workflows (TASK-746).
….4.50 The runner pinned ao-cli rev ba9b0307, where environment/prepare shares the 60s ENVIRONMENT_RPC_TIMEOUT — a Railway node (serviceCreate + image pull + deploy + dial-home, default 300s) can NEVER prepare in 60s, so every delegated coding run was killed mid-prepare (leaking a half-created node, or ghosting when the 60s expired pre-create). Repin to 021a874e = ba9b0307 + cherry-picked 4b1b4cc2 (ENVIRONMENT_PREPARE_TIMEOUT=360s, REQUIREMENT-049). Diagnosed via fable+codex. TASK-811.
…gation terminalizes (v0.4.51)
TASK-811 H1: direct_execute emitted workflow_id=None on the Err arm, so a failed
delegated run (e.g. environment/prepare failure) could not be terminalized by the
daemon and ghosted as 'Running' forever (sweep shields delegated). Thread the
resolved workflow id out of execute_workflow_with_hub via a sink set as soon as
the row is minted/loaded; the Err arm now falls back to it. The daemon can now mark
the run Failed and its stderr ('workflow execution failed: <err>') surfaces the
real prepare error. Diagnosed via fable+codex.
…s restart (v0.4.52, rc.28 reconciler companion, TASK-933)
The REQ-052 remote-animus session path (delegate_workflow_via_session) is the
production delegated-coding path: it prepares ONE node, runs the whole workflow
via environment/exec_session, and tears the node down. It persisted NOTHING
about the node, so animus-cli v0.7.0-rc.28's restart reconciler — which reaps /
reattaches a delegated node off SessionCheckpoint.environment — was inert for
coding runs: a daemon restart leaked the node (a full standalone animus keeps
running) and a re-dispatch prepared a SECOND one.
- phase_session: mirror rc.28's on-disk schema into the runner's local
checkpoint module (its on-disk twin): EnvironmentBinding { environment_id,
handle, bound_at, torn_down } + SessionCheckpoint.environment (serde
default/skip, backward compatible), update_session_environment /
mark_environment_torn_down (signature-identical to the daemon twin) +
find_reusable_binding. JSON shape asserted byte-compatible with the released
rc.28 reader.
- workflow_session (remote-animus session, the coding path): after prepare (or
reuse) succeeds and BEFORE exec_session, persist the node binding as a Running
checkpoint at the reconciler's phase (current_phase_id); after a successful
teardown, mark torn_down; on terminal, close the checkpoint. On re-dispatch,
probe the persisted handle's liveness (side-effect-free exec of `true`) and
REUSE the live node (skip prepare) or prepare fresh when dead — mirroring
rc.28's probe_delegate. Composed with v0.4.48/51's terminal-drive (does not
duplicate it); rc.28's auto_resume already SKIPS env-bound checkpoints, so the
synthetic Running checkpoint is never mis-resumed locally.
- Cargo: 0.4.51 -> 0.4.52. KEEPS the line's ao-cli pin 021a874e and
animus-environment-protocol tag v0.7.0-rc.9 (EnvironmentHandle serializes
identically to rc.6/rc.28). No pin changes.
Tests: binding round-trips with rc.28's exact fields + JSON keys; teardown
marks torn_down (idempotent, binding-gated); find_reusable_binding selects
live/matching only; the session path persists a Running checkpoint carrying the
rc.28 shape. cargo build + test (231 lib + integration, 0 failed) + clippy
clean; --no-default-features (feature-off) also builds.
…resume-companion-oneid feat(runner): persist env-node binding + reuse/resume live node across restart (v0.4.52, rc.28 reconciler companion, TASK-933)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Aligns the workflow runner with Protocol v0.7.0-rc.12 and CLI v0.7.0-rc.30. Hardens delegated session lifecycle by invalidating dead resident-host generations without replaying exec_session, propagates actor context across prepare and exec_session with strict node-side ANIMUS_ACTOR_JSON parsing, and preserves paused sessions as resumable live checkpoints.\n\nValidation: cargo fmt, cargo check, clippy -D warnings, release build, and all-target tests (290 passed; 5 pre-existing ignored). Independent review found no blockers.