"""Issue #63: same-cwd conversation-id resolution in stop/purge/resume and same-type creation ordering for post-discovered agents. """ from __future__ import annotations import json import os import sqlite3 import subprocess import sys import time from pathlib import Path from typing import List, Optional import pytest import yaml REPO = Path(__file__).resolve().parents[1] SKILLS = REPO / ".agents" / "skills" if str(SKILLS) not in sys.path: sys.path.insert(0, str(SKILLS)) UA = "a6300001-0000-4000-8000-000000000001" UB = "b6300002-0000-4000-8000-000000000002" UC = "c6300003-0000-4000-8000-000000000003" UX = "d6300004-0000-4000-8000-000000000004" pytestmark = pytest.mark.usefixtures("own_workspace_root", "no_herdr_probe") @pytest.fixture def no_herdr_probe(monkeypatch): """In-process id_order.wait never asks the real herdr about a dead-owner marker's session: it reads as gone unless a test passes pane_state.""" from lib_py.agents import id_order monkeypatch.setattr(id_order, "herdr_session_state", lambda server, session, timeout: False) # --------------------------------------------------------------------------- helpers def _agy_conv(home: Path, uuid: str, mtime: Optional[float] = None) -> Path: db = Path(home) / ".gemini" / "antigravity-cli" / "conversations" / f"{uuid}.db" db.parent.mkdir(parents=True, exist_ok=True) conn = sqlite3.connect(str(db)) conn.execute("CREATE TABLE steps (id INTEGER)") conn.execute("INSERT INTO steps VALUES (1)") conn.commit() conn.close() if mtime is not None: os.utime(db, (mtime, mtime)) return db def _agy_full_artifacts(home: Path, uuid: str) -> List[Path]: base = Path(home) / ".gemini" / "antigravity-cli" db = _agy_conv(home, uuid) wal = Path(str(db) + "-wal") shm = Path(str(db) + "-shm") wal.write_text("wal") shm.write_text("shm") brain = base / "brain" / uuid brain.mkdir(parents=True, exist_ok=True) (brain / "note.md").write_text("note") ann = base / "annotations" / f"{uuid}.pbtxt" ann.parent.mkdir(parents=True, exist_ok=True) ann.write_text("pbtxt") lock = base / "presence" / f"{uuid}.lock" lock.parent.mkdir(parents=True, exist_ok=True) lock.write_text("") return [db, wal, shm, brain, ann, lock] def _agy_cache(home: Path, mapping: dict) -> None: lc = Path(home) / ".gemini" / "antigravity-cli" / "cache" / "last_conversations.json" lc.parent.mkdir(parents=True, exist_ok=True) lc.write_text(json.dumps({str(k): v for k, v in mapping.items()})) def _claude_transcript(claude_dir: Path, ws: Path, uuid: str, mtime: Optional[float] = None) -> Path: from lib_py.verify_session import workspace_key pdir = Path(claude_dir) / workspace_key(str(ws)) pdir.mkdir(parents=True, exist_ok=True) f = pdir / f"{uuid}.jsonl" f.write_text(json.dumps({"sessionId": uuid, "cwd": str(ws), "type": "user"}) + "\n") if mtime is not None: os.utime(f, (mtime, mtime)) return f def _row(name: str, agent: str, cwd: Path, epoch: int, pid: int = 4242, **extra) -> dict: own = {"agy": "agy_conversation_id_own", "opencode": "opencode_session_id_own", "codex": "codex_session_id_own", "hermes": "hermes_conversation_id_own", "claude": "claude_session_id_own", "grok": "grok_session_id_own"}[agent] row = {"name": name, "status": "running", "role": "creator", "agent": agent, "herdr_session_epoch": epoch, own: None, "session_id_source": "pending-discovery", "session_id_verified": False, "last_visible_status": "unverified", "pane": {"cwd": str(cwd), "pid": pid, "cmd": agent}} row.update(extra) return row # --------------------------------------------------------------------------- 2.1 workspace_uuid def test_find_workspace_uuid_skips_exclude_ids(tmp_path): """rev 2 (B): find_workspace_uuid skips exclude_ids across sibling rows and discover().""" from lib_py.workspace_uuid import find_workspace_uuid home = tmp_path / "home" claude_dir = home / ".claude" / "projects" ws = tmp_path / "ws" ws.mkdir(parents=True) now = time.time() _claude_transcript(claude_dir, ws, UA, mtime=now) _claude_transcript(claude_dir, ws, UB, mtime=now - 10) state = { "herdr_sessions": [ _row("ws-creator-claude", "claude", ws, int(now) - 100, status="stopped", claude_session_id_own=UA, session_id_source="discovered", session_id_verified=True), _row("ws-reviewer-claude", "claude", ws, int(now) - 100, status="running", claude_session_id_own=None), ] } assert find_workspace_uuid(str(ws), "claude", "ws-reviewer-claude", state, str(home), str(claude_dir), exclude_ids=()) == UA assert find_workspace_uuid(str(ws), "claude", "ws-reviewer-claude", state, str(home), str(claude_dir), exclude_ids={UA}) == UB def test_find_workspace_uuid_uses_explicit_home_and_claude_dir(tmp_path, monkeypatch): """rev 2 (B-sub): explicit home and claude_dir are forwarded to tier-1 revalidate.""" from lib_py.workspace_uuid import find_workspace_uuid monkeypatch.setenv("HOME", str(tmp_path / "wrong-home")) monkeypatch.delenv("HOME_DIR", raising=False) monkeypatch.delenv("CLAUDE_PROJECT_DIR", raising=False) home = tmp_path / "custom-home" claude_dir = tmp_path / "custom-claude" ws = tmp_path / "ws" ws.mkdir(parents=True) _agy_conv(home, UA) _claude_transcript(claude_dir, ws, UB) state = { "herdr_sessions": [ _row("ws-creator-agy", "agy", ws, 100, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), _row("ws-creator-claude", "claude", ws, 100, status="stopped", claude_session_id_own=UB, session_id_source="discovered", session_id_verified=True), ] } assert find_workspace_uuid(str(ws), "agy", "ws-creator-agy", state, home=str(home), claude_dir=str(claude_dir)) == UA assert find_workspace_uuid(str(ws), "claude", "ws-creator-claude", state, home=str(home), claude_dir=str(claude_dir)) == UB # --------------------------------------------------------------------------- 2.2 session_ids unit tests @pytest.fixture def owners(monkeypatch): """agy presence ownership without lsof/ps: {uuid: [holder pids]} and a pid -> ppid table the test fills in.""" from lib_py.agents.adapters import agy held, parents = {}, {} def holders(paths): if held.get("__lsof__") is None and "__lsof__" in held: return None return sorted({p for path in paths for u, pids in held.items() if u in path for p in pids}) monkeypatch.setattr(agy, "_holder_pids", holders) monkeypatch.setattr(agy, "_process_parents", lambda: dict(parents) or None) return held, parents def test_session_ids_decide_stop_and_resume_unit(tmp_path, owners): from lib_py.agents.session_ids import decide_resume, decide_stop home = tmp_path / "home" ws = tmp_path / "ws" ws.mkdir(parents=True) now = time.time() epoch = int(now) - 100 _agy_conv(home, UA, mtime=now - 5) _agy_conv(home, UB, mtime=now - 2) _agy_cache(home, {ws: UA}) held, parents = owners held[UA] = [5001] held[UB] = [6001] parents.update({5001: 1, 6001: 1}) state = { "herdr_sessions": [ _row("ws-creator-agy", "agy", ws, epoch, pid=5001), _row("ws-reviewer-agy", "agy", ws, epoch, pid=6001), ] } res_a = decide_stop(state, "ws-creator-agy", "agy", str(ws), True, pane_pid=5001, purge=False, home=str(home)) assert res_a["captured"] == UA and res_a["status"] == "captured" res_b = decide_stop(state, "ws-reviewer-agy", "agy", str(ws), True, pane_pid=6001, purge=False, home=str(home)) assert res_b["captured"] == "" and res_b["status"] == "owner-mismatch" res_dead = decide_stop(state, "ws-reviewer-agy", "agy", str(ws), False, pane_pid=None, purge=True, home=str(home)) assert res_dead["captured"] == "" and res_dead["purge_uuid"] == "" assert "herdr session was already gone" in res_dead["note"] # Two same-type rows in one cwd and B has no own id -> fresh spawn with NOTE state["herdr_sessions"][0]["status"] = "stopped" state["herdr_sessions"][0]["agy_conversation_id_own"] = UA state["herdr_sessions"][1]["status"] = "stopped" r_b = decide_resume(state, "ws-reviewer-agy", "agy", str(ws), herdr_alive=False, home=str(home)) assert r_b["uuid"] == "" and r_b["error"] is None and not r_b["drop_stale_id"] assert any("2 agy sessions share" in n for n in r_b["notes"]) def test_session_ids_own_id_priority_rules_unit(tmp_path): """rev 2 (A): own id never falls through to tier-2; broken registry refuses even when DB is missing.""" from lib_py.agents.session_ids import decide_resume, decide_stop home = tmp_path / "home" ws = tmp_path / "ws" ws.mkdir(parents=True) epoch = int(time.time()) - 100 # Missing UA.db on disk, but both rows claim UA state_dup = { "herdr_sessions": [ _row("ws-creator-agy", "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), _row("ws-reviewer-agy", "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), ] } r_dup = decide_resume(state_dup, "ws-reviewer-agy", "agy", str(ws), herdr_alive=False, home=str(home)) assert r_dup["error"] and "ws-creator-agy" in r_dup["error"] assert r_dup["notes"] == [] s_dup = decide_stop(state_dup, "ws-reviewer-agy", "agy", str(ws), False, purge=True, home=str(home)) assert s_dup["captured"] == "" and s_dup["purge_uuid"] == "" assert "also recorded by ws-creator-agy" in s_dup["note"] # Single row with stale own id UX (missing UX.db) while cache has UC _agy_conv(home, UC) _agy_cache(home, {ws: UC}) state_stale = { "herdr_sessions": [ _row("ws-creator-agy", "agy", ws, epoch, status="stopped", agy_conversation_id_own=UX, session_id_source="discovered", session_id_verified=True), ] } r_stale = decide_resume(state_stale, "ws-creator-agy", "agy", str(ws), herdr_alive=False, home=str(home)) assert r_stale["uuid"] == "" and r_stale["drop_stale_id"] is True and r_stale["error"] is None assert any(f"recorded agy id {UX} no longer verifies" in n for n in r_stale["notes"]) s_stale = decide_stop(state_stale, "ws-creator-agy", "agy", str(ws), False, purge=False, home=str(home)) assert s_stale["captured"] == "" and "no longer verifies; kept as is" in s_stale["note"] # --------------------------------------------------------------------------- Done Criteria 1 & 2 (stop & purge) def _fake_lsof_multi(bindir: Path, mapping: dict) -> None: """Fake lsof that prints the pid from `mapping` ({uuid_substr: pid}) for any matching path.""" cases = "\n".join(f' *{k}*) echo {v}; found=1 ;;' for k, v in mapping.items()) exe = Path(bindir) / "lsof" exe.write_text( "#!/bin/sh\n" "found=0\n" 'for a in "$@"; do\n' ' case "$a" in\n' f"{cases}\n" " esac\n" "done\n" '[ "$found" = "1" ] && exit 0\n' "exit 1\n" ) exe.chmod(0o755) def _mock_herdr_agent(state_file: Path, name: str, cwd: Path, pid: int, epoch: int, agent: str = "agy", pane_id: str = "") -> None: state = json.loads(Path(state_file).read_text()) pid_str = pane_id or f"w1:p_{name}" state.setdefault("agents", {})[name] = { "pid": pid, "cwd": str(cwd), "command": agent, "agent": agent, "pane_id": pid_str, "workspace_id": "w1", "created": epoch, "buffer": f"Antigravity {cwd}\n>", } Path(state_file).write_text(json.dumps(state)) def _write_rows(sandbox: Path, rows: list) -> Path: lib = sandbox / ".agents" / "skills" / "lib.sh" yaml_path = sandbox / ".mam" / "agent-sessions.yaml" mutation = f"d['herdr_sessions'] = {repr(rows)}\n" res = subprocess.run( ["bash", "-c", f'source "{lib}" && atomic_dump_yaml "{yaml_path}"'], input=mutation, capture_output=True, text=True, cwd=str(sandbox), timeout=60, ) assert res.returncode == 0, res.stdout + res.stderr return yaml_path def _read_rows(sandbox: Path) -> dict: yaml_path = sandbox / ".mam" / "agent-sessions.yaml" data = yaml.safe_load(yaml_path.read_text()) or {} return {r["name"]: r for r in data.get("herdr_sessions") or []} def _run_stop(sandbox: Path, session: str, agent: str = "agy", purge: bool = False) -> subprocess.CompletedProcess: script = sandbox / ".agents" / "skills" / "multi-agent-mux-stop" / "scripts" / "stop_session.sh" cmd = ["bash", str(script), "--session", session, "--agent", agent] if purge: cmd += ["--purge-conversation", "--yes"] env = dict(os.environ, SKS_QUIESCENT_TRIES="0", SKS_DIALOG_TIMEOUT="0", SKS_BUSY_TIMEOUT="0") return subprocess.run(cmd, capture_output=True, text=True, cwd=str(sandbox), env=env, timeout=120) @pytest.mark.parametrize("order", [("a_then_b", True), ("b_then_a", True), ("both_dead", False)]) def test_c1_stop_same_cwd_agy_sessions_records_only_verified_owner(mam_sandbox, mock_herdr, order): """Done Criterion 1: two agy sessions in one cwd; cache points to A's UA. Stop A->B and B->A with herdr alive records UA on A and None on B; when herdr is already dead for both, neither records an id and both leave a reason.""" label, alive = order ws = mam_sandbox / "same-ws" ws.mkdir(parents=True, exist_ok=True) now = time.time() epoch = int(now) - 100 _agy_full_artifacts(mam_sandbox, UA) _agy_full_artifacts(mam_sandbox, UB) _agy_cache(mam_sandbox, {ws: UA}) proc_a = subprocess.Popen(["sleep", "120"]) proc_b = subprocess.Popen(["sleep", "120"]) try: _fake_lsof_multi(mam_sandbox / "bin", {UA: proc_a.pid, UB: proc_b.pid}) name_a, name_b = "same-ws-creator-agy", "same-ws-reviewer-agy" _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch, pid=proc_a.pid), _row(name_b, "agy", ws, epoch, pid=proc_b.pid), ]) if alive: _mock_herdr_agent(mock_herdr, name_a, ws, proc_a.pid, epoch, pane_id="w1:p_a") _mock_herdr_agent(mock_herdr, name_b, ws, proc_b.pid, epoch, pane_id="w1:p_b") seq = [name_a, name_b] if label != "b_then_a" else [name_b, name_a] for sname in seq: res = _run_stop(mam_sandbox, sname, "agy") assert res.returncode == 0, res.stdout + res.stderr rows = _read_rows(mam_sandbox) if alive: assert rows[name_a]["agy_conversation_id_own"] == UA assert rows[name_a]["resumable"] is True assert rows[name_b]["agy_conversation_id_own"] is None assert rows[name_b]["last_visible_status"].startswith("stop: ") else: assert rows[name_a]["agy_conversation_id_own"] is None assert rows[name_b]["agy_conversation_id_own"] is None assert "herdr session was already gone" in rows[name_a]["last_visible_status"] assert "herdr session was already gone" in rows[name_b]["last_visible_status"] finally: for p in (proc_a, proc_b): p.kill() p.wait() @pytest.mark.parametrize("broken_dup", [False, True]) def test_c2_purge_never_deletes_another_rows_conversation(mam_sandbox, mock_herdr, broken_dup): """Done Criterion 2: A is stopped with UA; B has no own id (or also has UA in a broken registry) and herdr is dead. Purging B leaves A's conversation intact.""" ws = mam_sandbox / "same-ws" ws.mkdir(parents=True, exist_ok=True) epoch = int(time.time()) - 100 a_files = _agy_full_artifacts(mam_sandbox, UA) _agy_cache(mam_sandbox, {ws: UA}) name_a, name_b = "same-ws-creator-agy", "same-ws-reviewer-agy" row_a = _row(name_a, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True) row_b = _row(name_b, "agy", ws, epoch, status="running", agy_conversation_id_own=(UA if broken_dup else None)) _write_rows(mam_sandbox, [row_a, row_b]) res = _run_stop(mam_sandbox, name_b, "agy", purge=True) assert res.returncode == 0, res.stdout + res.stderr assert "nothing purged" in res.stdout for p in a_files: assert p.exists(), f"A's artifact was deleted by B's purge: {p}" rows = _read_rows(mam_sandbox) assert name_b not in rows assert rows[name_a]["agy_conversation_id_own"] == UA # --------------------------------------------------------------------------- Done Criteria 3, 4, 7 & rev 2 (resume) def _run_resume(sandbox: Path, ws: Path, session: str, agent: str = "agy", dry_run: bool = False) -> subprocess.CompletedProcess: script = sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resume_session.sh" cmd = ["bash", str(script), "--workspace", str(ws), "--agent", agent, "--session", session] if dry_run: cmd.append("--dry-run") return subprocess.run(cmd, capture_output=True, text=True, cwd=str(sandbox), timeout=120) def test_c3_resume_unpinned_second_agy_row_starts_fresh_and_running_branch_keeps_epoch( mam_sandbox, mock_herdr, mock_agents ): """Done Criterion 3: two stopped agy rows in one cwd (A=UA, B=None): dry-run and real resume of B spawn without --conversation UA and emit NOTE; when B's herdr is already alive without an id, resume does not record UA and preserves B's herdr_session_epoch.""" ws = mam_sandbox / "same-ws" ws.mkdir(parents=True, exist_ok=True) epoch_a = int(time.time()) - 300 epoch_b = int(time.time()) - 200 _agy_full_artifacts(mam_sandbox, UA) _agy_cache(mam_sandbox, {ws: UA}) name_a, name_b = "same-ws-creator-agy", "same-ws-reviewer-agy" _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch_a, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), _row(name_b, "agy", ws, epoch_b, status="stopped", agy_conversation_id_own=None), ]) dry = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=True) assert dry.returncode == 0, dry.stdout + dry.stderr assert "--conversation" not in dry.stdout assert f"NOTE: {name_b} has no recorded agy id and 2 agy sessions share" in dry.stderr real = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=False) assert real.returncode == 0, real.stdout + real.stderr rows = _read_rows(mam_sandbox) assert rows[name_b]["status"] == "running" assert rows[name_b]["agy_conversation_id_own"] is None assert "--conversation" not in rows[name_b]["pane"]["cmd_full"] # Running branch: B is already alive in herdr without an own id; epoch must stay fixed. fixed_epoch = 1700000123 _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch_a, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), _row(name_b, "agy", ws, fixed_epoch, status="running", agy_conversation_id_own=None), ]) _mock_herdr_agent(mock_herdr, name_b, ws, 9999, fixed_epoch, pane_id="w1:p_b") alive_res = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=False) assert alive_res.returncode == 0, alive_res.stdout + alive_res.stderr rows = _read_rows(mam_sandbox) assert rows[name_b]["agy_conversation_id_own"] is None assert rows[name_b]["herdr_session_epoch"] == fixed_epoch assert rows[name_b]["last_visible_status"].startswith("already running; no id recorded at ") def test_c4_pinned_post_discovered_and_preassigned_agents_unchanged(mam_sandbox, mock_herdr, mock_agents): """Done Criterion 4: agy/opencode rows with own_id keep it on stop and resume with it; claude/grok stop/resume and find_workspace_uuid behave as before.""" ws = mam_sandbox / "c4-ws" ws.mkdir(parents=True, exist_ok=True) epoch = int(time.time()) - 100 _agy_full_artifacts(mam_sandbox, UA) _claude_transcript(mam_sandbox / ".claude" / "projects", ws, UB) # Seed opencode DB for ses_c4 oc_db = mam_sandbox / ".local" / "share" / "opencode" / "opencode.db" oc_db.parent.mkdir(parents=True, exist_ok=True) conn = sqlite3.connect(str(oc_db)) conn.execute("CREATE TABLE IF NOT EXISTS session_v2 (id TEXT PRIMARY KEY, parent_id TEXT, directory TEXT, time_created INTEGER)") conn.execute("INSERT OR REPLACE INTO session_v2 VALUES (?, NULL, ?, ?)", ("ses_c4", str(ws), epoch * 1000)) conn.commit() conn.close() _write_rows(mam_sandbox, [ _row("c4-creator-agy", "agy", ws, epoch, status="running", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), _row("c4-creator-opencode", "opencode", ws, epoch, status="running", opencode_session_id_own="ses_c4", session_id_source="discovered", session_id_verified=True), _row("c4-creator-claude", "claude", ws, epoch, status="running", claude_session_id_own=UB, session_id_source="assigned", session_id_verified=True), ]) for sname, agent in (("c4-creator-agy", "agy"), ("c4-creator-opencode", "opencode"), ("c4-creator-claude", "claude")): st = _run_stop(mam_sandbox, sname, agent) assert st.returncode == 0, st.stdout + st.stderr rows = _read_rows(mam_sandbox) assert rows["c4-creator-agy"]["agy_conversation_id_own"] == UA and rows["c4-creator-agy"]["resumable"] is True assert rows["c4-creator-opencode"]["opencode_session_id_own"] == "ses_c4" and rows["c4-creator-opencode"]["resumable"] is True assert rows["c4-creator-claude"]["claude_session_id_own"] == UB and rows["c4-creator-claude"]["resumable"] is True r_agy = _run_resume(mam_sandbox, ws, "c4-creator-agy", "agy", dry_run=True) assert r_agy.returncode == 0 and f"--conversation {UA}" in r_agy.stdout r_oc = _run_resume(mam_sandbox, ws, "c4-creator-opencode", "opencode", dry_run=True) assert r_oc.returncode == 0 and "--session ses_c4" in r_oc.stdout r_cl = _run_resume(mam_sandbox, ws, "c4-creator-claude", "claude", dry_run=True) assert r_cl.returncode == 0 and f"-r {UB}" in r_cl.stdout @pytest.mark.parametrize("db_exists", [True, False]) def test_c7_resume_refuses_when_recorded_id_is_claimed_by_another_row( mam_sandbox, mock_herdr, mock_agents, db_exists ): """Done Criterion 7 & rev 2 (A): when both A and B record UA (whether UA.db exists on disk or is missing), resume B refuses with exit 1 before spawning, names A on stderr, and emits no fresh-spawn NOTE.""" ws = mam_sandbox / "c7-ws" ws.mkdir(parents=True, exist_ok=True) epoch = int(time.time()) - 100 if db_exists: _agy_full_artifacts(mam_sandbox, UA) _agy_cache(mam_sandbox, {ws: UA}) name_a, name_b = "c7-ws-creator-agy", "c7-ws-reviewer-agy" _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), _row(name_b, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), ]) before = _read_rows(mam_sandbox) res = _run_resume(mam_sandbox, ws, name_b, "agy", dry_run=False) assert res.returncode == 1, res.stdout + res.stderr assert name_a in res.stderr and UA in res.stderr assert "NOTE:" not in res.stderr state = json.loads(Path(mock_herdr).read_text()) assert name_b not in state.get("agents", {}) assert _read_rows(mam_sandbox) == before def test_rev2_stale_own_id_single_row_never_overwrites_with_cache_and_clears_on_resume( mam_sandbox, mock_herdr, mock_agents ): """rev 2 (A)-3: single agy row with stale own_id UX (UX.db missing) while cache and disk hold UC. Stop keeps UX; resume starts a fresh conversation without UC, emits NOTE, and clears agy_conversation_id_own to None so #40 can pin later.""" ws = mam_sandbox / "stale-ws" ws.mkdir(parents=True, exist_ok=True) epoch = int(time.time()) - 100 _agy_full_artifacts(mam_sandbox, UC) _agy_cache(mam_sandbox, {ws: UC}) name = "stale-ws-creator-agy" _write_rows(mam_sandbox, [ _row(name, "agy", ws, epoch, status="running", agy_conversation_id_own=UX, session_id_source="discovered", session_id_verified=True), ]) st = _run_stop(mam_sandbox, name, "agy") assert st.returncode == 0, st.stdout + st.stderr row_after_stop = _read_rows(mam_sandbox)[name] assert row_after_stop["agy_conversation_id_own"] == UX assert "no longer verifies; kept as is" in row_after_stop["last_visible_status"] res = _run_resume(mam_sandbox, ws, name, "agy", dry_run=False) assert res.returncode == 0, res.stdout + res.stderr assert f"NOTE: {name}'s recorded agy id {UX} no longer verifies" in res.stderr row_after_resume = _read_rows(mam_sandbox)[name] assert row_after_resume["agy_conversation_id_own"] is None assert row_after_resume["session_id_source"] == "pending-discovery" assert row_after_resume["session_id_verified"] is False assert f"recorded id {UX} no longer verified" in row_after_resume["last_visible_status"] assert UC not in row_after_resume["pane"]["cmd_full"] def test_rev2_drop_stale_id_refuses_to_clear_id_claimed_by_another_row(mam_sandbox): """rev 2: update_yaml_resumed.sh --drop-stale-id keeps own_id if another row also claims it.""" ws = mam_sandbox / "dup-drop-ws" ws.mkdir(parents=True, exist_ok=True) epoch = int(time.time()) - 100 name_a, name_b = "dup-creator-agy", "dup-reviewer-agy" _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA), _row(name_b, "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA), ]) upd = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "update_yaml_resumed.sh" res = subprocess.run( ["bash", str(upd), "--session", name_b, "--uuid", "", "--agent", "agy", "--workspace", str(ws), "--drop-stale-id"], capture_output=True, text=True, cwd=str(mam_sandbox), timeout=60, ) assert res.returncode == 0, res.stdout + res.stderr assert _read_rows(mam_sandbox)[name_b]["agy_conversation_id_own"] == UA def test_rev2_claude_unpinned_sibling_row_skips_creator_claimed_id(mam_sandbox, mock_herdr, mock_agents): """rev 2 (B): ws-creator-claude is stopped with UA; ws-reviewer-claude has no own_id (--wrapper row shape); disk has UA.jsonl and UB.jsonl. Resume and stop of ws-reviewer-claude pick UB instead of creator's UA.""" ws = mam_sandbox / "claude-sib-ws" ws.mkdir(parents=True, exist_ok=True) now = time.time() epoch = int(now) - 100 cdir = mam_sandbox / ".claude" / "projects" _claude_transcript(cdir, ws, UA, mtime=now) _claude_transcript(cdir, ws, UB, mtime=now - 10) name_a, name_b = "claude-sib-ws-creator-claude", "claude-sib-ws-reviewer-claude" _write_rows(mam_sandbox, [ _row(name_a, "claude", ws, epoch, status="stopped", claude_session_id_own=UA, session_id_source="discovered", session_id_verified=True), _row(name_b, "claude", ws, epoch, status="running", claude_session_id_own=None), ]) dry = _run_resume(mam_sandbox, ws, name_b, "claude", dry_run=True) assert dry.returncode == 0, dry.stdout + dry.stderr st = _run_stop(mam_sandbox, name_b, "claude") assert st.returncode == 0, st.stdout + st.stderr assert _read_rows(mam_sandbox)[name_b]["claude_session_id_own"] == UB dry_after = _run_resume(mam_sandbox, ws, name_b, "claude", dry_run=True) assert dry_after.returncode == 0, dry_after.stdout + dry_after.stderr assert f"-r {UB}" in dry_after.stdout # --------------------------------------------------------------------------- Done Criterion 6 (id_order & create/resume wait) def test_c6_create_waits_or_refuses_while_same_cwd_post_discovered_row_has_no_id( mam_sandbox, mock_herdr, mock_agents ): """Done Criterion 6 (a) & (b): A is running without an id and has a live pane pid. (a) With MAM_ID_ORDER_WAIT_S=0, create B exits 1 without spawning or adding a row and names A on stderr. (b) With MAM_ID_ORDER_WAIT_S=30, once A records its id in a background thread, create B proceeds (rc 0) and records herdr_session_epoch >= the pin time.""" import threading ws = mam_sandbox / "c6-ws" ws.mkdir(parents=True, exist_ok=True) epoch_a = int(time.time()) - 100 name_a, name_b = "c6-ws-creator-agy", "c6-ws-reviewer-agy" create_sh = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-create" / "scripts" / "create_session.sh" holder = subprocess.Popen(["sleep", "120"]) try: _write_rows(mam_sandbox, [_row(name_a, "agy", ws, epoch_a, pid=holder.pid)]) _mock_herdr_agent(mock_herdr, name_a, ws, holder.pid, epoch_a, pane_id="w1:p_a") # (a) Wait cap 0 -> immediate refusal env0 = dict(os.environ, MAM_ID_ORDER_WAIT_S="0") res0 = subprocess.run( ["bash", str(create_sh), "--workspace", str(ws), "--agent", "agy", "--role", "reviewer", "--session", name_b, "--no-onboard"], capture_output=True, text=True, cwd=str(mam_sandbox), env=env0, timeout=120, ) assert res0.returncode == 1, res0.stdout + res0.stderr assert name_a in res0.stderr assert name_b not in _read_rows(mam_sandbox) herdr_state = json.loads(Path(mock_herdr).read_text()) assert name_b not in herdr_state.get("agents", {}) # Also check resume --dry-run does not wait even with MAM_ID_ORDER_WAIT_S=0 _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch_a, pid=holder.pid), _row(name_b, "agy", ws, epoch_a, status="stopped", agy_conversation_id_own=UB), ]) _agy_full_artifacts(mam_sandbox, UB) dry_res = subprocess.run( ["bash", str(mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resume_session.sh"), "--workspace", str(ws), "--agent", "agy", "--session", name_b, "--dry-run"], capture_output=True, text=True, cwd=str(mam_sandbox), env=env0, timeout=60, ) assert dry_res.returncode == 0, dry_res.stdout + dry_res.stderr # Reset to only A running without id for (b) _write_rows(mam_sandbox, [_row(name_a, "agy", ws, epoch_a, pid=holder.pid)]) pinned_at = [] def _pin_a_later(): time.sleep(1.5) pinned_at.append(int(time.time())) _write_rows(mam_sandbox, [ _row(name_a, "agy", ws, epoch_a, pid=holder.pid, agy_conversation_id_own=UA, session_id_source="discovered", session_id_verified=True), ]) t = threading.Thread(target=_pin_a_later) t.start() try: env30 = dict(os.environ, MAM_ID_ORDER_WAIT_S="30") res_ok = subprocess.run( ["bash", str(create_sh), "--workspace", str(ws), "--agent", "agy", "--role", "reviewer", "--session", name_b, "--no-onboard"], capture_output=True, text=True, cwd=str(mam_sandbox), env=env30, timeout=120, ) finally: t.join(timeout=10) assert res_ok.returncode == 0, res_ok.stdout + res_ok.stderr assert "id-order: waiting for" in res_ok.stderr and name_a in res_ok.stderr rows = _read_rows(mam_sandbox) assert name_b in rows and rows[name_b]["status"] == "running" assert pinned_at and rows[name_b]["herdr_session_epoch"] >= pinned_at[0] finally: holder.kill() holder.wait() def test_c6_id_order_markers_non_blocking_cases_and_uuid_recheck(tmp_path, monkeypatch): """Done Criterion 6 (c), (d), and Criterion 7 uuid recheck during wait + knob parsing.""" from lib_py.agents import id_order home = tmp_path / "home" ws = tmp_path / "ws" ws_other = tmp_path / "ws-other" ws.mkdir(parents=True) ws_other.mkdir(parents=True) yaml_path = tmp_path / "agent-sessions.yaml" yaml_path.write_text(yaml.safe_dump({"herdr_sessions": []})) monkeypatch.setenv("HOME", str(home)) monkeypatch.setenv("AGENT_SESSIONS_YAML", str(yaml_path)) # Knob parsing rules assert id_order.wait_s({}) == 600 assert id_order.wait_s({"MAM_ID_ORDER_WAIT_S": "0"}) == 0 assert id_order.wait_s({"MAM_ID_ORDER_WAIT_S": " 05 "}) == 5 for bad in ("abc", "-1", "9" * 10, ""): assert id_order.wait_s({"MAM_ID_ORDER_WAIT_S": bad}) == 600, bad # (c) Marker-only "being created": blocks while owner_pid is alive, passes once dead dpath = id_order.order_dir(str(ws), "agy", home=str(home)) proc = subprocess.Popen(["sleep", "120"]) try: id_order.write_marker(dpath, "creating-a-agy", proc.pid, str(ws), "agy") assert id_order.wait(str(ws), "agy", "creating-b-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0) == 1 finally: proc.kill() proc.wait() # Dead owner_pid marker is pruned and wait succeeds assert id_order.wait(str(ws), "agy", "creating-b-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0) == 0 # Clean up creating-b-agy's own marker for next sub-checks for f in Path(dpath).glob("*.json"): f.unlink() # (d) Non-blocking cases: claude, different folder, different agent, dead pane pid, pinned row dead_proc = subprocess.Popen(["true"]) dead_proc.wait() live_proc = subprocess.Popen(["sleep", "120"]) try: state_nonblock = { "herdr_sessions": [ _row("other-cwd-agy", "agy", ws_other, 100, pid=live_proc.pid), _row("same-cwd-codex", "codex", ws, 100, pid=live_proc.pid), _row("dead-pane-agy", "agy", ws, 100, pid=dead_proc.pid), _row("pinned-agy", "agy", ws, 100, pid=live_proc.pid, agy_conversation_id_own=UA), ] } yaml_path.write_text(yaml.safe_dump(state_nonblock)) assert id_order.wait(str(ws), "claude", "new-claude", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0) == 0 assert id_order.wait(str(ws), "agy", "new-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0) == 0 # Criterion 7: wait --uuid refuses when another row already claimed that uuid assert id_order.wait(str(ws), "agy", "resuming-agy", os.getpid(), uuid=UA, yaml_path=str(yaml_path), home=str(home), wait_s_val=0) == 1 finally: live_proc.kill() live_proc.wait() def test_portability_python39_ast_and_bash_syntax(): """Static guard: new/updated Python modules parse under Python 3.9 and shell scripts pass bash -n.""" import ast for rel in ( "lib_py/workspace_uuid.py", "lib_py/agents/session_ids.py", "lib_py/agents/id_order.py", ): src = (SKILLS / rel).read_text(encoding="utf-8") ast.parse(src, filename=rel, feature_version=(3, 9)) for rel in ( "lib.sh", "multi-agent-mux-stop/scripts/stop_session.sh", "multi-agent-mux-resume/scripts/resolve_session_id.sh", "multi-agent-mux-resume/scripts/resume_session.sh", "multi-agent-mux-resume/scripts/update_yaml_resumed.sh", "multi-agent-mux-create/scripts/create_session.sh", ): res = subprocess.run(["bash", "-n", str(SKILLS / rel)], capture_output=True, text=True, timeout=30) assert res.returncode == 0, f"{rel}: {res.stderr}" def test_resolve_session_id_without_session_arg_returns_workspace_id(mam_sandbox): """resolve_session_id.sh and decide_resume with name='' return the workspace's stopped row id.""" from lib_py.agents.session_ids import decide_resume ws = mam_sandbox / "opt-sess-ws" ws.mkdir(parents=True, exist_ok=True) epoch = int(time.time()) - 100 _agy_full_artifacts(mam_sandbox, UA) _agy_cache(mam_sandbox, {ws: UA}) _claude_transcript(mam_sandbox / ".claude" / "projects", ws, UB) rows = [ _row("opt-sess-ws-creator-agy", "agy", ws, epoch, status="stopped", agy_conversation_id_own=UA), _row("opt-sess-ws-creator-claude", "claude", ws, epoch, status="stopped", claude_session_id_own=UB), ] _write_rows(mam_sandbox, rows) state = {"herdr_sessions": rows} assert decide_resume(state, "", "agy", str(ws), home=str(mam_sandbox))["uuid"] == UA assert decide_resume( state, "", "claude", str(ws), home=str(mam_sandbox), claude_dir=str(mam_sandbox / ".claude" / "projects"), )["uuid"] == UB script = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-resume" / "scripts" / "resolve_session_id.sh" res_agy = subprocess.run( ["bash", str(script), "--workspace", str(ws), "--agent", "agy"], cwd=str(mam_sandbox), capture_output=True, text=True, timeout=30, ) assert res_agy.returncode == 0 assert res_agy.stdout.strip() == UA res_claude = subprocess.run( ["bash", str(script), "--workspace", str(ws), "--agent", "claude"], cwd=str(mam_sandbox), capture_output=True, text=True, timeout=30, ) assert res_claude.returncode == 0 assert res_claude.stdout.strip() == UB def test_id_order_old_stopped_row_with_own_id_does_not_release_fresh_startup_marker(tmp_path, monkeypatch): """A pre-admission stopped row's own_id must not release a live startup marker for a fresh resume (stale own_id) or a create reusing an old stopped session name.""" import threading from lib_py.agents import id_order from lib_py.agents.session_ids import decide_resume home = tmp_path / "home" ws = tmp_path / "ws" ws.mkdir(parents=True) yaml_path = tmp_path / "agent-sessions.yaml" monkeypatch.setenv("HOME", str(home)) monkeypatch.setenv("AGENT_SESSIONS_YAML", str(yaml_path)) for mode, old_own in (("stale_resume", UX), ("recreate_stopped", UA)): if mode == "recreate_stopped": _agy_conv(home, UA, time.time() - 50) old_epoch = int(time.time()) - 100 old_stop = int(time.time()) - 10 row_a = _row( "same-ws-creator-agy", "agy", ws, old_epoch, status="stopped", agy_conversation_id_own=old_own, stopped_at_epoch=old_stop, ) state = {"herdr_sessions": [row_a]} yaml_path.write_text(yaml.safe_dump(state)) if mode == "stale_resume": dec = decide_resume(state, "same-ws-creator-agy", "agy", str(ws), home=str(home)) assert dec["uuid"] == "" and dec["drop_stale_id"] is True proc_a = subprocess.Popen(["sleep", "120"]) proc_b = subprocess.Popen(["sleep", "120"]) try: # Admit A (fresh startup, uuid=''): writes live marker while old stopped row is in state assert id_order.wait( str(ws), "agy", "same-ws-creator-agy", proc_a.pid, uuid="", yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 0 # Immediate check at cap 0: B must be blocked even before A updates its YAML row assert id_order.wait( str(ws), "agy", "same-ws-reviewer-agy", proc_b.pid, uuid="", yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 1 # Concurrent check: B waits while A transitions old-stopped -> running-unpinned -> running-pinned b_rc: list = [] def _run_b(): rc = id_order.wait( str(ws), "agy", "same-ws-reviewer-agy", proc_b.pid, uuid="", yaml_path=str(yaml_path), home=str(home), wait_s_val=10, poll=0.1, ) b_rc.append(rc) t = threading.Thread(target=_run_b) t.start() time.sleep(0.35) assert b_rc == [], f"B unblocked prematurely while A was still old stopped row ({mode})" # A registers its fresh running row without an own_id running_unpinned = _row( "same-ws-creator-agy", "agy", ws, old_epoch + 100, pid=proc_a.pid, status="running", agy_conversation_id_own=None, ) yaml_path.write_text(yaml.safe_dump({"herdr_sessions": [running_unpinned]})) time.sleep(0.35) assert b_rc == [], f"B unblocked prematurely while A was running unpinned ({mode})" # A records its new conversation id UB -> B unblocks running_pinned = _row( "same-ws-creator-agy", "agy", ws, old_epoch + 100, pid=proc_a.pid, status="running", agy_conversation_id_own=UB, ) yaml_path.write_text(yaml.safe_dump({"herdr_sessions": [running_pinned]})) t.join(timeout=5) assert b_rc == [0], f"B did not unblock after A recorded its new id ({mode})" finally: proc_a.kill() proc_b.kill() proc_a.wait() proc_b.wait() def test_canonical_cwd_distinguishes_underscore_hyphen_and_nested_paths_with_symlink_control(tmp_path, monkeypatch): """Distinct directories whose lossy workspace_key collides (project_a vs project-a, nested/sub vs nested-sub) do not block or merge rows; a symlink alias to the same directory does match.""" from lib_py.agents import id_order from lib_py.agents.session_ids import decide_resume, same_type_rows home = tmp_path / "home" ws_under = tmp_path / "project_a" ws_hyphen = tmp_path / "project-a" ws_nested = tmp_path / "nested" / "sub" ws_flat = tmp_path / "nested-sub" for p in (ws_under, ws_hyphen, ws_nested, ws_flat): p.mkdir(parents=True) ws_alias = tmp_path / "project_a_link" ws_alias.symlink_to(ws_under) yaml_path = tmp_path / "agent-sessions.yaml" monkeypatch.setenv("HOME", str(home)) monkeypatch.setenv("AGENT_SESSIONS_YAML", str(yaml_path)) _agy_conv(home, UA, time.time() - 10) _agy_conv(home, UB, time.time() - 5) _agy_cache(home, {ws_under: UA, ws_hyphen: UB}) proc = subprocess.Popen(["sleep", "120"]) try: state_live = { "herdr_sessions": [ _row("under-agy", "agy", ws_under, 100, pid=proc.pid, status="running"), _row("nested-agy", "agy", ws_nested, 100, pid=proc.pid, status="running"), ] } yaml_path.write_text(yaml.safe_dump(state_live)) # Colliding workspace_key directories do NOT block each other assert id_order.wait( str(ws_hyphen), "agy", "hyphen-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 0 assert id_order.wait( str(ws_flat), "agy", "flat-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 0 # Symlink alias to ws_under DOES see live unpinned under-agy and blocks assert id_order.wait( str(ws_alias), "agy", "alias-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 1 finally: proc.kill() proc.wait() state_stopped = { "herdr_sessions": [ _row("under-agy", "agy", ws_under, 100, status="stopped"), _row("hyphen-agy", "agy", ws_hyphen, 100, status="stopped"), _row("nested-agy", "agy", ws_nested, 100, status="stopped"), _row("flat-agy", "agy", ws_flat, 100, status="stopped"), ] } assert [r["name"] for r in same_type_rows(state_stopped, str(ws_hyphen), "agy")] == ["hyphen-agy"] assert [r["name"] for r in same_type_rows(state_stopped, str(ws_flat), "agy")] == ["flat-agy"] assert [r["name"] for r in same_type_rows(state_stopped, str(ws_alias), "agy")] == ["under-agy"] # Single stopped row in ws_hyphen uses single-row workspace discovery (UB) instead of 2-row fresh spawn res_hyphen = decide_resume(state_stopped, "hyphen-agy", "agy", str(ws_hyphen), home=str(home)) assert res_hyphen["uuid"] == UB assert res_hyphen["notes"] == [] def test_id_order_resolved_but_unrecorded_resume_uuid_reserves_startup_until_recorded(tmp_path, monkeypatch): """When a single stopped post-discovered row with no own_id resolves a uuid from disk, id_order.wait reserves startup (writes a marker) until that uuid is recorded on the row.""" from lib_py.agents import id_order from lib_py.agents.session_ids import decide_resume home = tmp_path / "home" ws = tmp_path / "ws" ws.mkdir(parents=True) yaml_path = tmp_path / "agent-sessions.yaml" monkeypatch.setenv("HOME", str(home)) monkeypatch.setenv("AGENT_SESSIONS_YAML", str(yaml_path)) _agy_conv(home, UA, time.time() - 10) _agy_cache(home, {ws: UA}) old_epoch = int(time.time()) - 100 row_a = _row("same-ws-creator-agy", "agy", ws, old_epoch, status="stopped", agy_conversation_id_own=None) state = {"herdr_sessions": [row_a]} yaml_path.write_text(yaml.safe_dump(state)) dec = decide_resume(state, "same-ws-creator-agy", "agy", str(ws), home=str(home)) assert dec["uuid"] == UA proc_a = subprocess.Popen(["sleep", "120"]) proc_b = subprocess.Popen(["sleep", "120"]) try: # Admit A with resolved (but not yet recorded) uuid UA -> writes startup marker assert id_order.wait( str(ws), "agy", "same-ws-creator-agy", proc_a.pid, uuid=UA, yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 0 # While A has not yet recorded UA on its row, B must be refused at cap 0 assert id_order.wait( str(ws), "agy", "same-ws-reviewer-agy", proc_b.pid, uuid="", yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 1 # Once A records UA on its row (via update_yaml_resumed.sh), B is released recorded_a = _row( "same-ws-creator-agy", "agy", ws, old_epoch + 50, pid=proc_a.pid, status="running", agy_conversation_id_own=UA, ) yaml_path.write_text(yaml.safe_dump({"herdr_sessions": [recorded_a]})) assert id_order.wait( str(ws), "agy", "same-ws-reviewer-agy", proc_b.pid, uuid="", yaml_path=str(yaml_path), home=str(home), wait_s_val=0, ) == 0 finally: proc_a.kill() proc_b.kill() proc_a.wait() proc_b.wait() def test_id_order_dead_owner_marker_blocks_while_its_unregistered_herdr_session_is_alive( tmp_path, monkeypatch, capsys ): """PR #69 review: a create/resume killed after spawning its pane but before registering its row leaves a dead-owner marker. It keeps blocking while the marker's herdr session is alive, goes once the row is registered (the row decides) or the session is gone; a live attempt on the same session name is not pruned; a running row without a pane pid blocks (fail-closed).""" from lib_py.agents import id_order home = tmp_path / "home" ws = tmp_path / "ws" ws.mkdir(parents=True) yaml_path = tmp_path / "agent-sessions.yaml" yaml_path.write_text(yaml.safe_dump({"herdr_sessions": []})) monkeypatch.setenv("HOME", str(home)) monkeypatch.setenv("AGENT_SESSIONS_YAML", str(yaml_path)) dpath = id_order.order_dir(str(ws), "agy", home=str(home)) marker_a = Path(dpath) / "orphan-a-agy.json" alive_sessions = {"orphan-a-agy"} unknown_sessions = set() probes = [] def pane_state(server, session, timeout): probes.append((server, session)) if session in unknown_sessions: return None return session in alive_sessions def wait_b(**kw): return id_order.wait(str(ws), "agy", "new-b-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0, pane_state=pane_state, **kw) dead = subprocess.Popen(["true"]) dead.wait() live = subprocess.Popen(["sleep", "120"]) try: # Orphan: owner dead, no row, herdr session alive -> blocks, marker kept, kill hint printed id_order.write_marker(dpath, "orphan-a-agy", dead.pid, str(ws), "agy", herdr_server="srv-x") capsys.readouterr() assert wait_b() == 1 assert marker_a.exists() assert ("srv-x", "orphan-a-agy") in probes err = capsys.readouterr().err assert "HERDR_SESSION_NAME=srv-x herdr kill-session -t orphan-a-agy" in err # No authoritative answer from herdr -> kept and blocks as well unknown_sessions.add("orphan-a-agy") assert wait_b() == 1 assert marker_a.exists() assert "herdr could not tell whether its session is gone" in capsys.readouterr().err unknown_sessions.clear() # Row registered for the attempt -> marker pruned without a probe; the row decides probes.clear() epoch = int(time.time()) yaml_path.write_text(yaml.safe_dump({"herdr_sessions": [ _row("orphan-a-agy", "agy", ws, epoch, pid=live.pid)]})) assert wait_b() == 1 assert not marker_a.exists() and probes == [] yaml_path.write_text(yaml.safe_dump({"herdr_sessions": [ _row("orphan-a-agy", "agy", ws, epoch, pid=live.pid, agy_conversation_id_own=UA)]})) assert wait_b() == 0 # Session gone -> pruned, B admitted yaml_path.write_text(yaml.safe_dump({"herdr_sessions": []})) id_order.write_marker(dpath, "orphan-a-agy", dead.pid, str(ws), "agy", herdr_server="srv-x") alive_sessions.clear() assert wait_b() == 0 assert not marker_a.exists() # A live attempt on the same session name is kept and blocks; a dead one is taken over same = Path(dpath) / "same-agy.json" id_order.write_marker(dpath, "same-agy", live.pid, str(ws), "agy") assert id_order.wait(str(ws), "agy", "same-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0, pane_state=pane_state) == 1 assert json.loads(same.read_text())["owner_pid"] == live.pid live.kill() live.wait() assert id_order.wait(str(ws), "agy", "same-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=0, pane_state=pane_state) == 0 assert json.loads(same.read_text())["owner_pid"] == os.getpid() same.unlink() # Non-blocking review note: a running same-cwd row without a pane pid blocks no_pid = _row("no-pid-agy", "agy", ws, epoch) no_pid["pane"].pop("pid") yaml_path.write_text(yaml.safe_dump({"herdr_sessions": [no_pid]})) assert wait_b() == 1 finally: live.kill() live.wait() def test_id_order_create_waits_for_orphan_herdr_session_of_a_killed_create( mam_sandbox, mock_herdr, mock_agents ): """PR #69 review reproduction: A's create spawned its pane and was killed before appending its row (dead-owner marker, live herdr agent, no row). create B refuses while A's herdr session is alive and proceeds once it is gone.""" from lib_py.agents import id_order ws = mam_sandbox / "orphan-ws" ws.mkdir(parents=True, exist_ok=True) name_a, name_b = "orphan-ws-creator-agy", "orphan-ws-reviewer-agy" create_sh = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-create" / "scripts" / "create_session.sh" dead = subprocess.Popen(["true"]) dead.wait() holder = subprocess.Popen(["sleep", "120"]) try: dpath = id_order.order_dir(str(ws), "agy", home=str(mam_sandbox)) id_order.write_marker(dpath, name_a, dead.pid, str(ws), "agy", herdr_server="orphan-ws") _mock_herdr_agent(mock_herdr, name_a, ws, holder.pid, int(time.time()), pane_id="w1:p_a") def create_b(): return subprocess.run( ["bash", str(create_sh), "--workspace", str(ws), "--agent", "agy", "--role", "reviewer", "--session", name_b, "--no-onboard"], capture_output=True, text=True, cwd=str(mam_sandbox), env=dict(os.environ, MAM_ID_ORDER_WAIT_S="3"), timeout=120, ) res = create_b() assert res.returncode == 1, res.stdout + res.stderr assert name_a in res.stderr and "kill-session -t " + name_a in res.stderr assert "its herdr session is alive" in res.stderr assert name_b not in _read_rows(mam_sandbox) assert name_b not in json.loads(Path(mock_herdr).read_text()).get("agents", {}) state = json.loads(Path(mock_herdr).read_text()) state["agents"].pop(name_a) Path(mock_herdr).write_text(json.dumps(state)) res = create_b() assert res.returncode == 0, res.stdout + res.stderr assert _read_rows(mam_sandbox)[name_b]["status"] == "running" finally: holder.kill() holder.wait() def test_id_order_herdr_probes_stay_within_the_wait_cap(tmp_path, monkeypatch): """PR #69 review r2 P2: probes of dead-owner markers get only the time left until the wait deadline, shared by all markers, so a slow herdr cannot hold the wait (and its lock) past the cap; cap 0 asks nothing (unknown, kept).""" from lib_py.agents import id_order home = tmp_path / "home" ws = tmp_path / "ws" ws.mkdir(parents=True) yaml_path = tmp_path / "agent-sessions.yaml" yaml_path.write_text(yaml.safe_dump({"herdr_sessions": []})) monkeypatch.setenv("HOME", str(home)) monkeypatch.setenv("AGENT_SESSIONS_YAML", str(yaml_path)) dpath = id_order.order_dir(str(ws), "agy", home=str(home)) dead = subprocess.Popen(["true"]) dead.wait() for name in ("orphan-1-agy", "orphan-2-agy"): id_order.write_marker(dpath, name, dead.pid, str(ws), "agy", herdr_server="srv") budgets = [] def slow_probe(server, session, timeout): budgets.append(timeout) if timeout <= 0: return None time.sleep(min(timeout, 5.0)) return None for cap in (0, 1, 3): budgets.clear() t0 = time.monotonic() rc = id_order.wait(str(ws), "agy", "new-agy", os.getpid(), yaml_path=str(yaml_path), home=str(home), wait_s_val=cap, poll=0.05, pane_state=slow_probe) elapsed = time.monotonic() - t0 bound = cap assert rc == 1, cap assert elapsed < bound + 0.6, (cap, elapsed) assert budgets and max(budgets) <= bound + 0.01, (cap, budgets) if cap == 0: assert elapsed < 0.3 and all(b <= 0 for b in budgets), (elapsed, budgets) assert len(list(Path(dpath).glob("orphan-*.json"))) == 2, cap def test_id_order_real_probe_against_mock_herdr_fails_closed(mam_sandbox, mock_herdr): """PR #69 review r2 P1: lib.sh mam_id_order_wait with the real probe (sandbox copy, mock herdr). A listed orphan blocks; an unlisted one is pruned; a failed `agent list` (server_not_running included) keeps the marker and refuses; with MAM_ID_ORDER_WAIT_S=0 herdr is not asked and the marker is kept.""" from lib_py.agents import id_order ws = mam_sandbox / "probe-ws" ws.mkdir(parents=True, exist_ok=True) name_a, name_b = "probe-ws-creator-agy", "probe-ws-reviewer-agy" lib = mam_sandbox / ".agents" / "skills" / "lib.sh" dpath = id_order.order_dir(str(ws), "agy", home=str(mam_sandbox)) marker_a = Path(dpath) / f"{name_a}.json" dead = subprocess.Popen(["true"]) dead.wait() holder = subprocess.Popen(["sleep", "120"]) def set_mock(**kw): state = json.loads(Path(mock_herdr).read_text()) state.pop("force_error_code", None) state["agents"].pop(name_a, None) state.update(kw) Path(mock_herdr).write_text(json.dumps(state)) def wait_b(cap="2"): for f in Path(dpath).glob(f"{name_b}.json"): f.unlink() id_order.write_marker(dpath, name_a, dead.pid, str(ws), "agy", herdr_server="probe-ws") return subprocess.run( ["bash", "-c", 'source "$1" && mam_id_order_wait "$2" agy "$3"', "_", str(lib), str(ws), name_b], capture_output=True, text=True, cwd=str(mam_sandbox), timeout=120, env=dict(os.environ, MAM_ID_ORDER_WAIT_S=cap, HERDR_SESSION_NAME="probe-ws"), ) try: _mock_herdr_agent(mock_herdr, name_a, ws, holder.pid, int(time.time()), pane_id="w1:p_a") res = wait_b() assert res.returncode == 1 and "its herdr session is alive" in res.stderr, res.stderr assert marker_a.exists() set_mock(force_error_code="protocol_mismatch") res = wait_b() assert res.returncode == 1, res.stderr assert "herdr could not tell whether its session is gone" in res.stderr assert marker_a.exists() set_mock(force_error_code="server_not_running") res = wait_b() assert res.returncode == 1, res.stderr assert "herdr could not tell whether its session is gone" in res.stderr assert marker_a.exists() set_mock() res = wait_b(cap="0") assert res.returncode == 1, res.stderr assert "MAM_ID_ORDER_WAIT_S=0 leaves no time to ask herdr" in res.stderr assert marker_a.exists() res = wait_b() assert res.returncode == 0, res.stderr assert not marker_a.exists() finally: holder.kill() holder.wait()