Files
cc-ci-orchestrator/cc-ci-plan/launch-upgrader.py
autonomic-bot 28ef7e44ab launch-upgrader: add stall-detect + auto-resume watchdog (opencode-go limit)
The opencode-go subscription's rolling usage-limit (429) ends the 'opencode run'
agent loop mid-run; it does NOT self-resume. Add:
- resume: continue the SAME session (context preserved) via 'opencode run -s <id>
  --continue' — finds the session from the web server, kills the idle proc safely
  (via /proc scan, never pkill -f self-match), relaunches in the tmux session.
- babysit: poll the session log; on a stall (>15min idle) wait out any 429
  retry-after then auto-resume. Spawned automatically by an opencode 'start'.

So a usage-limit pause now self-heals instead of needing a manual nudge.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-23 01:26:24 +00:00

360 lines
18 KiB
Python

#!/usr/bin/env python3
"""
cc-ci upgrader launcher — one-shot weekly recipe-upgrade job agent.
The upgrader runs /upgrade-all to completion, then stops and stays idle so the
run + summary remain viewable in the web UI. The next weekly run starts a fresh
session (start clears any idle/finished session).
Usage:
launch-upgrader.py start use-or-create: leave an in-flight run alone, else start fresh
launch-upgrader.py fresh always kill any existing session and start fresh
launch-upgrader.py stop kill the session
launch-upgrader.py status show session state
launch-upgrader.py attach tmux attach to the session
Env:
LOOP_BACKEND opencode (default) | claude — also accepts UPGRADER_BACKEND
LOOP_MODEL model flag (overrides UPGRADER_MODEL); default tracks backend —
opencode→opencode-go/glm-5.2, claude→sonnet
UPGRADER_MODEL provider/model for opencode, e.g. opencode-go/glm-5.2 (OpenCode Go
subscription) or tinfoil/deepseek-v4-pro; sonnet etc. for claude
UPGRADER_ARGS extra args passed to /upgrade-all (e.g. "n8n ghost", "--dry-run")
claude backend:
CLAUDE_BIN, CLAUDE_FLAGS, REMOTE_CONTROL
opencode backend:
OPENCODE_BIN, OPENCODE_SERVER, OPENCODE_SHARE (1=attach to web server + public --share link)
"""
import os, sys, subprocess, re
from datetime import datetime
from pathlib import Path
# ── config ────────────────────────────────────────────────────────────────────
SESSION = os.environ.get("UPGRADER_SESSION", "cc-ci-upgrader")
WORKDIR = os.environ.get("UPGRADER_DIR", "/srv/cc-ci")
LOG_DIR = os.environ.get("LOG_DIR", "/srv/cc-ci/.cc-ci-logs")
# LOOP_BACKEND / LOOP_MODEL take precedence (unified control from the operator).
# Default backend+model is the OpenCode Go subscription on glm-5.2; override either via env.
# The model default tracks the backend so an explicit `LOOP_BACKEND=claude` (without a model)
# still gets a sensible claude model rather than a glm id.
BACKEND = os.environ.get("LOOP_BACKEND", os.environ.get("UPGRADER_BACKEND", "opencode"))
_DEFAULT_MODEL = "opencode-go/glm-5.2" if BACKEND == "opencode" else "sonnet"
MODEL = os.environ.get("LOOP_MODEL", os.environ.get("UPGRADER_MODEL", _DEFAULT_MODEL))
CLAUDE_BIN = os.environ.get("CLAUDE_BIN", "claude")
CLAUDE_FLAGS = os.environ.get("CLAUDE_FLAGS", "--dangerously-skip-permissions")
REMOTE_CONTROL = os.environ.get("REMOTE_CONTROL", "1") == "1"
OPENCODE_BIN = os.environ.get("OPENCODE_BIN", "/home/loops/.local/bin/opencode")
OPENCODE_SERVER = os.environ.get("OPENCODE_SERVER", "http://127.0.0.1:4096")
# Web visibility for the opencode backend: attach the session to the shared opencode
# web server (viewable at http://oc.commoninternet.net, tailnet-only) AND optionally
# create a public opencode.ai --share link. Default both on so the run is monitorable.
OPENCODE_SHARE = os.environ.get("OPENCODE_SHARE", "1") == "1"
UPGRADER_ARGS = os.environ.get("UPGRADER_ARGS", "")
# ── helpers ───────────────────────────────────────────────────────────────────
def log(msg):
ts = datetime.now().strftime("%H:%M:%S")
print(f"[upgrader {ts}] {msg}", flush=True)
def die(msg):
log(f"ERROR: {msg}")
sys.exit(1)
def session_alive():
return subprocess.run(
["tmux", "has-session", "-t", SESSION], capture_output=True
).returncode == 0
def session_busy():
"""True while a turn is actively in flight (not idle/finished/wedged)."""
r = subprocess.run(["tmux", "capture-pane", "-pt", SESSION],
capture_output=True, text=True)
pane = r.stdout if r.returncode == 0 else ""
return bool(re.search(r"esc to interrupt|⠋|⠙|⠹|⠸|⠼|⠴|⠦|⠧|⠇|⠏|Running tool", pane))
def kill_session():
subprocess.run(["tmux", "kill-session", "-t", SESSION], capture_output=True)
# ── kickoff prompt ────────────────────────────────────────────────────────────
def build_kickoff():
args_note = f" with arguments: {UPGRADER_ARGS}" if UPGRADER_ARGS else ""
return f"""\
*** cc-ci UPGRADER — weekly recipe-upgrade job ***
You are the cc-ci Upgrader: a ONE-SHOT job agent, NOT a perpetual loop. Run the
recipe-upgrade sequence to completion, then STOP. Your cwd is {WORKDIR}; reach the CI
server with `ssh cc-ci`; creds are in {WORKDIR}/.testenv; skills in {WORKDIR}/.claude/skills/.
DO THIS:
1. Invoke the /upgrade-all skill in DEFAULT mode{args_note}
(read {WORKDIR}/.claude/skills/upgrade-all/SKILL.md for the full procedure). It surveys
every enrolled recipe and, for each upgradeable one, runs /recipe-upgrade in DEFAULT
mode — recipe PR only, verified by posting `!testme` on the PR (results visible in the
PR, iterate up to 3x). A genuinely stale test gets an explanatory PR COMMENT, never a
test edit.
2. Process recipes via per-recipe SUBAGENTS so your own context stays light. If your
context usage climbs (~80%), run /compact before continuing.
3. Write + push the weekly summary (the PR list is the actionable output for the operator).
4. WHEN THE RUN IS COMPLETE: STOP. Print the final summary (lead with the PR list) and an
`UPGRADE RUN COMPLETE` line, then go idle. Do NOT loop, do NOT re-run, and do NOT kill
your own session — leave it up so the operator can review the output in the web UI.
Next week's run starts a fresh session (the launcher clears this idle one).
GUARDRAILS: NEVER merge any PR. NEVER weaken a test. DEFAULT mode only — do NOT pass
--with-tests (updating cc-ci tests is the operator's per-recipe opt-in). Single-writer:
dedicated branches + separate clones, never push main, never touch the build loops'
/cc-ci /cc-ci-adv clones. The shared Swarm is stateful — go sequentially.
"""
# ── launch ────────────────────────────────────────────────────────────────────
def start(mode="use-or-create"):
import shutil
if not shutil.which("tmux"):
die("tmux not found")
Path(LOG_DIR).mkdir(parents=True, exist_ok=True)
if session_alive():
if mode == "use-or-create" and session_busy():
log(f"{SESSION} already running a job (busy) — leaving it")
return
log(f"{SESSION} exists but idle/stale (or fresh requested) — killing it first")
kill_session()
import time; time.sleep(1)
kf = Path(LOG_DIR) / f".kickoff-{SESSION}.txt"
kf.write_text(build_kickoff())
model_flag = f"--model '{MODEL}'" if MODEL else ""
log(f"starting {SESSION} (backend={BACKEND}, model={MODEL}, args='{UPGRADER_ARGS or '<none>'}')")
if BACKEND == "claude":
if not shutil.which(CLAUDE_BIN):
die(f"claude CLI not found — set CLAUDE_BIN (currently: {CLAUDE_BIN})")
rc = f"--remote-control '{SESSION}'" if REMOTE_CONTROL else ""
cmd = f"{CLAUDE_BIN} {rc} {model_flag} {CLAUDE_FLAGS} \"$(cat '{kf}')\""
elif BACKEND == "opencode":
if not Path(OPENCODE_BIN).exists():
die(f"opencode not found at {OPENCODE_BIN}")
# NOTE: -m/--model and --attach/--title/--share are flags on the `run` SUBCOMMAND,
# so they must come AFTER `run` (a global `opencode --model X run` is ignored).
share_flag = "--share" if OPENCODE_SHARE else ""
cmd = (
f"set -a; . /srv/cc-ci/.testenv; set +a; "
f"{OPENCODE_BIN} run {model_flag} {share_flag} --attach '{OPENCODE_SERVER}' "
f"--title '{SESSION}' \"$(cat '{kf}')\""
)
log(f" attached to {OPENCODE_SERVER} → http://oc.commoninternet.net (tailnet only)"
+ (" +public --share link (printed in the session)" if OPENCODE_SHARE else ""))
else:
die(f"unknown LOOP_BACKEND '{BACKEND}' — use 'claude' or 'opencode'")
subprocess.run(["tmux", "new-session", "-d", "-s", SESSION, "-c", WORKDIR, cmd])
subprocess.run(["tmux", "pipe-pane", "-o", "-t", SESSION,
f"cat >> '{LOG_DIR}/{SESSION}.log'"])
log(f"started. attach: tmux attach -t {SESSION} log: {LOG_DIR}/{SESSION}.log")
# For the opencode backend, spawn a detached babysitter that auto-resumes the run if the
# opencode-go usage-limit (429) stalls it mid-run (it does NOT self-resume). See babysit().
if BACKEND == "opencode" and os.environ.get("UPGRADER_BABYSIT", "1") == "1":
subprocess.Popen(["python3", os.path.realpath(__file__), "babysit"],
stdout=open(f"{LOG_DIR}/{SESSION}-babysit.log", "a"),
stderr=subprocess.STDOUT, start_new_session=True)
log(" babysitter spawned — auto-resume on usage-limit stalls")
# ── opencode stall-detect + auto-resume watchdog ────────────────────────────────
# The opencode-go subscription enforces a rolling usage-limit (HTTP 429 + retry-after). When it
# trips mid-run, the `opencode run` agent loop ENDS and does NOT self-resume. This watchdog detects
# the stall (the session log stops growing), waits out the limit, and resumes the SAME session —
# context preserved — via `opencode run -s <id> --continue`. Standalone: launch-upgrader.py {resume|babysit}.
import json as _json, urllib.request as _ureq, time as _time
STALL_MIN = float(os.environ.get("UPGRADER_STALL_MIN", "15")) # log-idle minutes ⇒ stalled
CHECK_EVERY = int(os.environ.get("UPGRADER_CHECK_SEC", "180")) # babysitter poll cadence
DONE_MARKER = "UPGRADE RUN COMPLETE"
GO_ENDPOINT = "https://opencode.ai/zen/go/v1/chat/completions"
AUTH_JSON = os.path.expanduser("~/.local/share/opencode/auth.json")
LOG_FILE = f"{LOG_DIR}/{SESSION}.log"
def _server_get(path):
try:
with _ureq.urlopen(OPENCODE_SERVER + path, timeout=15) as r:
return _json.load(r)
except Exception:
return None
def _session_id():
"""Newest top-level opencode session titled like SESSION (the run we manage)."""
rows = _server_get("/session") or []
rows = rows if isinstance(rows, list) else rows.get("data", [])
cands = [s for s in rows if s.get("title") == SESSION and not (s.get("parentID") or s.get("parentId"))]
cands.sort(key=lambda s: (s.get("time") or {}).get("created") or 0, reverse=True)
return cands[0]["id"] if cands else None
def _log_idle_min():
try:
return (_time.time() - os.path.getmtime(LOG_FILE)) / 60.0
except Exception:
return None
def _go_key():
try:
return (_json.load(open(AUTH_JSON)).get("opencode-go") or {}).get("key")
except Exception:
return None
def _limit_retry_after():
"""0 if the opencode-go endpoint is available (HTTP 200); else the 429 retry-after seconds."""
key = _go_key()
if not key:
return 0
body = _json.dumps({"model": (MODEL or "").split("/")[-1] or "glm-5.2", "max_tokens": 8,
"messages": [{"role": "user", "content": "hi"}]}).encode()
req = _ureq.Request(GO_ENDPOINT, data=body, method="POST",
headers={"Authorization": "Bearer " + key, "content-type": "application/json"})
try:
_ureq.urlopen(req, timeout=20).read(); return 0
except _ureq.HTTPError as e:
if e.code == 429:
try: return max(1, int(e.headers.get("retry-after", "300")))
except Exception: return 300
return 0
except Exception:
return 0
def _run_pids():
"""PIDs of live `opencode run` procs (via /proc scan — never matches this process)."""
me, out = os.getpid(), []
for p in os.listdir("/proc"):
if not p.isdigit() or int(p) == me:
continue
try:
cl = open(f"/proc/{p}/cmdline", "rb").read().split(b"\0")
except Exception:
continue
if b"opencode" in (b" ".join(cl)) and b"run" in cl and b"--attach" in cl:
out.append(int(p))
return out
def _completed():
try:
with open(LOG_FILE, errors="ignore") as f:
f.seek(0, 2); f.seek(max(0, f.tell() - 20000)); return DONE_MARKER in f.read()
except Exception:
return False
def resume(reason="manual"):
"""Resume the managed opencode session from where it stopped (context preserved)."""
import signal
sid = _session_id()
if not sid:
log(f"resume: no top-level '{SESSION}' session on {OPENCODE_SERVER} — cannot resume"); return False
log(f"resume ({reason}): continuing session {sid}")
for pid in _run_pids():
try: os.kill(pid, signal.SIGTERM)
except Exception: pass
_time.sleep(2); kill_session(); _time.sleep(1)
kf = Path(LOG_DIR) / f".kickoff-{SESSION}-resume.txt"
kf.write_text(
"The opencode-go usage limit has reset (or the run stalled). You were mid-way through the weekly "
"cc-ci /upgrade-all run. CONTINUE from where you left off — do NOT start over. Process the enrolled "
"recipes not yet done this week, alphabetically; SKIP ones already done (their PRs exist — extend, "
"never duplicate). Per recipe: run /recipe-upgrade in DEFAULT mode via a subagent, verify with "
"!testme, open/extend the recipe PR (NEVER merge, NEVER weaken a test), <= DRONE_RUNNER_CAPACITY "
"concurrent. immich has a tag+digest image abra can't parse — do the upstream-direct cross-check "
"(recipe-upgrade SKILL §1), don't silently skip it. When all remaining recipes are done: "
"write+push the weekly summary, then `python3 /srv/cc-ci/cc-ci-plan/launch-report.py fresh`, print "
"'" + DONE_MARKER + "', and go idle.")
share = "--share" if OPENCODE_SHARE else ""
cmd = (f"set -a; . /srv/cc-ci/.testenv; set +a; {OPENCODE_BIN} run -s {sid} --continue "
f"--model '{MODEL}' {share} --attach '{OPENCODE_SERVER}' \"$(cat '{kf}')\"")
subprocess.run(["tmux", "new-session", "-d", "-s", SESSION, "-c", WORKDIR, cmd])
subprocess.run(["tmux", "pipe-pane", "-o", "-t", SESSION, f"cat >> '{LOG_FILE}'"])
log(f"resume: relaunched {SESSION} (session {sid})"); return True
def babysit():
"""Watch the opencode upgrader; on a stall, wait out any usage-limit then resume the session.
Exits when the run prints UPGRADE RUN COMPLETE. Spawned by an opencode `start`; also standalone."""
log(f"babysit: watching {SESSION} (stall>{STALL_MIN}min log-idle, poll {CHECK_EVERY}s)")
misses = 0
while True:
_time.sleep(CHECK_EVERY)
if _completed():
log("babysit: run completed — exiting"); return
idle = _log_idle_min()
if idle is None or idle <= STALL_MIN:
misses = 0; continue
# Log has been static > STALL_MIN — the run is stalled (limit or wedge), not working.
retry = _limit_retry_after()
if retry > 0:
wait = min(retry + 30, 3600)
log(f"babysit: stalled {idle:.0f}min + usage-limited (retry-after {retry}s) — waiting {wait}s")
_time.sleep(wait); continue
# Not limited but stalled — confirm it's really wedged (two consecutive misses), then resume.
misses += 1
if misses >= 2:
log(f"babysit: stalled {idle:.0f}min, limit clear — auto-resuming")
resume("babysit auto-resume"); misses = 0
# ── main ──────────────────────────────────────────────────────────────────────
def main():
cmd = sys.argv[1] if len(sys.argv) > 1 else "start"
if cmd == "start":
start("use-or-create")
elif cmd == "fresh":
start("fresh")
elif cmd == "stop":
if session_alive():
log(f"killing {SESSION}")
kill_session()
else:
log(f"{SESSION} not running")
elif cmd == "status":
if session_alive():
busy = "busy" if session_busy() else "idle/finishing"
log(f"{SESSION}: RUNNING ({busy})")
subprocess.run(
f"ps -eo pid,etime,args | grep '[r]emote-control {SESSION}' || true",
shell=True)
else:
log(f"{SESSION}: stopped")
log(f"backend: {BACKEND} model: {MODEL} args: '{UPGRADER_ARGS or '<none>'}'")
elif cmd == "attach":
os.execvp("tmux", ["tmux", "attach", "-t", SESSION])
elif cmd == "resume":
resume("manual")
elif cmd == "babysit":
babysit()
else:
print(f"""cc-ci upgrader launcher — one-shot weekly recipe-upgrade job
launch-upgrader.py start use-or-create (leave busy run alone, else start fresh)
launch-upgrader.py fresh always kill existing + start fresh
launch-upgrader.py stop kill the session
launch-upgrader.py status show session state
launch-upgrader.py attach tmux attach
launch-upgrader.py resume continue the opencode session from where it stalled (-s <id> --continue)
launch-upgrader.py babysit watch + auto-resume the opencode run across usage-limit (429) stalls
Backend: {BACKEND} (LOOP_BACKEND or UPGRADER_BACKEND env var)
Model: {MODEL} (LOOP_MODEL or UPGRADER_MODEL env var)
Args: {UPGRADER_ARGS or '<none>'} (UPGRADER_ARGS env var, passed to /upgrade-all)
claude: viewable at claude.ai/code
opencode: viewable at http://oc.commoninternet.net server={OPENCODE_SERVER}
""")
if __name__ == "__main__":
main()