shaiwatcher/wrapper/wrapper.py

import os, sys, time, shutil, subprocess, signal, json, pathlib, re, datetime
from typing import Tuple

# ---------- Config (env) ----------
REPO_URL    = os.getenv("REPO_URL", "").strip()
REPO_BRANCH = os.getenv("REPO_BRANCH", "main").strip()
REPO_TOKEN  = os.getenv("REPO_TOKEN", "").strip()  # optional

# prefer CHECK_TIME_UTC; fall back to old RECHECK_UTC
CHECK_TIME_UTC = os.getenv("CHECK_TIME_UTC", os.getenv("RECHECK_UTC", "03:00")).strip()

# 1 = ignore test-only bumps (e.g. v1.2.3.4-T3 -> v1.2.3.4-T4)
IGNORE_TEST_LEVEL = os.getenv("IGNORE_TEST_LEVEL", "1").strip() == "1"

PIP_INSTALL = os.getenv("PIP_INSTALL_REQUIREMENTS", "1").strip() == "1"

CACHE_DIR = pathlib.Path("/cache/app")
TMP_DIR   = pathlib.Path("/cache/tmp")
PREV_DIR  = pathlib.Path("/cache/prev")
DATA_DIR  = pathlib.Path("/data")

RUN_TIMEOUT = int(os.getenv("WRAPPER_STOP_TIMEOUT", "25"))
ROLLBACK_MAX_FAILS = 3

# ---------- Helpers ----------
def log(msg: str):
    print(f"[wrapper] {msg}", flush=True)

def run(*cmd, cwd=None, check=True) -> subprocess.CompletedProcess:
    log(f"$ {' '.join(cmd)}")
    return subprocess.run(cmd, cwd=cwd, check=check, text=True, capture_output=True)

def ensure_git():
    try:
        run("git","--version")
    except subprocess.CalledProcessError as e:
        log(f"git missing? {e.stderr}")
        sys.exit(1)

def utc_now() -> datetime.datetime:
    return datetime.datetime.utcnow()

def next_utc(hhmm: str) -> float:
    hh, mm = map(int, hhmm.split(":"))
    now = utc_now()
    tgt = now.replace(hour=hh, minute=mm, second=0, microsecond=0)
    if tgt <= now:
        tgt = tgt + datetime.timedelta(days=1)
    return (tgt - now).total_seconds()

_VERSION_RE = re.compile(r'^\s*VERSION\s*=\s*[\'"]([^\'"]+)[\'"]', re.M)
def extract_version_from(path: pathlib.Path) -> str:
    try:
        txt = path.read_text(encoding="utf-8", errors="ignore")
        m = _VERSION_RE.search(txt)
        return m.group(1).strip() if m else "v0.0.0.0"
    except Exception:
        return "v0.0.0.0"

def parse_version(ver: str) -> Tuple[int,int,int,int,bool]:
    # Format: vMajor.Minor.Enh.Patch[-T...]
    # Example: v1.2.3.4-T7
    test = "-T" in ver
    core = ver.split("-T")[0].lstrip("v")
    parts = [int(p or 0) for p in core.split(".")+["0","0","0","0"]][:4]
    return parts[0], parts[1], parts[2], parts[3], test

def should_update(old: str, new: str) -> bool:
    """
    Update if the numeric tuple increases.
    Ignore updates that change *only* the test suffix (e.g., v1.2.3.4-T1 -> v1.2.3.4-T2).
    """
    oM,oE,oN,oP,ot = parse_version(old)
    nM,nE,nN,nP,nt = parse_version(new)
    if (oM,oE,oN,oP) != (nM,nE,nN,nP):
        return True
    # numeric parts same → only test suffix changed
    return not IGNORE_TEST_LEVEL and (ot != nt)

def clone_or_fetch(target: pathlib.Path):
    if target.exists() and (target / ".git").exists():
        try:
            run("git","fetch","--all","-p", cwd=target)
            run("git","reset","--hard", f"origin/{REPO_BRANCH}", cwd=target)
            return
        except Exception as e:
            log(f"fetch failed, recloning: {e}")
            shutil.rmtree(target, ignore_errors=True)

    url = REPO_URL
    # optional token (only if provided)
    if REPO_TOKEN and REPO_URL.startswith("https://"):
        url = REPO_URL.replace("https://", f"https://{REPO_TOKEN}@")
    run("git","clone","--branch",REPO_BRANCH,"--depth","1", url, str(target))

def prime_tmp_then_decide():
    # Pull to TMP, compare versions (bot.py VERSION), decide if we swap CACHE
    TMP_DIR.mkdir(parents=True, exist_ok=True)
    shutil.rmtree(TMP_DIR, ignore_errors=True)
    clone_or_fetch(TMP_DIR)

    new_ver = extract_version_from(TMP_DIR / "bot.py")
    old_ver = extract_version_from(CACHE_DIR / "bot.py") if (CACHE_DIR / "bot.py").exists() else "v0.0.0.0"
    log(f"cached version: {old_ver}  /  remote version: {new_ver}")

    if not (CACHE_DIR / "bot.py").exists():
        # First time
        do_swap = True
        reason = f"first fetch -> {new_ver}"
    else:
        do_swap = should_update(old_ver, new_ver)
        reason = f"update allowed: {do_swap} (old={old_ver}, new={new_ver})"

    return do_swap, old_ver, new_ver, reason

def swap_cache_to_prev():
    PREV_DIR.mkdir(parents=True, exist_ok=True)
    shutil.rmtree(PREV_DIR, ignore_errors=True)
    if CACHE_DIR.exists():
        shutil.move(str(CACHE_DIR), str(PREV_DIR))

def copy_tmp_to_cache():
    shutil.rmtree(CACHE_DIR, ignore_errors=True)
    shutil.copytree(TMP_DIR, CACHE_DIR, dirs_exist_ok=False)

def pip_install(cwd: pathlib.Path):
    req = cwd / "requirements.txt"
    if PIP_INSTALL and req.exists():
        try:
            run(sys.executable, "-m", "pip", "install", "-r", str(req))
        except subprocess.CalledProcessError as e:
            log("pip install failed (will continue anyway)")
            log(e.stdout + "\n" + e.stderr)

def set_boot_env(status: str, old_ver: str, new_ver: str, commit: str = "", subject: str = ""):
    # Env passed to the bot; bot should read and post to modlog on_ready
    os.environ["SHAI_BOOT_STATUS"]  = status
    os.environ["SHAI_BOOT_OLDVER"]  = old_ver
    os.environ["SHAI_BOOT_NEWVER"]  = new_ver
    os.environ["SHAI_BUILD_COMMIT"] = commit
    os.environ["SHAI_BUILD_SUBJECT"]= subject

def get_head_info(cwd: pathlib.Path) -> Tuple[str,str]:
    try:
        c1 = run("git","rev-parse","--short","HEAD", cwd=cwd, check=True)
        sha = c1.stdout.strip()
        c2 = run("git","log","-1","--pretty=%s", cwd=cwd, check=True)
        subj = c2.stdout.strip()
        return (sha, subj)
    except Exception:
        return ("", "")

def start_bot(cwd: pathlib.Path) -> subprocess.Popen:
    env = os.environ.copy()
    # Make sure data dir exists (the bot should use SHAI_DATA or SHAI_DATA_FILE or config)
    env.setdefault("SHAI_DATA", "/data/data.json")
    # Run from the cached code directory
    return subprocess.Popen([sys.executable, "-u", "bot.py"], cwd=cwd, env=env)

def graceful_restart(proc: subprocess.Popen) -> bool:
    try:
        proc.terminate()
        try:
            proc.wait(timeout=RUN_TIMEOUT)
            return True
        except subprocess.TimeoutExpired:
            proc.kill()
            proc.wait(timeout=10)
            return True
    except Exception:
        return False

def run_loop():
    ensure_git()

    # initial fetch/decide
    updated, old_ver, new_ver, reason = prime_tmp_then_decide()
    if updated:
        log(f"updating cache: {reason}")
        swap_cache_to_prev()
        copy_tmp_to_cache()
    else:
        log(f"no update: {reason}")

    # pip install if needed (requirements.txt inside repo)
    pip_install(CACHE_DIR)

    # boot status env
    sha, subj = get_head_info(CACHE_DIR)
    if updated:
        set_boot_env(
            f"Successfully fetched, cached, and booted new version",
            old_ver, new_ver, sha, subj
        )
    else:
        msg = "Successfully booted from cached version"
        if sha or subj:
            msg += " (repo reachable)"
        set_boot_env(msg, old_ver, new_ver, sha, subj)

    # start bot
    proc = start_bot(CACHE_DIR)
    log(f"bot started pid={proc.pid}")

    consecutive_failures = 0

    while True:
        # sleep until next check (03:00 UTC by default)
        delay = next_utc(CHECK_TIME_UTC)
        log(f"sleeping {int(delay)}s until {RECHECK_UTC} UTC for update check")
        time.sleep(delay)

        # check for update
        try:
            upd, cur_ver, remote_ver, why = prime_tmp_then_decide()
            log(f"nightly check: {why}")
            if not upd:
                # no update -> continue loop
                continue

            # graceful restart into new version
            log("updating to new version at nightly window")
            ok = graceful_restart(proc)
            if not ok:
                log("warning: bot did not stop cleanly")

            # swap and boot
            swap_cache_to_prev()
            copy_tmp_to_cache()
            pip_install(CACHE_DIR)
            sha, subj = get_head_info(CACHE_DIR)
            set_boot_env("Successfully fetched, cached, and booted new version", cur_ver, remote_ver, sha, subj)
            proc = start_bot(CACHE_DIR)
            log(f"bot restarted on new version pid={proc.pid}")
            consecutive_failures = 0

        except Exception as e:
            log(f"nightly update failed: {e}")
            consecutive_failures += 1
            if consecutive_failures < ROLLBACK_MAX_FAILS and PREV_DIR.exists() and (PREV_DIR / "bot.py").exists():
                log("attempting rollback to previous cached version")
                if proc.poll() is None:
                    graceful_restart(proc)
                shutil.rmtree(CACHE_DIR, ignore_errors=True)
                shutil.copytree(PREV_DIR, CACHE_DIR, dirs_exist_ok=False)
                pip_install(CACHE_DIR)
                set_boot_env("Rolled back to last known working version", "-", extract_version_from(CACHE_DIR / "bot.py"))
                proc = start_bot(CACHE_DIR)
            elif consecutive_failures >= ROLLBACK_MAX_FAILS:
                log("critical: failed 3 times to update/restart; entering freeze mode")
                # Optional: DM owner could be done in a tiny fallback bot here if OWNER_ID provided.
                # For now, just idle to allow SSH/exec into container.
                try:
                    if proc.poll() is None:
                        graceful_restart(proc)
                except Exception:
                    pass
                while True:
                    time.sleep(3600)

if __name__ == "__main__":
    run_loop()