cc-ci/runner/harness/warm.py

"""Warm-infrastructure harness primitive (Phase 2w / WC1+).

Phase 2w keeps a small set of apps "warm" at STABLE domains (distinct from the cold per-run
`<recipe[:4]>-<6hex>` scheme — see DECISIONS.md Phase-2w):

- **live-warm** — actually deployed and running (keycloak today): a shared SSO provider that
  dependent runs use instead of co-deploying a fresh provider. The per-run *realm* (not the app) is
  the isolation unit — created at run start, deleted at run end (see harness.sso WC1 helpers).
- **data-warm** (W1+) — undeployed-when-idle canonicals whose data volume is retained.

This module owns the stable-domain scheme + the "is the warm provider actually usable right now?"
probe + the live-app-hex scan used to reap orphan realms concurrency-safely. It deliberately does NOT
deploy the warm provider — that's the declarative Nix reconciler's job (nix/modules/warm-keycloak.nix).
The harness only *uses* a warm provider when one is up, and falls back to cold co-deploy otherwise.
"""

from __future__ import annotations

import re
import ssl
import subprocess
import urllib.error
import urllib.request

# Recipes that, when declared as a dep, are served from a shared live-warm instance at a stable
# domain instead of being co-deployed per run. Maps dep-recipe -> stable domain.
WARM_DOMAINS = {
    "keycloak": "warm-keycloak.ci.commoninternet.net",
}

# Health probe per warm provider: (path, ok-codes). Mirrors the recipe_meta health contract.
_WARM_HEALTH = {
    "keycloak": ("/realms/master", (200,)),
}

_CTX = ssl.create_default_context()
_CTX.check_hostname = False
_CTX.verify_mode = ssl.CERT_NONE

# A cold per-run stack name looks like "<tag>-<6hex>_ci_commoninternet_net_<svc>"; extract the hex.
_STACK_HEX_RE = re.compile(r"^[a-z0-9]{1,4}-([0-9a-f]{6})_ci_commoninternet_net_")


def warm_domain(recipe: str) -> str | None:
    """The stable warm domain for a dep recipe, or None if this recipe is not served warm."""
    return WARM_DOMAINS.get(recipe)


def is_warm_up(recipe: str, domain: str | None = None, timeout: int = 10) -> bool:
    """True iff the warm provider for `recipe` answers its health endpoint right now. Used to decide
    whether to use the warm path or fall back to cold co-deploy. Conservative: any error → False."""
    domain = domain or warm_domain(recipe)
    if not domain:
        return False
    path, ok = _WARM_HEALTH.get(recipe, ("/", (200, 301, 302)))
    req = urllib.request.Request(f"https://{domain}{path}", method="GET")
    try:
        with urllib.request.urlopen(req, timeout=timeout, context=_CTX) as r:
            return r.status in ok
    except urllib.error.HTTPError as e:
        return e.code in ok
    except Exception:  # noqa: BLE001 — down / unreachable / TLS / DNS → not usable
        return False


def live_app_hexes() -> set[str]:
    """The set of 6hex suffixes of currently-deployed cold per-run app stacks. Used to reap orphan
    realms safely: a realm whose hex maps to a live stack belongs to an in-flight run and is kept.
    Reads docker service names directly so it works even when an app's .env was already removed."""
    out: set[str] = set()
    try:
        res = subprocess.run(
            ["docker", "service", "ls", "--format", "{{.Name}}"],
            capture_output=True,
            text=True,
            timeout=30,
        )
    except Exception:  # noqa: BLE001
        return out
    for name in res.stdout.splitlines():
        m = _STACK_HEX_RE.match(name.strip())
        if m:
            out.add(m.group(1))
    return out


def reap_orphan_realms(recipe: str, domain: str | None = None) -> list[str]:
    """Reap per-run realms on the warm provider left behind by crashed/killed dependent runs. Safe
    under concurrency: realms whose hex maps to a currently-live app stack are kept. Returns the
    realms actually deleted; [] on any error (best-effort run-start cleanup, never fatal)."""
    domain = domain or warm_domain(recipe)
    if recipe != "keycloak" or not domain:
        return []
    from . import sso  # local import avoids import cycle at module load

    try:
        return sso.reap_orphaned_realms(domain, live_app_hexes())
    except Exception:  # noqa: BLE001 — reaping is hygiene, not correctness-critical
        return []


def realm_for(parent_recipe: str, parent_domain: str) -> str:
    """The per-run realm name for a dependent run: "<parent_recipe>-<6hex>" where the 6hex is the
    parent's per-run domain label suffix. Unique per (parent, pr, ref) so concurrent dependents never
    collide on a shared keycloak, and traceable back to the app stack for reaping/debugging."""
    label = parent_domain.split(".", 1)[0]  # "lasu-0a6fb2"
    m = re.search(r"-([0-9a-f]{6})$", label)
    suffix = m.group(1) if m else label
    return f"{parent_recipe}-{suffix}"