furtka/furtka/reconciler.py

from dataclasses import dataclass
from pathlib import Path

from furtka import deps, dockerops
from furtka.manifest import ManifestError, load_manifest
from furtka.scanner import scan

_ON_START_TIMEOUT_SECONDS = 30.0


@dataclass(frozen=True)
class Action:
    kind: str  # "ensure_volume" | "compose_up" | "hook" | "skip" | "error"
    target: str
    detail: str = ""

    def describe(self) -> str:
        if self.detail:
            return f"{self.kind:14s} {self.target}  ({self.detail})"
        return f"{self.kind:14s} {self.target}"


def reconcile(apps_root: Path, dry_run: bool = False) -> list[Action]:
    """Walk the apps tree and bring docker into the desired state.

    Apps are visited in dependency order — providers before consumers — so a
    consumer's `on_start` hook runs against an already-up provider. Within a
    tier, order stays alphabetical for deterministic boot logs. Apps with
    unresolvable `requires` (missing provider, broken manifest cycle) are
    visited last; reconcile's per-app isolation still kicks in if they fail.

    Failures during one app's reconcile (Docker errors, missing binary, …) are
    captured as Action(kind='error', …) and do NOT abort the whole sweep — the
    other apps still get reconciled. Callers inspect the returned actions to
    decide overall success.
    """
    actions: list[Action] = []
    results = scan(apps_root)
    for result in deps.installed_topo_order(results):
        if not result.ok:
            actions.append(Action("skip", result.path.name, result.error or ""))
            continue
        m = result.manifest
        try:
            for vol_short in m.volumes:
                full = m.volume_name(vol_short)
                actions.append(Action("ensure_volume", full))
                if not dry_run:
                    dockerops.ensure_volume(full)
            hook_failed = False
            for req in m.requires:
                if not req.on_start:
                    continue
                hook_label = f"{m.name}:{req.app}:on_start"
                actions.append(Action("hook", hook_label, req.on_start))
                if dry_run:
                    continue
                try:
                    _fire_on_start_hook(m, req, apps_root)
                except (
                    dockerops.DockerError,
                    FileNotFoundError,
                    OSError,
                    ManifestError,
                ) as e:
                    actions.append(
                        Action("error", m.name, f"on_start({req.app}): {e}")
                    )
                    hook_failed = True
                    break
            if hook_failed:
                # Skip compose_up: a consumer whose provider's contract didn't
                # get re-established (e.g. missing MQTT user) starting up
                # blindly is worse than not starting it. The provider stays up
                # and other apps in the sweep keep going.
                continue
            actions.append(Action("compose_up", m.name))
            if not dry_run:
                dockerops.compose_up(result.path, m.name)
        except (dockerops.DockerError, FileNotFoundError, OSError) as e:
            # Catch broad enough to cover: docker daemon down, docker binary
            # missing on the box, compose file unreadable. Narrow enough that
            # programmer errors (KeyError etc.) still surface.
            actions.append(Action("error", m.name, str(e)))
    return actions


def _fire_on_start_hook(consumer, req, apps_root: Path) -> None:
    """Run a single `on_start` hook against the provider's running container.

    Reconciler-local helper — kept narrow on purpose so reconcile's main loop
    stays scannable. Errors propagate; the caller decorates with the per-app
    Action("error", ...) and skips compose_up for this consumer.
    """
    provider_dir = apps_root / req.app
    provider_manifest_path = provider_dir / "manifest.json"
    if not provider_manifest_path.is_file():
        raise FileNotFoundError(
            f"required app {req.app!r} is not installed"
        )
    # Validate provider manifest loads (otherwise scanner would have skipped
    # it and we'd still try to exec — fail loud here instead).
    load_manifest(provider_manifest_path, expected_name=req.app)
    hook_abs = provider_dir / req.on_start
    if not hook_abs.is_file():
        raise FileNotFoundError(
            f"on_start hook {req.on_start!r} missing in provider {req.app}"
        )
    service = deps.provider_exec_service(provider_dir, req.app)
    dockerops.compose_exec_script(
        provider_dir,
        req.app,
        service,
        hook_abs,
        env={
            "FURTKA_CONSUMER_APP": consumer.name,
            "FURTKA_CONSUMER_VERSION": consumer.version,
        },
        timeout=_ON_START_TIMEOUT_SECONDS,
    )


def has_errors(actions: list[Action]) -> bool:
    return any(a.kind == "error" for a in actions)