computer-playbook/cli/deploy/development/deploy.py

from __future__ import annotations

import argparse
import json
import os
import subprocess
from typing import Any, Mapping

from cli.create.inventory.services_disabler import (
    find_provider_roles,
    parse_services_disabled,
)

from .common import (
    make_compose,
    repo_root_from_here,
    resolve_container,
    resolve_deploy_ids_for_apps,
)
from .inventory import filter_plan_to_variant, plan_dev_inventory_matrix


def _env_variant() -> int | None:
    raw = os.environ.get("VARIANT", "").strip()
    if not raw:
        return None
    try:
        return int(raw)
    except ValueError:
        raise SystemExit(
            f"VARIANT environment variable must be an integer, got {raw!r}"
        )


def _env_full_cycle() -> bool:
    return os.environ.get("FULL_CYCLE", "").strip().lower() == "true"


def _run_deploy(
    compose,
    *,
    deploy_ids: list[str],
    debug: bool,
    passthrough: list[str],
    inventory_dir: str,
    extra_ansible_vars: Mapping[str, Any] | None = None,
) -> int:
    inv_root = str(inventory_dir).rstrip("/")
    inv_file = f"{inv_root}/devices.yml"
    pw_file = f"{inv_root}/.password"

    cmd = [
        "python3",
        "-m",
        "cli.deploy.dedicated",
        inv_file,
        "-p",
        pw_file,
        "-vv",
        "--assert",
        "true",
        "--diff",
        "--id",
        *deploy_ids,
    ]
    if debug:
        cmd.insert(cmd.index("--diff") + 1, "--debug")

    # Ansible extra-vars (`-e key=value`) for caller-supplied runtime toggles
    # (typically the async-pass override). JSON-encoding the value preserves
    # bool/int/list semantics across the shell hop into Ansible.
    if extra_ansible_vars:
        for key, value in extra_ansible_vars.items():
            cmd.extend(["-e", f"{key}={json.dumps(value)}"])

    if passthrough:
        cmd.extend(passthrough)

    extra_env: dict[str, str] = {
        # Force ANSI colors even when no TTY is allocated (CI default).
        "ANSIBLE_FORCE_COLOR": "1",
        "PY_COLORS": "1",
        "TERM": "xterm-256color",
    }
    services_disabled = os.environ.get("SERVICES_DISABLED", "")
    if services_disabled:
        extra_env["SERVICES_DISABLED"] = services_disabled

    # The Playwright E2E gate now keys on `RUNTIME` from the inventory's
    # host_vars (baked at init time by `cli.deploy.development.init`).
    # We no longer need to forward GITHUB_ACTIONS / ACT /
    # INFINITO_MAKE_DEPLOY / INFINITO_SKIP_E2E into the container at deploy
    # time: by then the runtime decision is already serialised into the
    # inventory the deploy stage consumes.

    # Live stream output for immediate visibility.
    r = compose.exec(
        cmd,
        check=False,
        live=True,
        extra_env=extra_env,
    )

    return int(r.returncode)


def _purge_app_entities(*, container: str, app_ids: list[str]) -> None:
    """Run the per-app cleanup script for every app whose variant just
    changed between matrix-deploy rounds.

    `scripts/tests/deploy/local/purge/entity.sh` removes the
    application's containers, networks, and Ansible-managed state on
    the host so the next round boots from a clean slate. Failures are
    surfaced (the matrix MUST NOT silently mix variant state across
    rounds).
    """
    if not app_ids:
        return
    repo_root = repo_root_from_here()
    purge_script = (
        repo_root / "scripts" / "tests" / "deploy" / "local" / "purge" / "entity.sh"
    )
    env = os.environ.copy()
    env["APPS"] = ",".join(app_ids)
    env["INFINITO_CONTAINER"] = container
    print(
        "=== matrix-deploy: purging entities between rounds for "
        f"{', '.join(app_ids)} ==="
    )
    subprocess.run(
        ["bash", str(purge_script)],
        cwd=str(repo_root),
        env=env,
        check=True,
    )


def add_parser(sub: argparse._SubParsersAction) -> None:
    p = sub.add_parser(
        "deploy", help="Run deploy inside the infinito container (requires inventory)."
    )
    p.add_argument(
        "--inventory-dir",
        default=os.environ.get("INVENTORY_DIR"),
        required=os.environ.get("INVENTORY_DIR") is None,
        help=(
            "Inventory directory base (default: $INVENTORY_DIR). When the "
            "primary apps declare more than one matrix-deploy variant, the "
            "wrapper iterates the sibling folders `<dir>-0`, `<dir>-1`, ... "
            "produced by the matching `init` step; otherwise the directory "
            "is used as-is."
        ),
    )

    g = p.add_mutually_exclusive_group(required=True)
    g.add_argument(
        "--apps",
        help="One or more application ids (will include run_after deps automatically).",
    )
    g.add_argument(
        "--id",
        nargs="+",
        default=None,
        help="Explicit application ids (space-separated).",
    )
    p.add_argument(
        "--debug",
        action=argparse.BooleanOptionalAction,
        default=False,
        help="Enable/disable Ansible debug mode (default: disabled).",
    )
    p.add_argument(
        "--variant",
        type=int,
        default=_env_variant(),
        help=(
            "Pin the matrix deploy to a single round (zero-based index), "
            "skipping inter-round cleanup. Useful for redeploying one "
            "specific variant without iterating the whole matrix. Defaults "
            "to the VARIANT environment variable when set, otherwise full-"
            "matrix mode."
        ),
    )
    p.add_argument(
        "--full-cycle",
        action=argparse.BooleanOptionalAction,
        default=_env_full_cycle(),
        help=(
            "After each round's regular deploy, immediately re-run the "
            "deploy with `-e ASYNC_ENABLED=true` (the async update pass). "
            "Pass 1 + Pass 2 stay co-located on the SAME variant so the "
            "async re-deploy always runs against the host state the "
            "matching sync deploy just produced. Defaults to the "
            "FULL_CYCLE environment variable (true|false) when set."
        ),
    )
    p.add_argument(
        "ansible_args",
        nargs=argparse.REMAINDER,
        help="Passthrough args appended to `cli.deploy.dedicated` (use `--` to separate).",
    )
    p.set_defaults(_handler=handler)


def handler(args: argparse.Namespace) -> int:
    compose = make_compose()

    if args.apps:
        deploy_ids = resolve_deploy_ids_for_apps(compose, args.apps)
        primary_app_ids = [
            a.strip() for a in args.apps.replace(",", " ").split() if a.strip()
        ]
    else:
        deploy_ids = list(args.id or [])
        primary_app_ids = list(deploy_ids)

    # Remove any app IDs that were disabled via SERVICES_DISABLED so the deploy
    # list stays consistent with the inventory created by the init step.
    raw_disabled = os.environ.get("SERVICES_DISABLED", "").strip()
    if raw_disabled:
        services = parse_services_disabled(raw_disabled)
        roles_dir = compose.repo_root / "roles"
        provider_map = find_provider_roles(services, roles_dir)
        disabled_app_ids = set(provider_map.values())
        deploy_ids = [app_id for app_id in deploy_ids if app_id not in disabled_app_ids]
        primary_app_ids = [
            app_id for app_id in primary_app_ids if app_id not in disabled_app_ids
        ]

    # argparse.REMAINDER includes the leading '--' if present; drop it
    passthrough = list(args.ansible_args or [])
    if passthrough and passthrough[0] == "--":
        passthrough = passthrough[1:]

    # Matrix-deploy: the init step produced one inventory folder per round
    # (`<dir>-0`, `<dir>-1`, ...; or just `<dir>` when there is a single
    # round) with the per-app variant data already baked into each folder's
    # `host_vars`. Here we just iterate the same plan and deploy against
    # each folder. Between rounds we purge exactly the apps whose target
    # variant changed; apps that stayed on variant 0 are NOT purged. The
    # final round is followed by no purge so the last state remains
    # available for inspection / follow-up specs.
    plan = plan_dev_inventory_matrix(
        roles_dir=str(compose.repo_root / "roles"),
        include=primary_app_ids,
        base_inventory_dir=str(args.inventory_dir),
    )
    try:
        plan = filter_plan_to_variant(plan, args.variant)
    except ValueError as exc:
        raise SystemExit(f"--variant: {exc}")

    # INFINITO_CONTAINER is the single SPOT — defaults.sh keeps it in
    # lock-step with INFINITO_DISTRO across matrix iterations. Read it
    # strictly here; no fallback derivation, no env-vs-arg ambiguity.
    container_name = resolve_container()

    rc = 0
    previous_round_variants: dict[str, int] = {}
    for round_index, inv_dir, round_variants in plan:
        # Round 0 always deploys the full include set (the baseline state
        # of every primary app + its dependencies). Rounds R>0 only
        # re-deploy apps that ACTUALLY have a variant N for that round
        # (i.e. `round_variants[app] == round_index`); apps that fall back
        # to variant 0 are left at whatever state the previous round
        # produced. This avoids wasted re-deploys of apps that have no
        # real variant for the current round.
        if round_index == 0:
            round_deploy_ids = deploy_ids
        else:
            round_deploy_ids = sorted(
                app_id for app_id, idx in round_variants.items() if idx == round_index
            )
            if not round_deploy_ids:
                print(
                    f"=== matrix-deploy: round {round_index + 1}/{len(plan)} "
                    f"skipped (no apps with a real variant {round_index}) ==="
                )
                previous_round_variants = round_variants
                continue

        # Cleanup applies only to apps we are about to re-deploy this
        # round AND whose variant changed since the previous round.
        # `previous_round_variants` only becomes truthy after the first
        # iteration completes, so single-folder modes (N=1 or `--variant`
        # pinned) skip the purge entirely. Apps that are not being
        # re-deployed retain their state from earlier rounds, so purging
        # them would leave the host in an inconsistent half-applied state.
        if previous_round_variants:
            changed = sorted(
                app_id
                for app_id in round_deploy_ids
                if previous_round_variants.get(app_id, 0)
                != round_variants.get(app_id, 0)
            )
            if changed:
                _purge_app_entities(container=container_name, app_ids=changed)

        pass_label = (
            f"matrix-deploy: round {round_index + 1}/{len(plan)} "
            f"inv={inv_dir} variants={round_variants} apps={round_deploy_ids}"
        )
        print(f"=== {pass_label} PASS 1 (sync) ===")
        rc = _run_deploy(
            compose,
            deploy_ids=round_deploy_ids,
            debug=bool(args.debug),
            passthrough=passthrough,
            inventory_dir=inv_dir,
        )
        if rc != 0:
            return rc

        # When `--full-cycle` is set, the async update pass runs IMMEDIATELY
        # against the same variant's host state, before we move to the next
        # round. This co-locates PASS 1 and PASS 2 per variant so the async
        # re-deploy never accidentally targets a host that was left in a
        # different variant by a prior round.
        if bool(args.full_cycle):
            print(f"=== {pass_label} PASS 2 (async) ===")
            rc = _run_deploy(
                compose,
                deploy_ids=round_deploy_ids,
                debug=bool(args.debug),
                passthrough=passthrough,
                inventory_dir=inv_dir,
                extra_ansible_vars={"ASYNC_ENABLED": True},
            )
            if rc != 0:
                return rc

        previous_round_variants = round_variants

    return rc