pic/tests/integration/test_apply_propagation.py

"""
Apply-propagation integration tests.

Verifies the full save → pending → apply → verify lifecycle:

  1. GET  /api/config/pending  — reports needs_restart
  2. PUT  /api/config          — a port change marks pending
  3. DELETE /api/config/pending — discard clears the flag
  4. POST /api/config/apply    — apply returns 200, clears pending,
                                 restarts only the affected container
  5. After restart, GET /api/config reflects the saved change

Routes used (confirmed in api/app.py):
  GET    /api/config/pending   — {needs_restart, changed_at, changes, containers}
  DELETE /api/config/pending   — discard without restart
  POST   /api/config/apply     — trigger restart; returns {message, restart_in_progress}
  GET    /health               — {status: "healthy"} (used to poll for recovery)

Why calendar.port?
------------------
Changing calendar.port is the safest apply-test trigger because:
  - It falls under the _PORT_CHANGE_MAP in app.py and therefore sets
    needs_restart=true pointing only at the ['radicale'] container.
  - The API container itself is NOT in that list, so the Flask process
    stays up during apply — no connection gap to handle.
  - cell_name / domain changes are applied immediately to DNS and do NOT
    set needs_restart, so they cannot be used to exercise the pending path.

Run with:  pytest tests/integration/test_apply_propagation.py -v
"""
import time
import sys
import os

import pytest
import requests
from requests.exceptions import ConnectionError, Timeout

sys.path.insert(0, os.path.dirname(__file__))
from conftest import API_BASE, _resolve_admin_pass

# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------

_POLL_INTERVAL = 2    # seconds between health polls
_HEALTH_TIMEOUT = 90  # max seconds to wait for healthy after apply

# Two distinct valid calendar ports used as before/after values.
# Neither conflicts with any other default service port.
_CAL_PORT_A = 5232  # the standard Radicale default
_CAL_PORT_B = 5233  # an alternate safe value used as the "changed" state


# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------


_S = None

@pytest.fixture(scope='module', autouse=True)
def _auth_session():
    global _S
    _S = requests.Session()
    _S.headers['Content-Type'] = 'application/json'
    r = _S.post(f"{API_BASE}/api/auth/login",
                json={'username': 'admin', 'password': _resolve_admin_pass()})
    assert r.status_code == 200, f"Login failed: {{r.text}}"

def get(path, **kw):
    return _S.get(f"{API_BASE}{path}", **kw)


def put(path, **kw):
    return _S.put(f"{API_BASE}{path}", **kw)


def post(path, **kw):
    return _S.post(f"{API_BASE}{path}", **kw)


def delete(path, **kw):
    return _S.delete(f"{API_BASE}{path}", **kw)


def wait_for_healthy(timeout: int = _HEALTH_TIMEOUT) -> bool:
    """
    Poll GET /health until it returns {"status": "healthy"} or timeout expires.
    Connection errors are swallowed so the loop survives a container restart.
    Returns True if healthy within timeout, False otherwise.
    """
    deadline = time.monotonic() + timeout
    while time.monotonic() < deadline:
        try:
            r = requests.get(f"{API_BASE}/health", timeout=5)
            if r.status_code == 200 and r.json().get("status") == "healthy":
                return True
        except (ConnectionError, Timeout):
            pass  # API may be momentarily unreachable — keep trying
        time.sleep(_POLL_INTERVAL)
    return False


def pending_state() -> dict:
    """Return the current /api/config/pending response body."""
    return get("/api/config/pending").json()


def current_calendar_port() -> int:
    """Read calendar.port from the live config."""
    cfg = get("/api/config").json()
    svc = cfg.get("service_configs", {}).get("calendar", {})
    return int(svc.get("port", _CAL_PORT_A))


def set_calendar_port(port: int) -> requests.Response:
    """PUT calendar.port and return the response."""
    return put("/api/config", json={"calendar": {"port": port}})


# ---------------------------------------------------------------------------
# TestPendingState  (no container restarts)
# ---------------------------------------------------------------------------

class TestPendingState:
    """Tests that verify pending-state semantics without triggering an apply."""

    def test_pending_starts_false(self):
        """
        After discarding any stale pending changes, GET /api/config/pending
        must return needs_restart=false.
        """
        delete("/api/config/pending")
        data = pending_state()
        assert data["needs_restart"] is False, (
            f"Expected needs_restart=false at baseline, got: {data}"
        )

    def test_save_config_sets_pending(self):
        """
        PUT /api/config with a changed calendar.port must flip needs_restart
        to true.  The change is discarded (not applied) so no restart occurs.
        """
        original_port = current_calendar_port()
        # Pick whichever alternate port is not currently in use.
        new_port = _CAL_PORT_B if original_port == _CAL_PORT_A else _CAL_PORT_A

        # Start from a clean state.
        delete("/api/config/pending")
        assert pending_state()["needs_restart"] is False, "Could not clear pending state"

        try:
            r = set_calendar_port(new_port)
            assert r.status_code == 200, f"PUT /api/config failed: {r.text}"

            data = pending_state()
            assert data["needs_restart"] is True, (
                f"Expected needs_restart=true after port change, got: {data}"
            )
            assert isinstance(data["changes"], list)
            assert len(data["changes"]) > 0, (
                "changes list is empty even though needs_restart is true"
            )
            # The pending restart should be scoped to the radicale container.
            assert "radicale" in data.get("containers", []) or data.get("containers") == ["*"], (
                f"Expected 'radicale' in pending containers, got: {data.get('containers')}"
            )
        finally:
            set_calendar_port(original_port)
            delete("/api/config/pending")

    def test_discard_clears_pending(self):
        """
        DELETE /api/config/pending after a config change must reset
        needs_restart to false and empty the changes list without
        restarting any containers.
        """
        original_port = current_calendar_port()
        new_port = _CAL_PORT_B if original_port == _CAL_PORT_A else _CAL_PORT_A

        delete("/api/config/pending")  # start clean

        try:
            r = set_calendar_port(new_port)
            assert r.status_code == 200, f"PUT /api/config failed: {r.text}"
            assert pending_state()["needs_restart"] is True, (
                "needs_restart not set after port change — cannot test discard"
            )

            dr = delete("/api/config/pending")
            assert dr.status_code == 200, (
                f"DELETE /api/config/pending returned {dr.status_code}: {dr.text}"
            )
            body = dr.json()
            assert "message" in body, f"Discard response missing 'message': {body}"

            data = pending_state()
            assert data["needs_restart"] is False, (
                f"needs_restart still true after discard: {data}"
            )
            assert data["changes"] == [], (
                f"changes list not empty after discard: {data}"
            )
        finally:
            set_calendar_port(original_port)
            delete("/api/config/pending")


# ---------------------------------------------------------------------------
# TestApplyAndVerify  (triggers actual container restart)
# ---------------------------------------------------------------------------

class TestApplyAndVerify:
    """
    Tests that call POST /api/config/apply.

    Because a calendar.port change only restarts the 'radicale' container
    (not the API container), the API stays up throughout.  wait_for_healthy()
    is still called after each apply to confirm full readiness before making
    assertions.

    Every test restores the original calendar port in a finally block.
    """

    def test_apply_endpoint_exists(self):
        """
        POST /api/config/apply must return 200 with a 'message' key even
        when there is nothing pending (documented no-op behaviour).
        """
        delete("/api/config/pending")

        r = post("/api/config/apply")
        assert r.status_code == 200, (
            f"Expected 200 from POST /api/config/apply (no-op), "
            f"got {r.status_code}: {r.text}"
        )
        body = r.json()
        assert "message" in body, f"Response body missing 'message' key: {body}"

    def test_apply_clears_pending(self):
        """
        After saving a config change and calling POST /api/config/apply,
        GET /api/config/pending must return needs_restart=false.

        app.py clears the pending flag synchronously before spawning the
        restart thread, so the flag is cleared as soon as the apply HTTP
        response is received — regardless of when containers finish starting.
        """
        original_port = current_calendar_port()
        new_port = _CAL_PORT_B if original_port == _CAL_PORT_A else _CAL_PORT_A

        delete("/api/config/pending")

        try:
            r = set_calendar_port(new_port)
            assert r.status_code == 200, f"PUT /api/config failed: {r.text}"
            assert pending_state()["needs_restart"] is True, (
                "needs_restart not set — cannot verify that apply clears it"
            )

            ar = post("/api/config/apply")
            assert ar.status_code == 200, (
                f"POST /api/config/apply returned {ar.status_code}: {ar.text}"
            )
            apply_body = ar.json()
            assert "message" in apply_body, (
                f"Apply response missing 'message': {apply_body}"
            )

            # Wait for the API to confirm healthy (radicale restart in background).
            recovered = wait_for_healthy()
            assert recovered, (
                f"API did not return healthy within {_HEALTH_TIMEOUT}s after apply"
            )

            data = pending_state()
            assert data["needs_restart"] is False, (
                f"needs_restart still true after apply + recovery: {data}"
            )

        finally:
            set_calendar_port(original_port)
            delete("/api/config/pending")
            wait_for_healthy(_HEALTH_TIMEOUT)

    def test_cell_name_change_persists_after_apply(self):
        """
        Full lifecycle: change calendar.port → apply → wait for healthy →
        GET /api/config must return the new port value.

        This confirms that the configuration written to disk before apply
        survives the container restart cycle and is not rolled back.
        cell_name is used as a secondary check: because cell_name changes
        are applied immediately to DNS (not via pending), we verify it was
        not inadvertently cleared by the apply path.
        """
        original_port = current_calendar_port()
        original_cfg = get("/api/config").json()
        original_name = original_cfg["cell_name"]

        new_port = _CAL_PORT_B if original_port == _CAL_PORT_A else _CAL_PORT_A

        delete("/api/config/pending")

        try:
            # 1. Save a new calendar port.
            r = set_calendar_port(new_port)
            assert r.status_code == 200, f"PUT /api/config failed: {r.text}"

            # 2. Confirm config is saved before apply.
            saved_port = current_calendar_port()
            assert saved_port == new_port, (
                f"calendar.port not saved before apply: got {saved_port}"
            )

            # 3. Confirm pending is set.
            assert pending_state()["needs_restart"] is True, (
                "needs_restart not set after calendar.port change"
            )

            # 4. Apply.
            ar = post("/api/config/apply")
            assert ar.status_code == 200, (
                f"POST /api/config/apply returned {ar.status_code}: {ar.text}"
            )
            apply_body = ar.json()
            assert "message" in apply_body, (
                f"Apply response missing 'message': {apply_body}"
            )

            # 5. Wait for healthy.
            recovered = wait_for_healthy()
            assert recovered, (
                f"API did not return healthy within {_HEALTH_TIMEOUT}s after apply"
            )

            # 6. Verify the port change persists in the running config.
            post_apply_port = current_calendar_port()
            assert post_apply_port == new_port, (
                f"calendar.port reverted after apply: "
                f"expected {new_port}, got {post_apply_port}"
            )

            # 7. Confirm that cell_name was not inadvertently cleared.
            post_apply_cfg = get("/api/config").json()
            assert post_apply_cfg["cell_name"] == original_name, (
                f"cell_name changed unexpectedly during apply: "
                f"expected {original_name!r}, got {post_apply_cfg['cell_name']!r}"
            )

            # 8. Confirm pending is cleared.
            assert pending_state()["needs_restart"] is False, (
                "needs_restart still true after apply + recovery"
            )

        finally:
            # Restore the original calendar port.
            set_calendar_port(original_port)
            post("/api/config/apply")
            wait_for_healthy(_HEALTH_TIMEOUT)
            delete("/api/config/pending")