CIS490/tests/test_host_health.py

"""Tests for the receiver's /v1/host-health endpoints.

PUT /v1/host-health/<host>  → store the lab host's daily doctor JSON
GET /v1/host-health         → aggregate snapshot across all hosts

The endpoint is deliberately NOT gated by version_gate — sick hosts
running stale code still need to be able to report sickness.
"""

from __future__ import annotations

import json
from pathlib import Path

import pytest
from starlette.testclient import TestClient

from receiver.app import make_app
from receiver.store import EpisodeStore


@pytest.fixture
def app(tmp_path: Path):
    store = EpisodeStore(
        store_root=tmp_path / "store",
        incoming_root=tmp_path / "incoming",
        index_path=tmp_path / "index.jsonl",
    )
    return make_app(
        store=store, max_episode_bytes=10_000_000,
        bearer_token=None,
        health_root=tmp_path / "host-health",
    )


@pytest.fixture
def app_with_bearer(tmp_path: Path):
    store = EpisodeStore(
        store_root=tmp_path / "store",
        incoming_root=tmp_path / "incoming",
        index_path=tmp_path / "index.jsonl",
    )
    return make_app(
        store=store, max_episode_bytes=10_000_000,
        bearer_token="s3cret",
        health_root=tmp_path / "host-health",
    )


@pytest.fixture
def app_no_health(tmp_path: Path):
    """health_root=None ⇒ endpoint returns 404."""
    store = EpisodeStore(
        store_root=tmp_path / "store",
        incoming_root=tmp_path / "incoming",
        index_path=tmp_path / "index.jsonl",
    )
    return make_app(
        store=store, max_episode_bytes=10_000_000,
        bearer_token=None,
        health_root=None,
    )


def test_put_health_stores_doctor_json(app, tmp_path: Path) -> None:
    snapshot = {
        "role": "lab-host",
        "checks": [
            {"name": "install: VERSION stamp", "status": "ok", "detail": "main@abc123"},
            {"name": "shipper: recent ship results", "status": "fail",
             "detail": "12 412s in last 10 min"},
        ],
    }
    with TestClient(app) as c:
        r = c.put("/v1/host-health/lab1", json=snapshot)
    assert r.status_code == 200
    assert r.json()["status"] == "stored"
    assert r.json()["host_id"] == "lab1"

    # File written
    target = tmp_path / "host-health" / "lab1.json"
    assert target.exists()
    body = json.loads(target.read_text())
    assert body["host_id"] == "lab1"
    assert body["doctor"] == snapshot
    assert "received_at_wall" in body


def test_put_health_overwrites_previous_snapshot(app, tmp_path: Path) -> None:
    """The endpoint stores the LATEST snapshot per host, not history."""
    with TestClient(app) as c:
        r1 = c.put("/v1/host-health/lab1", json={"checks": [{"v": 1}]})
        r2 = c.put("/v1/host-health/lab1", json={"checks": [{"v": 2}]})
    assert r1.status_code == 200 and r2.status_code == 200
    body = json.loads((tmp_path / "host-health" / "lab1.json").read_text())
    assert body["doctor"] == {"checks": [{"v": 2}]}


def test_put_health_rejects_invalid_host_id(app) -> None:
    """is_valid_id rejects path-traversal-y strings + extreme lengths.
    URL-routing-level traversal (../) is normalised by the test
    client before reaching us, so probe with a chunk of bad chars
    that survive routing."""
    with TestClient(app) as c:
        r = c.put("/v1/host-health/lab%20with%20space", json={"checks": []})
    assert r.status_code == 400


def test_put_health_rejects_non_json_body(app) -> None:
    with TestClient(app) as c:
        r = c.put("/v1/host-health/lab1", content=b"not json",
                  headers={"Content-Type": "application/json"})
    assert r.status_code == 400


def test_put_health_rejects_array_body(app) -> None:
    """Body must be an OBJECT, not a list — the doctor's --json output
    is always {role, checks: [...]}, never bare list."""
    with TestClient(app) as c:
        r = c.put("/v1/host-health/lab1", json=["x", "y"])
    assert r.status_code == 400


def test_put_health_404_when_disabled(app_no_health) -> None:
    """Receivers without health_root configured return 404 — lets a
    deployment opt out without removing the routes."""
    with TestClient(app_no_health) as c:
        r = c.put("/v1/host-health/lab1", json={"checks": []})
    assert r.status_code == 404


def test_put_health_not_gated_by_version_gate(tmp_path: Path) -> None:
    """A sick host with stale code MUST still be able to report
    sickness. Confirm we don't check X-Cis490-Code-Commit on the
    health endpoint."""
    store = EpisodeStore(
        store_root=tmp_path / "store",
        incoming_root=tmp_path / "incoming",
        index_path=tmp_path / "index.jsonl",
    )
    # Build a gate that rejects everything to prove we don't run it.
    class _RejectAll:
        def check(self, commit): return False, "not-in-window"
        def head(self): return None
        def valid_count(self): return 0
    app = make_app(
        store=store, max_episode_bytes=10_000_000,
        bearer_token=None,
        version_gate=_RejectAll(),
        health_root=tmp_path / "host-health",
    )
    with TestClient(app) as c:
        r = c.put("/v1/host-health/lab1", json={"checks": []})
    assert r.status_code == 200


def test_put_health_requires_bearer_when_configured(app_with_bearer) -> None:
    with TestClient(app_with_bearer) as c:
        r = c.put("/v1/host-health/lab1", json={"checks": []})
        assert r.status_code == 401
        r2 = c.put("/v1/host-health/lab1", json={"checks": []},
                   headers={"Authorization": "Bearer s3cret"})
        assert r2.status_code == 200


def test_get_fleet_health_returns_all_snapshots(app, tmp_path: Path) -> None:
    with TestClient(app) as c:
        c.put("/v1/host-health/lab1", json={"checks": [{"v": 1}]})
        c.put("/v1/host-health/lab2", json={"checks": [{"v": 2}]})
        r = c.get("/v1/host-health")
    assert r.status_code == 200
    body = r.json()
    hosts = {h["host_id"]: h for h in body["hosts"]}
    assert hosts["lab1"]["doctor"] == {"checks": [{"v": 1}]}
    assert hosts["lab2"]["doctor"] == {"checks": [{"v": 2}]}


def test_get_fleet_health_empty_when_no_reports(app) -> None:
    with TestClient(app) as c:
        r = c.get("/v1/host-health")
    assert r.status_code == 200
    assert r.json() == {"hosts": []}


def test_get_fleet_health_ignores_temp_files(app, tmp_path: Path) -> None:
    """Atomic-write tmpfiles (.lab1.json.tmp) shouldn't show up in the
    aggregate listing if a write was in-flight."""
    health_dir = tmp_path / "host-health"
    health_dir.mkdir()
    (health_dir / ".lab1.json.tmp").write_text('{"host_id": "lab1"}')
    (health_dir / "lab2.json").write_text(
        '{"host_id": "lab2", "doctor": {"checks": []}}'
    )
    with TestClient(app) as c:
        r = c.get("/v1/host-health")
    body = r.json()
    hosts = [h.get("host_id") for h in body["hosts"]]
    assert hosts == ["lab2"]