homeassistant-stelloauth-addon/tests/test_runtime.sh

#!/usr/bin/env bash
set -Eeuo pipefail

repo_root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
image="homeassistant-stelloauth-addon:test"
run_id="$(date +%s)-$$"
first_container="stelloauth-runtime-${run_id}-first"
second_container="stelloauth-runtime-${run_id}-second"
tmp_dir="$(mktemp -d)"
artifacts_dir="${repo_root}/artifacts"
options_file="${tmp_dir}/options.json"

sentinels=(
    "sentinel-email@example.invalid"
    "SENTINEL_PASSWORD_9a34"
    "SENTINEL_COOKIE_7b21"
    "SENTINEL_OAUTH_CODE_5c88"
    "SENTINEL_ACCESS_TOKEN_1d62"
    "SENTINEL_REFRESH_TOKEN_4e73"
)

cleanup() {
    set +e
    for container in "$first_container" "$second_container"; do
        if docker container inspect "$container" >/dev/null 2>&1; then
            if [ "$(docker inspect --format '{{.State.Running}}' "$container" 2>/dev/null)" = "true" ]; then
                docker stop --time 10 "$container" >/dev/null 2>&1
            fi
            docker rm "$container" >/dev/null 2>&1
        fi
    done
    rm -r "$tmp_dir"
}
trap cleanup EXIT

fail() {
    printf 'runtime test failed: %s\n' "$*" >&2
    exit 1
}

write_options() {
    python3 - "$options_file" <<'PY'
import json
import pathlib
import sys

options = {
    "queue_timeout": "60s",
    "rate_limit_count": 5,
    "rate_limit_duration": "1h",
}
pathlib.Path(sys.argv[1]).write_text(
    json.dumps(options, separators=(",", ":")) + "\n",
    encoding="utf-8",
)
PY
}

start_container() {
    local container="$1"
    docker run --detach \
        --name "$container" \
        --platform linux/amd64 \
        --mount "type=bind,src=${options_file},dst=/data/options.json,readonly" \
        --publish 127.0.0.1::8080 \
        "$image" >/dev/null
}

host_port() {
    local container="$1"
    local mapping
    mapping="$(docker port "$container" 8080/tcp)"
    [ -n "$mapping" ] || fail "container 8080 has no host mapping"
    printf '%s\n' "${mapping##*:}"
}

probe_root() {
    local port="$1"
    python3 - "$port" <<'PY'
import sys
import urllib.request

opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
with opener.open(f"http://127.0.0.1:{sys.argv[1]}/", timeout=2) as response:
    if response.status != 200:
        raise SystemExit(f"root status {response.status}")
    response.read()
PY
}

wait_ready() {
    local container="$1"
    local port="$2"
    local deadline=$((SECONDS + 90))
    while (( SECONDS < deadline )); do
        if [ "$(docker inspect --format '{{.State.Running}}' "$container")" != "true" ]; then
            docker logs "$container" >&2
            fail "$container exited during readiness"
        fi
        if docker logs "$container" 2>&1 | grep -Fq "Stelloauth listening on 0.0.0.0:8080"; then
            if probe_root "$port" >/dev/null 2>&1; then
                return
            fi
        fi
        sleep 1
    done
    docker logs "$container" >&2
    fail "$container did not become ready within 90 seconds"
}

normalize_process_file() {
    local top_file="$1"
    python3 - "$top_file" <<'PY'
import pathlib
import sys

commands = []
for raw_line in pathlib.Path(sys.argv[1]).read_text(encoding="utf-8").splitlines()[1:]:
    fields = raw_line.split(maxsplit=1)
    command = " ".join(fields[1].split()) if len(fields) == 2 else ""
    tokens = command.split()
    while tokens and (
        tokens[0] == "/run/rosetta/rosetta"
        or pathlib.Path(tokens[0]).name.startswith("qemu-")
    ):
        tokens = tokens[1:]

    python_prefixes = (
        [],
        ["python3"],
        ["/usr/local/bin/python3"],
        ["/usr/local/bin/python3", "python3"],
    )
    supervisor_commands = [
        [*prefix, "/usr/local/bin/addon-supervisor"]
        for prefix in python_prefixes
    ]
    cloak_commands = [
        [
            *prefix,
            "/usr/local/bin/cloakserve",
            "--headless=true",
            "--idle-timeout=30",
            "--data-dir=/tmp/cloakserve",
        ]
        for prefix in python_prefixes
    ]
    stelloauth_commands = (
        ["/usr/local/bin/stelloauth"],
        ["/usr/local/bin/stelloauth", "/usr/local/bin/stelloauth"],
    )

    if tokens in supervisor_commands:
        commands.append("addon-supervisor")
    elif tokens in cloak_commands:
        commands.append("cloakserve")
    elif tokens in stelloauth_commands:
        commands.append("stelloauth")
    elif command:
        commands.append(f"unexpected:{command}")

print("\n".join(sorted(commands)))
PY
}

normalized_process_commands() {
    local container="$1"
    local top_file="${tmp_dir}/${container}-processes.txt"
    docker top "$container" -eo pid,args > "$top_file"
    normalize_process_file "$top_file"
}

capture_process_baseline() {
    local container="$1"
    local expected current
    local deadline=$((SECONDS + 10))
    expected=$'addon-supervisor\ncloakserve\nstelloauth'
    while (( SECONDS < deadline )); do
        current="$(normalized_process_commands "$container")"
        if [ "$current" = "$expected" ]; then
            printf '%s\n' "$current"
            return
        fi
        sleep 0.25
    done
    printf 'expected startup process baseline:\n%s\ncurrent process commands:\n%s\n' \
        "$expected" "$current" >&2
    docker top "$container" >&2
    fail "$container did not reach the expected startup process baseline"
}

assert_processes_return_to_baseline() {
    local container="$1"
    local baseline="$2"
    local current
    local deadline=$((SECONDS + 10))
    while (( SECONDS < deadline )); do
        current="$(normalized_process_commands "$container")"
        if [ "$current" = "$baseline" ]; then
            return
        fi
        sleep 0.25
    done
    printf 'expected process baseline after CDP close:\n%s\ncurrent process commands:\n%s\n' \
        "$baseline" "$current" >&2
    docker top "$container" >&2
    fail "$container retained browser or profile processes after CDP close"
}

assert_loopback_cdp_listener() {
    local container="$1"
    local artifact="$2"
    docker exec "$container" cat /proc/net/tcp > "$artifact"
    docker exec -i "$container" python3 - <<'PY'
expected = f"0100007F:{9222:04X}"
wildcard = f"00000000:{9222:04X}"
if expected != "0100007F:2406":
    raise SystemExit(f"unexpected 9222 hexadecimal encoding: {expected}")

listeners = set()
with open("/proc/net/tcp", encoding="ascii") as handle:
    next(handle)
    for line in handle:
        fields = line.split()
        if len(fields) >= 4 and fields[3] == "0A":
            listeners.add(fields[1].upper())

if expected not in listeners:
    raise SystemExit(f"missing loopback CDP listener {expected}: {sorted(listeners)}")
if wildcard in listeners:
    raise SystemExit(f"wildcard CDP listener present: {wildcard}")
PY
}

probe_and_close_cdp() {
    local container="$1"
    docker exec -i "$container" python3 - <<'PY'
import json
import urllib.request

opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
version_url = "http://127.0.0.1:9222/json/version?fingerprint=runtime-readiness"
close_url = "http://127.0.0.1:9222/fingerprint/runtime-readiness/close"
with opener.open(version_url, timeout=10) as response:
    if response.status != 200:
        raise SystemExit(f"CDP version status {response.status}")
    document = json.load(response)
websocket_url = document.get("webSocketDebuggerUrl")
if not isinstance(websocket_url, str) or not websocket_url:
    raise SystemExit("CDP response lacks webSocketDebuggerUrl")
request = urllib.request.Request(close_url, data=b"", method="POST")
with opener.open(request, timeout=10) as response:
    if response.status != 200:
        raise SystemExit(f"CDP close status {response.status}")
    response.read()
PY
}

post_invalid_worker() {
    local port="$1"
    local response_artifact="$2"
    python3 - "$port" "$response_artifact" <<'PY'
import json
import pathlib
import sys
import urllib.error
import urllib.request

port, artifact = sys.argv[1:]
body = {
    "url": (
        "https://example.invalid/am/oauth2/authorize"
        "?redirect_uri=sentinel%3A%2F%2Fcallback"
        "&code=SENTINEL_OAUTH_CODE_5c88"
        "&access_token=SENTINEL_ACCESS_TOKEN_1d62"
        "&refresh_token=SENTINEL_REFRESH_TOKEN_4e73"
        "&cookie=SENTINEL_COOKIE_7b21"
    ),
    "email": "sentinel-email@example.invalid",
    "password": "SENTINEL_PASSWORD_9a34",
    "cookie": "SENTINEL_COOKIE_7b21",
    "oauth_code": "SENTINEL_OAUTH_CODE_5c88",
    "access_token": "SENTINEL_ACCESS_TOKEN_1d62",
    "refresh_token": "SENTINEL_REFRESH_TOKEN_4e73",
}
request = urllib.request.Request(
    f"http://127.0.0.1:{port}/worker",
    data=json.dumps(body, separators=(",", ":")).encode(),
    headers={"Content-Type": "application/json"},
    method="POST",
)
opener = urllib.request.build_opener(urllib.request.ProxyHandler({}))
try:
    with opener.open(request, timeout=10) as response:
        status = response.status
        response_body = response.read()
except urllib.error.HTTPError as error:
    status = error.code
    response_body = error.read()
if status != 400:
    raise SystemExit(f"invalid worker status {status}, want 400")
pathlib.Path(artifact).write_bytes(response_body)
PY
}

assert_no_9222_mapping() {
    local container="$1"
    local mapping
    mapping="$(docker port "$container" 9222/tcp 2>/dev/null || true)"
    [ -z "$mapping" ] || fail "container 9222 is mapped: $mapping"
}

scan_logs() {
    local container="$1"
    local log_file="${tmp_dir}/${container}.log"
    docker logs "$container" > "$log_file" 2>&1
    for sentinel in "${sentinels[@]}"; do
        if grep -Fq "$sentinel" "$log_file"; then
            fail "$container logs contain sentinel $sentinel"
        fi
    done
    if grep -Fq "worker OAuth request" "$log_file"; then
        fail "$container began an OAuth flow for the rejected worker body"
    fi
}

stop_and_assert() {
    local container="$1"
    local timing_artifact="$2"
    local started_ns ended_ns elapsed state
    started_ns="$(python3 -c 'import time; print(time.monotonic_ns())')"
    docker stop --time 10 "$container" >/dev/null
    ended_ns="$(python3 -c 'import time; print(time.monotonic_ns())')"
    elapsed="$(python3 - "$started_ns" "$ended_ns" <<'PY'
import sys
print((int(sys.argv[2]) - int(sys.argv[1])) / 1_000_000_000)
PY
)"
    printf 'seconds=%s\n' "$elapsed" > "$timing_artifact"
    python3 - "$elapsed" <<'PY'
import sys
if float(sys.argv[1]) > 10.0:
    raise SystemExit(f"container stop exceeded 10 seconds: {sys.argv[1]}")
PY
    state="$(docker inspect --format '{{.State.Status}} {{.State.ExitCode}} {{.State.Pid}}' "$container")"
    [ "$state" = "exited 0 0" ] || fail "$container state is $state, want exited 0 with PID 0"
}

if [ "${1:-}" = "--normalize-processes" ]; then
    [ "$#" -eq 2 ] || fail "--normalize-processes requires one docker top file"
    normalize_process_file "$2"
    exit
fi
[ "$#" -eq 0 ] || fail "unexpected runtime test arguments"

mkdir -p "$artifacts_dir"
write_options

if [ "${SKIP_BUILD:-0}" != "1" ]; then
    docker buildx build \
        --platform linux/amd64 \
        --build-arg BUILD_ARCH=amd64 \
        --load \
        --tag "$image" \
        "$repo_root/stelloauth"
fi

start_container "$first_container"
first_port="$(host_port "$first_container")"
wait_ready "$first_container" "$first_port"
first_baseline="$(capture_process_baseline "$first_container")"
assert_no_9222_mapping "$first_container"
assert_loopback_cdp_listener "$first_container" "${artifacts_dir}/runtime-proc-net-tcp.txt"
probe_and_close_cdp "$first_container"
assert_processes_return_to_baseline "$first_container" "$first_baseline"
post_invalid_worker "$first_port" "${artifacts_dir}/runtime-invalid-worker-response.json"
scan_logs "$first_container"
stop_and_assert "$first_container" "${artifacts_dir}/runtime-first-stop.txt"
scan_logs "$first_container"

start_container "$second_container"
second_port="$(host_port "$second_container")"
wait_ready "$second_container" "$second_port"
second_baseline="$(capture_process_baseline "$second_container")"
assert_no_9222_mapping "$second_container"
assert_loopback_cdp_listener "$second_container" "${artifacts_dir}/runtime-restart-proc-net-tcp.txt"
probe_and_close_cdp "$second_container"
assert_processes_return_to_baseline "$second_container" "$second_baseline"

sleep 5
docker stats --no-stream "$second_container" > "${artifacts_dir}/runtime-docker-stats.txt"
docker top "$second_container" > "${artifacts_dir}/runtime-docker-top.txt"
docker image inspect "$image" --format '{{.Size}}' > "${artifacts_dir}/runtime-image-size-bytes.txt"
docker image inspect "$image" --format '{{json .Config.ExposedPorts}}' > "${artifacts_dir}/runtime-image-exposed-ports.json"
docker inspect "$second_container" --format '{{json .HostConfig.PortBindings}}' > "${artifacts_dir}/runtime-host-port-bindings.json"

stop_and_assert "$second_container" "${artifacts_dir}/runtime-second-stop.txt"
scan_logs "$second_container"

printf 'runtime acceptance PASS: root, CDP, loopback bind, invalid worker, redaction, stop, restart\n'