tests/test_resilience.py

"""The app must degrade gracefully, never crash: input validation,
fail-open cache, broad LLM fallback, global exception handler."""

import json

import pytest
import requests as requests_lib
from fastapi.testclient import TestClient

import main
from app import cache
from app.agents import synthesizer
from app.providers import gemini, groq, llm
from app.providers.gemini import ProviderError
from tests.test_synthesizer import make_context

client = TestClient(main.app)
crash_client = TestClient(main.app, raise_server_exceptions=False)


def stream_payloads(response):
    return [
        json.loads(line[6:])
        for line in response.text.splitlines()
        if line.startswith("data: ")
    ]


# ---- plan stream input validation ------------------------------------------

def test_invalid_arrive_by_returns_friendly_sse_error():
    response = client.get(
        "/api/plan/stream",
        params={"from": "Aliganj", "to": "Chowk", "arrive_by": "not-a-date"},
    )
    assert response.status_code == 200  # SSE, not a 500
    payloads = stream_payloads(response)
    assert payloads[0]["type"] == "error"
    assert "invalid" in payloads[0]["message"].lower()
    assert payloads[-1]["type"] == "done"


def test_past_arrive_by_returns_friendly_sse_error():
    response = client.get(
        "/api/plan/stream",
        params={"from": "Aliganj", "to": "Chowk", "arrive_by": "2020-01-01T09:30"},
    )
    payloads = stream_payloads(response)
    assert payloads[0]["type"] == "error"
    assert "past" in payloads[0]["message"].lower()


def test_far_future_arrive_by_returns_friendly_sse_error():
    response = client.get(
        "/api/plan/stream",
        params={"from": "Aliganj", "to": "Chowk", "arrive_by": "2099-01-01T09:30"},
    )
    payloads = stream_payloads(response)
    assert payloads[0]["type"] == "error"
    assert "week" in payloads[0]["message"].lower()


# ---- endpoint crash safety ----------------------------------------------------

def test_autocomplete_crash_returns_empty_not_500(monkeypatch):
    def boom(q):
        raise RuntimeError("unexpected bug key=secret123")

    monkeypatch.setattr(main.geocode, "autocomplete_place", boom)
    response = client.get("/api/autocomplete", params={"q": "hazra"})
    assert response.status_code == 200
    assert response.json() == {"suggestions": []}


def test_ask_crash_returns_clean_500_without_secrets(monkeypatch):
    monkeypatch.setattr(main.llm, "available", lambda: True)

    def boom(question):
        raise RuntimeError("boom at url?key=topsecret99")

    monkeypatch.setattr(main.orchestrator, "agent_ask", boom)
    response = client.post("/api/ask", json={"question": "hello"})
    assert response.status_code == 500
    assert "topsecret99" not in response.text
    assert "detail" in response.json()


def test_ask_empty_question_rejected(monkeypatch):
    monkeypatch.setattr(main.llm, "available", lambda: True)
    response = client.post("/api/ask", json={"question": "   "})
    assert response.status_code == 422


def test_global_handler_catches_unhandled_route_errors(monkeypatch):
    def boom():
        raise ValueError("internal bug key=hidden42")

    monkeypatch.setattr(main.llm, "available", boom)
    response = crash_client.get("/api/health")
    assert response.status_code == 500
    assert "hidden42" not in response.text
    assert "detail" in response.json()


# ---- fail-open cache ----------------------------------------------------------

def test_cache_read_failure_is_a_miss(monkeypatch):
    def broken_conn():
        raise RuntimeError("disk locked")

    monkeypatch.setattr(cache, "_get_conn", broken_conn)
    assert cache.get("anything") is None  # no exception


def test_cache_write_failure_is_silent(monkeypatch):
    def broken_conn():
        raise RuntimeError("disk full")

    monkeypatch.setattr(cache, "_get_conn", broken_conn)
    cache.set("key", "value", ttl_seconds=60)  # must not raise


def test_cached_function_still_works_when_cache_broken(monkeypatch):
    def broken_conn():
        raise RuntimeError("corrupted db")

    monkeypatch.setattr(cache, "_get_conn", broken_conn)

    @cache.cached(ttl_seconds=60)
    def compute(x):
        return x * 2

    assert compute(21) == 42  # function result delivered despite broken cache


# ---- LLM layer degradation ----------------------------------------------------

def test_synthesizer_survives_unexpected_llm_crash(monkeypatch, fake_keys):
    def boom(*args, **kwargs):
        raise ValueError("adapter bug, not a ProviderError")

    monkeypatch.setattr(synthesizer.llm, "chat", boom)
    verdict, provider = synthesizer.synthesize(make_context())
    assert provider == "deterministic"
    assert verdict["recommended_departure"] == "08:15"


def test_llm_falls_back_when_gemini_adapter_crashes(monkeypatch, fake_keys):
    def gemini_bug(*args, **kwargs):
        raise KeyError("unexpected response shape")

    monkeypatch.setattr(gemini, "chat", gemini_bug)
    monkeypatch.setattr(
        groq, "chat", lambda *a, **k: {"text": "ok", "tool_calls": [], "provider": "groq"}
    )
    result = llm.chat([{"role": "user", "content": "hi"}])
    assert result["provider"] == "groq"


def test_gemini_non_json_response_is_provider_error(monkeypatch, fake_keys):
    class BadResponse:
        status_code = 200
        text = "<html>gateway error</html>"

        def json(self):
            raise ValueError("not json")

    monkeypatch.setattr(gemini.requests, "post", lambda *a, **k: BadResponse())
    with pytest.raises(ProviderError, match="non-JSON"):
        gemini.chat([{"role": "user", "content": "hi"}])


def test_groq_non_json_response_is_provider_error(monkeypatch, fake_keys):
    class BadResponse:
        status_code = 200
        text = "<html>cloudflare</html>"

        def json(self):
            raise ValueError("not json")

    monkeypatch.setattr(groq.requests, "post", lambda *a, **k: BadResponse())
    with pytest.raises(ProviderError, match="non-JSON"):
        groq.chat([{"role": "user", "content": "hi"}])