tests/integration/test_failure_scenarios.py

"""Failure-scenario tests — verify graceful degradation."""

from __future__ import annotations

import asyncio
import base64

import pytest

from config.settings import Settings
from api.container import build_container
from models.jobs import JobKind, JobRequest
from models.providers import ProviderCapability
from providers.base import BaseProvider, ProviderResult


class FailingProvider(BaseProvider):
    """A provider that always raises an exception."""
    name = "failing_provider"
    capability = ProviderCapability.DETECTION

    def __init__(self, settings=None):
        super().__init__(settings=settings)

    def is_available(self) -> bool:
        return True

    def _run(self, pipeline_output):
        raise RuntimeError("Intentional failure for testing")


class NotConfiguredProvider(BaseProvider):
    """A provider that reports itself as not available."""
    name = "not_configured_provider"
    capability = ProviderCapability.DETECTION

    def is_available(self) -> bool:
        return False

    def _run(self, pipeline_output):
        raise RuntimeError("Should never be called")


@pytest.fixture
def container_with_failing_provider(test_settings):
    container = build_container(test_settings)
    # Register a failing provider into the registry
    container.registry._providers["failing_provider"] = FailingProvider(settings=test_settings)
    container.registry._providers["not_configured_provider"] = NotConfiguredProvider(settings=test_settings)
    return container


class TestGracefulDegradation:
    def test_failing_provider_does_not_crash_job(self, container_with_failing_provider, sample_image_b64):
        """A failing provider should be captured as failed evidence, not crash the job."""
        c = container_with_failing_provider
        # Use the failing provider explicitly
        req = JobRequest(
            kind=JobKind.DETECTION,
            image_base64=sample_image_b64,
            providers=["failing_provider"],
        )
        result = asyncio.run(c.detection_service.detect(req))
        assert result["success"] is True  # The JOB succeeded (with failures in evidence)
        report = result["report"]
        assert "failing_provider" in report["metadata"]["providers_failed"]
        assert len(report["evidence"]) >= 1
        # Find the failing provider's evidence
        failing_evidence = [e for e in report["evidence"] if e["provider"] == "failing_provider"]
        assert len(failing_evidence) == 1
        assert failing_evidence[0]["success"] is False
        assert "Intentional failure" in failing_evidence[0]["error"]

    def test_not_configured_provider_returns_not_configured(self, container_with_failing_provider, sample_image_b64):
        """A provider that's not available should return a NotConfigured result."""
        c = container_with_failing_provider
        req = JobRequest(
            kind=JobKind.DETECTION,
            image_base64=sample_image_b64,
            providers=["not_configured_provider"],
        )
        result = asyncio.run(c.detection_service.detect(req))
        assert result["success"] is True  # JOB succeeds
        report = result["report"]
        # Should have evidence with the not_configured error
        nc_evidence = [e for e in report["evidence"] if e["provider"] == "not_configured_provider"]
        assert len(nc_evidence) == 1
        assert nc_evidence[0]["success"] is False
        assert "NotConfigured" in nc_evidence[0]["error_type"]

    def test_mixed_success_and_failure_preserves_both(self, container_with_failing_provider, sample_image_b64):
        """If some providers succeed and others fail, both should be in evidence."""
        c = container_with_failing_provider
        req = JobRequest(
            kind=JobKind.DETECTION,
            image_base64=sample_image_b64,
            # Both haar (works) and failing_provider (fails) should run
            providers=["haar", "failing_provider"],
        )
        result = asyncio.run(c.detection_service.detect(req))
        report = result["report"]
        assert "haar" in report["metadata"]["providers_succeeded"]
        assert "failing_provider" in report["metadata"]["providers_failed"]
        assert len(report["evidence"]) == 2

    def test_invalid_image_returns_validation_error(self, container_with_failing_provider):
        """An invalid image should return a structured validation error."""
        c = container_with_failing_provider
        req = JobRequest(
            kind=JobKind.DETECTION,
            image_base64="!!!invalid base64!!!",
        )
        result = asyncio.run(c.detection_service.detect(req))
        assert result["success"] is False
        assert "error" in result
        assert result.get("error_type") == "ValidationError"

    def test_missing_image_returns_validation_error(self, container_with_failing_provider):
        """No image input should return a validation error."""
        c = container_with_failing_provider
        req = JobRequest(kind=JobKind.DETECTION)
        result = asyncio.run(c.detection_service.detect(req))
        assert result["success"] is False
        assert result.get("error_type") == "ValidationError"


class TestCircuitBreaker:
    def test_circuit_opens_after_threshold_failures(self, test_settings, sample_image_b64):
        """After N consecutive failures, the circuit should open."""
        # Lower threshold for test
        test_settings.circuit_breaker_failure_threshold = 3
        test_settings.circuit_breaker_recovery_seconds = 60
        container = build_container(test_settings)
        container.registry._providers["failing_provider"] = FailingProvider(settings=test_settings)

        # Trigger failures
        for _ in range(3):
            req = JobRequest(
                kind=JobKind.DETECTION,
                image_base64=sample_image_b64,
                providers=["failing_provider"],
            )
            asyncio.run(container.detection_service.detect(req))

        # Circuit should now be open
        assert container.health_monitor.is_available("failing_provider") is False

    def test_circuit_blocks_subsequent_invocations(self, test_settings, sample_image_b64):
        """When circuit is open, the orchestrator should skip the provider."""
        test_settings.circuit_breaker_failure_threshold = 2
        test_settings.circuit_breaker_recovery_seconds = 60
        container = build_container(test_settings)
        container.registry._providers["failing_provider"] = FailingProvider(settings=test_settings)

        # Trip the breaker
        for _ in range(2):
            req = JobRequest(
                kind=JobKind.DETECTION,
                image_base64=sample_image_b64,
                providers=["failing_provider"],
            )
            asyncio.run(container.detection_service.detect(req))

        # Now run again — provider should be skipped (no new evidence)
        req = JobRequest(
            kind=JobKind.DETECTION,
            image_base64=sample_image_b64,
            providers=["failing_provider"],
        )
        result = asyncio.run(container.detection_service.detect(req))
        report = result["report"]
        # When circuit is open, provider is skipped — so providers_invoked is empty
        assert "failing_provider" not in report["metadata"]["providers_invoked"]