ariadne/tests/test_hermes_infra_signals.py

from __future__ import annotations

import pytest

from ariadne.services import hermes_infra_signals as module


def _bundle(regions=None, tail=None, **jenkins):  # type: ignore[no-untyped-def]
    payload = {
        "job": "titan-iac",
        "console_failures": [{"marker": "ERROR:", "line_number": 10, "text": text} for text in regions or []],
        "console_tail": tail,
    }
    payload.update(jenkins)
    return {"incident_id": "titan-iac/12", "jenkins": payload, "log_evidence": {"records": []}}


@pytest.mark.parametrize("marker", module.INFRA_MARKERS)
def test_every_marker_is_detected_in_a_region(marker) -> None:
    bundle = _bundle(regions=[f"+ pip install -r requirements.txt\n{marker}\nbuild step failed"])

    assert module.has_transient_infra_signature(bundle) == (True, marker)


@pytest.mark.parametrize("marker", module.INFRA_MARKERS)
def test_every_marker_is_detected_in_the_tail(marker) -> None:
    bundle = _bundle(tail=f"Finished: FAILURE\n{marker.upper()}")

    assert module.has_transient_infra_signature(bundle) == (True, marker)


def test_real_dns_failure_from_pip_install() -> None:
    text = (
        "+ pip install -r requirements.txt\n"
        "WARNING: Retrying (Retry(total=4)) after connection broken by "
        "'NewConnectionError(...: Failed to establish a new connection: "
        "[Errno -3] Temporary failure in name resolution')\n"
        "ERROR: Could not install packages due to an OSError\n"
    )
    matched, marker = module.has_transient_infra_signature(_bundle(regions=[text]))

    assert matched is True
    assert marker == "failed to establish a new connection"


def test_real_scm_connect_failure_from_git_checkout() -> None:
    text = (
        "+ git fetch --tags --force --progress -- https://scm.bstein.dev/bstein/titan-iac.git\n"
        "fatal: unable to access 'https://scm.bstein.dev/bstein/titan-iac.git/': "
        "Failed to connect to scm.bstein.dev port 443 after 130626 ms: Connection timed out\n"
    )
    matched, marker = module.has_transient_infra_signature(_bundle(regions=[text]))

    assert matched is True
    assert marker == "failed to connect to"


def test_matching_is_case_insensitive() -> None:
    bundle = _bundle(regions=["Kubelet reported ErrImagePull for the agent container"])

    assert module.has_transient_infra_signature(bundle) == (True, "errimagepull")


def test_earliest_region_wins_over_later_regions_and_tail() -> None:
    bundle = _bundle(
        regions=["stage one: connection refused", "stage two: 502 bad gateway"],
        tail="Finished: FAILURE (i/o timeout)",
    )

    assert module.has_transient_infra_signature(bundle) == (True, "connection refused")


def test_earliest_marker_within_one_region_wins() -> None:
    text = "could not resolve host: registry.example\nthen later a 503 service unavailable"
    bundle = _bundle(regions=[text])

    assert module.has_transient_infra_signature(bundle) == (True, "could not resolve host")


def test_regions_are_searched_before_the_tail() -> None:
    bundle = _bundle(regions=["nothing infrastructural here"], tail="tls handshake timeout")

    assert module.has_transient_infra_signature(bundle) == (True, "tls handshake timeout")


def test_disk_exhaustion_is_deliberately_not_transient() -> None:
    text = "cp: error writing '/home/jenkins/agent/workspace/x': No space left on device"
    bundle = _bundle(regions=[text], tail=text)

    assert "no space left on device" not in module.INFRA_MARKERS
    assert module.has_transient_infra_signature(bundle) == (False, None)


def test_ordinary_test_failure_is_not_transient() -> None:
    text = "FAILED tests/test_discount.py::test_bulk - AssertionError: assert 0.9 == 0.8"
    bundle = _bundle(regions=[text], tail="short test summary info")

    assert module.has_transient_infra_signature(bundle) == (False, None)


@pytest.mark.parametrize(
    "bundle",
    [
        {},
        {"jenkins": None},
        {"jenkins": {}},
        {"jenkins": {"console_failures": None, "console_tail": None}},
        {"jenkins": {"console_failures": "not-a-list", "console_tail": 7}},
        {"jenkins": {"console_failures": ["not-a-dict", {"text": None}], "console_tail": ""}},
        {"jenkins": {"console_failures": [{}]}},
    ],
)
def test_empty_or_malformed_bundles_never_match(bundle) -> None:
    assert module.has_transient_infra_signature(bundle) == (False, None)


def test_non_dict_bundle_never_raises() -> None:
    assert module.has_transient_infra_signature(None) == (False, None)  # type: ignore[arg-type]
    assert module.has_transient_infra_signature("connection refused") == (False, None)  # type: ignore[arg-type]


def test_unreadable_bundle_is_swallowed() -> None:
    class Exploding(dict):
        def get(self, *args, **kwargs):  # type: ignore[no-untyped-def]
            raise RuntimeError("boom")

    assert module.has_transient_infra_signature(Exploding()) == (False, None)


def test_markers_are_lowercase_and_unique() -> None:
    assert list(module.INFRA_MARKERS) == [marker.lower() for marker in module.INFRA_MARKERS]
    assert len(set(module.INFRA_MARKERS)) == len(module.INFRA_MARKERS)