387 lines
12 KiB
Python
387 lines
12 KiB
Python
from __future__ import annotations
|
|
|
|
import json
|
|
|
|
import pytest
|
|
|
|
from src.services.agent99_sre_bridge import (
|
|
agent99_sre_single_flight_key,
|
|
bridge_alertmanager_to_agent99,
|
|
build_agent99_sre_alert,
|
|
post_agent99_sre_alert,
|
|
release_agent99_sre_single_flight,
|
|
resolve_agent99_alert_kind,
|
|
write_agent99_sre_alert,
|
|
)
|
|
|
|
|
|
def test_provider_freshness_alert_builds_agent99_contract() -> None:
|
|
payload = build_agent99_sre_alert(
|
|
alert_id="alert-20260709233000",
|
|
alertname="provider_freshness_signal",
|
|
severity="warning",
|
|
namespace="awoooi-prod",
|
|
target_resource="provider_freshness_signal",
|
|
message=(
|
|
"raw_signal_normalized controlled_runtime_candidate freshness "
|
|
"last seen missing direct ref verifier readback"
|
|
),
|
|
labels={
|
|
"service": "awoooi",
|
|
"token": "must-not-leak",
|
|
},
|
|
annotations={
|
|
"summary": "Provider freshness evidence is missing",
|
|
},
|
|
fingerprint="fp-provider-freshness",
|
|
alert_category="source_provider_freshness_triage",
|
|
notification_type="raw_signal_normalized",
|
|
source_url="https://awoooi.wooo.work/api/v1/webhooks/alertmanager",
|
|
)
|
|
|
|
assert payload["source"] == "awoooi-api-alertmanager"
|
|
assert payload["kind"] == "provider_freshness_signal"
|
|
assert payload["suggestedMode"] == "ProviderFreshness"
|
|
assert payload["controlledApply"] is False
|
|
assert payload["force"] is True
|
|
assert "provider-freshness" in payload["labels"]
|
|
assert "no-false-green" in payload["labels"]
|
|
assert "token=[redacted]" in payload["labels"]
|
|
assert "must-not-leak" not in json.dumps(payload)
|
|
assert "do not switch provider" in payload["instruction"]
|
|
assert payload["routing"]["routeSource"] == "structured_kind"
|
|
assert payload["routing"]["correlationKey"] == payload["routing"]["groupKey"]
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("alertname", "target", "message", "labels", "expected_kind", "expected_mode"),
|
|
[
|
|
(
|
|
"HostBackupFailed",
|
|
"backup",
|
|
"snapshot stale",
|
|
{},
|
|
"backup_health",
|
|
"BackupCheck",
|
|
),
|
|
(
|
|
"provider_freshness_signal",
|
|
"provider_freshness_signal",
|
|
"last_seen missing",
|
|
{},
|
|
"provider_freshness_signal",
|
|
"ProviderFreshness",
|
|
),
|
|
(
|
|
"RebootAutoRecoveryActiveBlocker",
|
|
"reboot-auto-recovery-slo",
|
|
"all_required_hosts_not_in_10_minute_reboot_window cpu pressure",
|
|
{"service": "awoooi"},
|
|
"host_recovery",
|
|
"Recover",
|
|
),
|
|
(
|
|
"PostgresqlDown",
|
|
"postgresql",
|
|
"database unavailable deployment awoooi",
|
|
{"service": "awoooi"},
|
|
"database_health",
|
|
"Status",
|
|
),
|
|
(
|
|
"AlertChainBroken",
|
|
"alertmanager",
|
|
"monitoring pipeline down awoooi",
|
|
{"service": "awoooi"},
|
|
"alert_chain_health",
|
|
"Status",
|
|
),
|
|
(
|
|
"HostDown",
|
|
"192.168.0.120",
|
|
"host_down vm_not_running",
|
|
{"host": "192.168.0.120"},
|
|
"host_recovery",
|
|
"Recover",
|
|
),
|
|
(
|
|
"KubePodCrashLooping",
|
|
"awoooi-api",
|
|
"k3s pod crashloopbackoff",
|
|
{"service": "awoooi"},
|
|
"awoooi_k3s",
|
|
"AwoooRepair",
|
|
),
|
|
],
|
|
)
|
|
def test_agent99_routing_replay_corpus(
|
|
alertname: str,
|
|
target: str,
|
|
message: str,
|
|
labels: dict[str, str],
|
|
expected_kind: str,
|
|
expected_mode: str,
|
|
) -> None:
|
|
payload = build_agent99_sre_alert(
|
|
alert_id=f"replay-{alertname}",
|
|
alertname=alertname,
|
|
severity="critical",
|
|
namespace="awoooi-prod",
|
|
target_resource=target,
|
|
message=message,
|
|
labels=labels,
|
|
fingerprint=f"fp-{alertname}",
|
|
)
|
|
|
|
assert payload["kind"] == expected_kind
|
|
assert payload["suggestedMode"] == expected_mode
|
|
|
|
|
|
def test_alert_kind_mapping_for_route_and_performance() -> None:
|
|
assert (
|
|
resolve_agent99_alert_kind(
|
|
alertname="PublicRoute502",
|
|
severity="critical",
|
|
namespace="awoooi-prod",
|
|
target_resource="awoooi-web",
|
|
message="HTTP 502",
|
|
)
|
|
== "public_route_502"
|
|
)
|
|
assert (
|
|
resolve_agent99_alert_kind(
|
|
alertname="HostHighCpuLoad",
|
|
severity="warning",
|
|
namespace="node",
|
|
target_resource="192.168.0.110",
|
|
message="load per core is high",
|
|
)
|
|
== "performance_pressure"
|
|
)
|
|
|
|
|
|
def test_write_agent99_sre_alert_to_configured_inbox(tmp_path) -> None:
|
|
payload = build_agent99_sre_alert(
|
|
alert_id="alert-write-smoke",
|
|
alertname="HostHighCpuLoad",
|
|
severity="warning",
|
|
namespace="node",
|
|
target_resource="192.168.0.110",
|
|
message="CPU load per core is high",
|
|
labels={"host": "192.168.0.110"},
|
|
)
|
|
|
|
written = write_agent99_sre_alert(payload, inbox_path=str(tmp_path))
|
|
|
|
assert written is not None
|
|
data = json.loads(written.read_text(encoding="utf-8"))
|
|
assert data["id"] == "awoooi-alertmanager-alert-write-smoke"
|
|
assert data["kind"] == "performance_pressure"
|
|
assert data["suggestedMode"] == "Perf"
|
|
|
|
|
|
def test_post_agent99_sre_alert_to_relay(monkeypatch) -> None:
|
|
payload = build_agent99_sre_alert(
|
|
alert_id="relay-smoke",
|
|
alertname="provider_freshness_signal",
|
|
severity="warning",
|
|
namespace="awoooi-prod",
|
|
target_resource="provider_freshness_signal",
|
|
message="freshness last_seen missing",
|
|
labels={"service": "awoooi", "authorization": "must-not-leak"},
|
|
)
|
|
seen: dict[str, object] = {}
|
|
|
|
class FakeResponse:
|
|
def getcode(self) -> int:
|
|
return 202
|
|
|
|
def read(self, _limit: int) -> bytes:
|
|
return b'{"ok":true}'
|
|
|
|
def __enter__(self) -> FakeResponse:
|
|
return self
|
|
|
|
def __exit__(self, *_args: object) -> None:
|
|
return None
|
|
|
|
def fake_urlopen(request, timeout): # type: ignore[no-untyped-def]
|
|
seen["url"] = request.full_url
|
|
seen["timeout"] = timeout
|
|
seen["headers"] = dict(request.header_items())
|
|
seen["payload"] = json.loads(request.data.decode("utf-8"))
|
|
return FakeResponse()
|
|
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.urllib.request.urlopen",
|
|
fake_urlopen,
|
|
)
|
|
|
|
receipt = post_agent99_sre_alert(
|
|
payload,
|
|
relay_url="http://192.168.0.99:8787/agent99/sre-alert/",
|
|
relay_token="relay-token",
|
|
timeout_seconds=1.5,
|
|
)
|
|
|
|
assert receipt == {"status": 202, "body": '{"ok":true}'}
|
|
assert seen["url"] == "http://192.168.0.99:8787/agent99/sre-alert/"
|
|
assert seen["timeout"] == 1.5
|
|
assert seen["payload"]["kind"] == "provider_freshness_signal"
|
|
assert seen["payload"]["suggestedMode"] == "ProviderFreshness"
|
|
assert "authorization=[redacted]" in seen["payload"]["labels"]
|
|
assert "must-not-leak" not in json.dumps(seen["payload"])
|
|
assert seen["headers"]["X-agent99-relay-token"] == "relay-token"
|
|
|
|
|
|
def test_agent99_single_flight_key_does_not_expose_fingerprint() -> None:
|
|
key = agent99_sre_single_flight_key("private/source/fingerprint")
|
|
|
|
assert key.startswith("agent99:sre_dispatch:single_flight:")
|
|
assert "private" not in key
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_agent99_bridge_suppresses_duplicate_single_flight(monkeypatch) -> None:
|
|
async def duplicate(*_args, **_kwargs) -> bool:
|
|
return False
|
|
|
|
dispatched: list[dict[str, object]] = []
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.try_acquire_agent99_sre_single_flight",
|
|
duplicate,
|
|
)
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert",
|
|
lambda payload: dispatched.append(payload) or "relay",
|
|
)
|
|
|
|
result = await bridge_alertmanager_to_agent99(
|
|
alert_id="duplicate-alert",
|
|
alertname="HostDown",
|
|
severity="critical",
|
|
namespace="node",
|
|
target_resource="192.168.0.120",
|
|
message="host_down",
|
|
fingerprint="same-fingerprint",
|
|
)
|
|
|
|
assert result["status"] == "suppressed"
|
|
assert result["reason"] == "single_flight_duplicate"
|
|
assert dispatched == []
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_agent99_bridge_dispatches_single_flight_winner(monkeypatch) -> None:
|
|
async def winner(*_args, **_kwargs) -> bool:
|
|
return True
|
|
|
|
dispatched: list[dict[str, object]] = []
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.try_acquire_agent99_sre_single_flight",
|
|
winner,
|
|
)
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert",
|
|
lambda payload: dispatched.append(payload) or "relay",
|
|
)
|
|
|
|
result = await bridge_alertmanager_to_agent99(
|
|
alert_id="winner-alert",
|
|
alertname="HostDown",
|
|
severity="critical",
|
|
namespace="node",
|
|
target_resource="192.168.0.120",
|
|
message="host_down",
|
|
fingerprint="winner-fingerprint",
|
|
)
|
|
|
|
assert result["status"] == "dispatched"
|
|
assert result["dispatch"] == "relay"
|
|
assert result["suggestedMode"] == "Recover"
|
|
assert len(dispatched) == 1
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_agent99_bridge_releases_lock_when_transport_fails(monkeypatch) -> None:
|
|
async def winner(*_args, **_kwargs) -> bool:
|
|
return True
|
|
|
|
released: list[tuple[str, str]] = []
|
|
|
|
async def release(fingerprint: str, alert_id: str) -> None:
|
|
released.append((fingerprint, alert_id))
|
|
|
|
def fail_transport(_payload: dict[str, object]) -> str:
|
|
raise RuntimeError("relay unavailable")
|
|
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.try_acquire_agent99_sre_single_flight",
|
|
winner,
|
|
)
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.release_agent99_sre_single_flight",
|
|
release,
|
|
)
|
|
monkeypatch.setattr(
|
|
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert",
|
|
fail_transport,
|
|
)
|
|
|
|
result = await bridge_alertmanager_to_agent99(
|
|
alert_id="transport-failure",
|
|
alertname="HostDown",
|
|
severity="critical",
|
|
namespace="node",
|
|
target_resource="192.168.0.120",
|
|
message="host_down",
|
|
fingerprint="transport-fingerprint",
|
|
)
|
|
|
|
assert result == {"status": "failed", "reason": "RuntimeError"}
|
|
assert released == [("transport-fingerprint", "transport-failure")]
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_agent99_single_flight_uses_redis_nx(monkeypatch) -> None:
|
|
seen: dict[str, object] = {}
|
|
|
|
class FakeRedis:
|
|
async def set(self, key, value, *, ex, nx): # type: ignore[no-untyped-def]
|
|
seen.update({"key": key, "value": value, "ex": ex, "nx": nx})
|
|
return True
|
|
|
|
monkeypatch.setattr("src.core.redis_client.get_redis", lambda: FakeRedis())
|
|
|
|
from src.services.agent99_sre_bridge import try_acquire_agent99_sre_single_flight
|
|
|
|
acquired = await try_acquire_agent99_sre_single_flight(
|
|
"source-fingerprint", "alert-1"
|
|
)
|
|
|
|
assert acquired is True
|
|
assert str(seen["key"]).startswith("agent99:sre_dispatch:single_flight:")
|
|
assert seen["value"] == "alert-1"
|
|
assert seen["ex"] == 600
|
|
assert seen["nx"] is True
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_agent99_single_flight_release_is_owner_checked(monkeypatch) -> None:
|
|
seen: dict[str, object] = {}
|
|
|
|
class FakeRedis:
|
|
async def eval(self, script, key_count, key, owner): # type: ignore[no-untyped-def]
|
|
seen.update(
|
|
{"script": script, "key_count": key_count, "key": key, "owner": owner}
|
|
)
|
|
return 1
|
|
|
|
monkeypatch.setattr("src.core.redis_client.get_redis", lambda: FakeRedis())
|
|
|
|
await release_agent99_sre_single_flight("source-fingerprint", "alert-1")
|
|
|
|
assert "redis.call('get'" in str(seen["script"])
|
|
assert seen["key_count"] == 1
|
|
assert seen["owner"] == "alert-1"
|