Files
awoooi/apps/api/tests/test_agent99_sre_bridge.py
ogt 6cf8429d17
All checks were successful
CD Pipeline / workflow-shape (push) Successful in 0s
CD Pipeline / cancel-stale-cd (push) Has been skipped
CD Pipeline / tests (push) Successful in 2m32s
CD Pipeline / build-and-deploy (push) Successful in 8m2s
CD Pipeline / post-deploy-checks (push) Successful in 2m22s
fix(automation): enforce durable alert closure
Restore D037 durable incident readback without weakening database failure semantics.

Fence Agent99 dispatch and terminal learning to canonical identities, durable leases, verifier receipts, and reconciler-owned checkpoints.

Drain backup and restore legacy receipts in bounded cohorts with strict transport, claim, projection, and no-false-green controls.

Keep SSH service refusal visible while separating it from broker network-policy reachability.
2026-07-14 09:40:53 +08:00

1032 lines
35 KiB
Python

from __future__ import annotations
import json
import pytest
from src.services.agent99_controlled_dispatch_ledger import (
build_agent99_dispatch_identity,
build_agent99_dispatch_receipt_envelope,
)
from src.services.agent99_sre_bridge import (
acquire_agent99_sre_single_flight,
agent99_sre_single_flight_key,
bridge_alertmanager_to_agent99,
build_agent99_sre_alert,
dispatch_agent99_sre_alert_with_receipt,
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"]
def test_explicit_backup_identity_wins_over_generic_freshness_words() -> None:
payload = build_agent99_sre_alert(
alert_id="backup-escrow-missing",
alertname="BackupCredentialEscrowEvidenceMissing",
severity="critical",
namespace="awoooi-prod",
target_resource="backup_restore",
message=(
"backup freshness missing; escrow evidence missing; "
"readback verifier required"
),
labels={
"event_type": "backup_restore_escrow_signal",
"lane": "backup_restore_escrow_triage",
},
fingerprint="fp-backup-freshness-collision",
)
assert payload["kind"] == "backup_health"
assert payload["suggestedMode"] == "BackupCheck"
assert payload["controlledApply"] is False
assert "backup-readback" in payload["labels"]
assert "no-false-green" in payload["labels"]
assert "read-only" in payload["labels"]
assert "do not run backup/restore" in payload["instruction"]
assert "remote delete" in payload["instruction"]
assert payload["resolution"]["policy"] == "mode_verifier_can_resolve"
@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",
),
(
"Host112GuestNotReady",
"host112-guest-recovery",
"Host 112 console LightDM VMware Tools and Wazuh are not ready",
{"host": "112", "service": "host112-guest-recovery"},
"host112_guest_recovery",
"Recover",
),
(
"KubePodCrashLooping",
"awoooi-api",
"k3s pod crashloopbackoff",
{"service": "awoooi"},
"awoooi_k3s",
"AwoooRepair",
),
(
"ColdStartGateBlocked",
"cold-start-gate",
"full-stack cold-start recovery gate requires remediation",
{"service": "awoooi"},
"host_recovery",
"Recover",
),
(
"AWOOOICICDFailed",
"AWOOOI deployment",
"build-and-deploy deployment failed",
{"service": "awoooi"},
"cicd_failure",
"Status",
),
],
)
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_agent99_resolution_policy_uses_mode_verifier_only_for_stateful_alerts() -> None:
cold_start = build_agent99_sre_alert(
alert_id="cold-start",
alertname="ColdStartGateBlocked",
severity="critical",
namespace="awoooi-prod",
target_resource="cold-start-gate",
message="cold-start gate blocked",
)
cicd = build_agent99_sre_alert(
alert_id="cicd-failure",
alertname="AWOOOICICDFailed",
severity="critical",
namespace="awoooi-prod",
target_resource="AWOOOI deployment",
message="build-and-deploy failed",
)
assert cold_start["resolution"]["policy"] == "mode_verifier_can_resolve"
assert cicd["resolution"]["policy"] == "external_source_receipt_required"
def test_host112_guest_recovery_precedes_wazuh_security_classification() -> None:
payload = build_agent99_sre_alert(
alert_id="host112-guest-not-ready",
alertname="Host112GuestNotReady",
severity="critical",
namespace="host112",
target_resource="host112-guest-recovery",
message="Host 112 Console VMware Tools Wazuh services are not ready",
labels={"host": "112", "service": "host112-guest-recovery"},
)
assert payload["kind"] == "host112_guest_recovery"
assert payload["suggestedMode"] == "Recover"
assert payload["controlledApply"] is True
assert payload["resolution"]["policy"] == "mode_verifier_can_resolve"
assert "do not reboot" in payload["instruction"]
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_dispatch_receipt_keeps_relay_acceptance_without_raw_body(monkeypatch) -> None:
payload = build_agent99_sre_alert(
alert_id="receipt-smoke",
alertname="ColdStartGateBlocked",
severity="critical",
namespace="awoooi-prod",
target_resource="cold-start-gate",
message="cold-start gate blocked",
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.settings.AGENT99_SRE_ALERT_RELAY_URL",
"http://192.168.0.99:8787/agent99/sre-alert/",
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.post_agent99_sre_alert",
lambda _payload: {
"status": 202,
"body": '{"ok":true,"inboxTriggered":true,"incomingFile":"private-path"}',
},
)
receipt = dispatch_agent99_sre_alert_with_receipt(payload)
assert receipt == {
"schema_version": "agent99_sre_dispatch_receipt_v1",
"status": "accepted_inbox_triggered",
"transport": "relay",
"alert_id": "awoooi-alertmanager-receipt-smoke",
"kind": "host_recovery",
"http_status": 202,
"accepted": True,
"inbox_triggered": True,
"delivery_certainty": "delivered",
"stores_raw_response": False,
}
assert "private-path" not in str(receipt)
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) -> dict[str, object]:
return {"acquired": False, "reason": "single_flight_duplicate"}
dispatched: list[dict[str, object]] = []
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
duplicate,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload)
or {
"status": "accepted_inbox_triggered",
"transport": "relay",
"accepted": True,
"inbox_triggered": True,
},
)
result = await bridge_alertmanager_to_agent99(
alert_id="duplicate-alert",
alertname="UnknownObservation",
severity="critical",
namespace="node",
target_resource="192.168.0.120",
message="observation only",
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) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
dispatched: list[dict[str, object]] = []
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
winner,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload)
or {
"status": "accepted_inbox_triggered",
"transport": "relay",
"accepted": True,
"inbox_triggered": True,
},
)
result = await bridge_alertmanager_to_agent99(
alert_id="winner-alert",
alertname="UnknownObservation",
severity="critical",
namespace="node",
target_resource="192.168.0.120",
message="observation only",
fingerprint="winner-fingerprint",
)
assert result["status"] == "dispatched"
assert result["dispatch"] == "relay"
assert result["dispatchReceipt"]["accepted"] is True
assert result["dispatchReceipt"]["inbox_triggered"] is True
assert result["suggestedMode"] == "Status"
assert len(dispatched) == 1
@pytest.mark.asyncio
async def test_agent99_bridge_releases_lock_when_receipt_is_not_accepted(
monkeypatch,
) -> None:
async def winner(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
released: list[tuple[str, str]] = []
async def release(fingerprint: str, alert_id: str) -> None:
released.append((fingerprint, alert_id))
monkeypatch.setattr(
"src.services.agent99_sre_bridge.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_with_receipt",
lambda _payload: {
"status": "relay_response_not_accepted",
"transport": "relay",
"accepted": False,
"inbox_triggered": False,
"delivery_certainty": "not_delivered",
},
)
result = await bridge_alertmanager_to_agent99(
alert_id="receipt-rejected",
alertname="UnknownObservation",
severity="critical",
namespace="node",
target_resource="192.168.0.120",
message="observation only",
fingerprint="receipt-rejected-fingerprint",
)
assert result["status"] == "failed"
assert result["reason"] == "relay_response_not_accepted"
assert result["dispatchReceipt"]["accepted"] is False
assert released == [("receipt-rejected-fingerprint", "receipt-rejected")]
@pytest.mark.asyncio
async def test_agent99_bridge_releases_lock_when_transport_fails(monkeypatch) -> None:
async def winner(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
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]) -> dict[str, object]:
raise RuntimeError("relay unavailable")
monkeypatch.setattr(
"src.services.agent99_sre_bridge.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_with_receipt",
fail_transport,
)
result = await bridge_alertmanager_to_agent99(
alert_id="transport-failure",
alertname="UnknownObservation",
severity="critical",
namespace="node",
target_resource="192.168.0.120",
message="observation only",
fingerprint="transport-fingerprint",
)
assert result["status"] == "failed"
assert result["reason"] == "RuntimeError"
assert result["dispatchPerformed"] is False
assert result["runtimeClosureVerified"] is False
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"
class _FakeControlledDispatchLedger:
def __init__(self) -> None:
self.reserved = False
self.receipt: dict[str, object] | None = None
self.complete_calls = 0
self.defer_calls = 0
async def reserve(self, *, identity, payload): # type: ignore[no-untyped-def]
if self.reserved:
return {
"status": "idempotent_completed",
"claimed": False,
"identity": identity.public_dict(),
"receipt": self.receipt,
}
self.reserved = True
return {
"status": "reserved",
"claimed": True,
"identity": identity.public_dict(),
"receipt": None,
"claim_token": "claim-attempt-1",
}
async def complete( # type: ignore[no-untyped-def]
self, *, identity, claim_token, dispatch_receipt, **kwargs
) -> dict[str, object]:
self.complete_calls += 1
self.receipt = {
**build_agent99_dispatch_receipt_envelope(
identity=identity,
dispatch_receipt=dispatch_receipt,
),
"receipt_persisted": True,
}
return self.receipt
async def mark_delivery_attempt_started( # type: ignore[no-untyped-def]
self, *, identity, claim_token
) -> dict[str, object]:
self.receipt = {
"status": "dispatch_delivery_unknown_reconcile_only",
"identity": identity.public_dict(),
"retry_policy": {"reconcile_only": True, "safe_to_retry": False},
"receipt_persisted": True,
"runtime_closure_verified": False,
}
return self.receipt
async def defer_claim_retryable( # type: ignore[no-untyped-def]
self, *, identity, claim_token, reason, retry_after_seconds
) -> dict[str, object]:
self.defer_calls += 1
self.receipt = {
"schema_version": "agent99_controlled_dispatch_receipt_v1",
"status": "dispatch_lock_retryable",
"identity": identity.public_dict(),
"dispatch_accepted": False,
"inbox_triggered": False,
"dispatch_promoted": False,
"controlled_apply_authorized": False,
"retry_policy": {
"safe_to_retry": True,
"retry_same_generation": True,
"requires_generation_advance": False,
"retry_after_seconds": retry_after_seconds,
"reason": reason,
},
"receipt_persisted": True,
"runtime_closure_verified": False,
}
return self.receipt
def test_agent99_dispatch_identity_is_stable_across_projections() -> None:
alertmanager = build_agent99_dispatch_identity(
project_id="awoooi",
incident_id="INC-20260711-11C751",
source_fingerprint="source-fingerprint",
route_id="agent99_cold_start_recovery",
execution_generation="1",
)
telegram = build_agent99_dispatch_identity(
project_id="awoooi",
incident_id="INC-20260711-11C751",
source_fingerprint="source-fingerprint",
route_id="agent99_cold_start_recovery",
execution_generation="1",
)
assert telegram.run_id == alertmanager.run_id
assert telegram.trace_id == alertmanager.trace_id
assert telegram.idempotency_key == alertmanager.idempotency_key
assert telegram.work_item_id == alertmanager.work_item_id
@pytest.mark.asyncio
async def test_mutating_dispatch_is_deduplicated_by_postgres_identity(
monkeypatch,
) -> None:
ledger = _FakeControlledDispatchLedger()
dispatched: list[dict[str, object]] = []
async def acquired(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
monkeypatch.setattr(
"src.services.agent99_sre_bridge.get_agent99_dispatch_ledger",
lambda: ledger,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
acquired,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload)
or {
"status": "accepted_inbox_triggered",
"transport": "relay",
"alert_id": str(payload["id"]),
"kind": str(payload["kind"]),
"accepted": True,
"inbox_triggered": True,
},
)
common = {
"alertname": "ColdStartGateBlocked",
"severity": "critical",
"namespace": "host_recovery",
"target_resource": "cold-start-gate",
"message": "cold-start gate blocked",
"fingerprint": "source-fingerprint",
"project_id": "awoooi",
"incident_id": "INC-20260711-11C751",
"route_id": "agent99_cold_start_recovery",
}
first = await bridge_alertmanager_to_agent99(alert_id="alertmanager-1", **common)
second = await bridge_alertmanager_to_agent99(alert_id="telegram-card-2", **common)
assert first["status"] == "dispatched"
assert second["status"] == "deduplicated"
assert len(dispatched) == 1
assert dispatched[0]["force"] is False
assert first["identity"] == second["identity"]
assert first["correlatedReceipt"]["runtime_execution_authorized"] is False
assert first["correlatedReceipt"]["runtime_closure_verified"] is False
assert first["correlatedReceipt"]["verifier"]["status"] == "pending"
assert first["correlatedReceipt"]["learning_writeback"]["status"] == (
"pending_verifier"
)
@pytest.mark.asyncio
async def test_mutating_dispatch_redis_failure_performs_zero_transport(
monkeypatch,
) -> None:
ledger = _FakeControlledDispatchLedger()
dispatched: list[dict[str, object]] = []
async def redis_unavailable(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": False, "reason": "redis_unavailable_fail_closed"}
monkeypatch.setattr(
"src.services.agent99_sre_bridge.get_agent99_dispatch_ledger",
lambda: ledger,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
redis_unavailable,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload),
)
result = await bridge_alertmanager_to_agent99(
alert_id="cold-start-redis-failure",
alertname="ColdStartGateBlocked",
severity="critical",
namespace="host_recovery",
target_resource="cold-start-gate",
message="cold-start gate blocked",
fingerprint="redis-failure-fingerprint",
incident_id="INC-20260711-11C751",
route_id="agent99_cold_start_recovery",
)
assert result["status"] == "retryable"
assert result["reason"] == "redis_unavailable_fail_closed"
assert result["dispatchPerformed"] is False
assert dispatched == []
assert ledger.complete_calls == 0
assert ledger.defer_calls == 1
assert result["correlatedReceipt"]["receipt_persisted"] is True
assert result["correlatedReceipt"]["runtime_closure_verified"] is False
@pytest.mark.asyncio
async def test_mutating_single_flight_redis_exception_is_fail_closed(
monkeypatch,
) -> None:
monkeypatch.setattr(
"src.core.redis_client.get_redis",
lambda: (_ for _ in ()).throw(RuntimeError("redis down")),
)
decision = await acquire_agent99_sre_single_flight(
"same-controlled-idempotency-key",
"same-run-owner",
fail_closed=True,
)
assert decision == {
"acquired": False,
"reason": "redis_unavailable_fail_closed",
}
@pytest.mark.asyncio
async def test_backupcheck_read_only_dispatch_has_durable_same_run_receipt(
monkeypatch,
) -> None:
ledger = _FakeControlledDispatchLedger()
dispatched: list[dict[str, object]] = []
async def acquired(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
monkeypatch.setattr(
"src.services.agent99_sre_bridge.get_agent99_dispatch_ledger",
lambda: ledger,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
acquired,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload)
or {
"schema_version": "agent99_sre_dispatch_receipt_v1",
"status": "accepted_inbox_triggered",
"transport": "relay",
"alert_id": str(payload["id"]),
"kind": str(payload["kind"]),
"accepted": True,
"inbox_triggered": True,
"delivery_certainty": "delivered",
},
)
result = await bridge_alertmanager_to_agent99(
alert_id="backup-readback",
alertname="BackupCredentialEscrowEvidenceMissing",
severity="warning",
namespace="awoooi-prod",
target_resource="backup_restore",
message="freshness escrow restore evidence required",
labels={"event_type": "backup_restore_escrow_signal"},
fingerprint="backup-fingerprint",
incident_id="INC-20260711-BACKUP",
route_id="agent99:backup_health:BackupCheck",
)
assert result["status"] == "dispatched"
assert result["identity"]["run_id"]
assert result["identity"]["trace_id"]
assert result["identity"]["work_item_id"]
assert result["correlatedReceipt"]["receipt_persisted"] is True
assert dispatched[0]["suggestedMode"] == "BackupCheck"
assert dispatched[0]["controlledApply"] is False
assert result["correlatedReceipt"]["controlled_apply_authorized"] is False
class _DeliveryUnknownLedger:
def __init__(self) -> None:
self.reserved = False
self.receipt: dict[str, object] | None = None
async def reserve(self, *, identity, payload): # type: ignore[no-untyped-def]
if self.reserved:
return {
"status": "idempotent_running",
"claimed": False,
"receipt": self.receipt,
}
self.reserved = True
self.receipt = {
"schema_version": "agent99_controlled_dispatch_receipt_v1",
"status": "dispatch_reserved",
"identity": identity.public_dict(),
"outbox": {"status": "pending_delivery"},
}
return {
"status": "reserved",
"claimed": True,
"claim_token": "unknown-delivery-claim",
"receipt": self.receipt,
}
async def complete(self, **_kwargs): # type: ignore[no-untyped-def]
return {
**(self.receipt or {}),
"status": "receipt_persistence_failed_no_runtime_closure",
"receipt_persisted": False,
}
async def mark_delivery_attempt_started( # type: ignore[no-untyped-def]
self, *, identity, claim_token
) -> dict[str, object]:
self.receipt = {
"schema_version": "agent99_controlled_dispatch_receipt_v1",
"status": "dispatch_delivery_unknown_reconcile_only",
"identity": identity.public_dict(),
"retry_policy": {"safe_to_retry": False, "reconcile_only": True},
"receipt_persisted": True,
"runtime_closure_verified": False,
}
return self.receipt
async def defer_claim_retryable(self, **_kwargs): # type: ignore[no-untyped-def]
raise AssertionError("lock is acquired in this test")
@pytest.mark.asyncio
async def test_accepted_transport_with_receipt_write_failure_is_not_replayed(
monkeypatch,
) -> None:
ledger = _DeliveryUnknownLedger()
dispatched: list[dict[str, object]] = []
async def acquired(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
monkeypatch.setattr(
"src.services.agent99_sre_bridge.get_agent99_dispatch_ledger",
lambda: ledger,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
acquired,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload)
or {
"status": "accepted_inbox_triggered",
"transport": "relay",
"alert_id": str(payload["id"]),
"kind": str(payload["kind"]),
"accepted": True,
"inbox_triggered": True,
"delivery_certainty": "delivered",
},
)
common = {
"alertname": "ColdStartGateBlocked",
"severity": "critical",
"namespace": "host_recovery",
"target_resource": "cold-start-gate",
"message": "cold-start gate blocked",
"fingerprint": "unknown-delivery-fingerprint",
"incident_id": "INC-20260711-11C751",
"route_id": "agent99:host_recovery:Recover",
}
first = await bridge_alertmanager_to_agent99(alert_id="first", **common)
second = await bridge_alertmanager_to_agent99(alert_id="second", **common)
assert first["reason"] == "dispatch_receipt_persistence_failed"
assert second["dispatchPerformed"] is False
assert len(dispatched) == 1
class _GenerationAdvanceLedger(_FakeControlledDispatchLedger):
async def reserve(self, *, identity, payload): # type: ignore[no-untyped-def]
if identity.execution_generation == "1":
return {
"status": "idempotent_failed",
"claimed": False,
"receipt": {
"status": "dispatch_not_delivered_retry_safe",
"identity": identity.public_dict(),
"retry_policy": {
"safe_to_retry": True,
"requires_generation_advance": True,
"max_generation": 3,
},
},
}
return await super().reserve(identity=identity, payload=payload)
@pytest.mark.asyncio
async def test_safe_not_delivered_failure_advances_generation_before_dispatch(
monkeypatch,
) -> None:
ledger = _GenerationAdvanceLedger()
dispatched: list[dict[str, object]] = []
async def acquired(*_args, **_kwargs) -> dict[str, object]:
return {"acquired": True, "reason": "acquired"}
monkeypatch.setattr(
"src.services.agent99_sre_bridge.get_agent99_dispatch_ledger",
lambda: ledger,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.acquire_agent99_sre_single_flight",
acquired,
)
monkeypatch.setattr(
"src.services.agent99_sre_bridge.dispatch_agent99_sre_alert_with_receipt",
lambda payload: dispatched.append(payload)
or {
"status": "accepted_inbox_triggered",
"transport": "relay",
"alert_id": str(payload["id"]),
"kind": str(payload["kind"]),
"accepted": True,
"inbox_triggered": True,
"delivery_certainty": "delivered",
},
)
result = await bridge_alertmanager_to_agent99(
alert_id="retry-generation",
alertname="ColdStartGateBlocked",
severity="critical",
namespace="host_recovery",
target_resource="cold-start-gate",
message="cold-start gate blocked",
fingerprint="retry-generation-fingerprint",
incident_id="INC-20260711-11C751",
route_id="agent99:host_recovery:Recover",
)
assert result["status"] == "dispatched"
assert result["identity"]["execution_generation"] == "2"
assert len(dispatched) == 1