feat: expose multi-host runtime readiness

This commit is contained in:
2026-08-02 05:29:47 +02:00
parent c241085806
commit b464d016b2
6 changed files with 519 additions and 104 deletions
+47 -5
View File
@@ -52,11 +52,13 @@ def test_module_operational_check_failure_is_isolated() -> None:
raise RuntimeError("secret detail")
result = routes._module_operational_checks(
_Registry(OperationalCheckProviderRegistration(
module_id="example",
check_id="example.failed",
provider=provider,
)),
_Registry(
OperationalCheckProviderRegistration(
module_id="example",
check_id="example.failed",
provider=provider,
)
),
force=True,
)[0]
@@ -82,3 +84,43 @@ def test_shared_runtime_cluster_missing_replicas_blocks_readiness() -> None:
assert check["state"] == "error"
assert check["readiness_critical"] is True
assert check["metrics"]["recovery_required"] == 1
def test_shared_runtime_cluster_skew_or_unserved_queue_blocks_readiness() -> None:
check = routes._runtime_cluster_check(
{
"available": True,
"state_profile": "shared",
"nodes": [],
"expected": {"api": 2, "worker": 2},
"active": {"api": 2, "worker": 2},
"composition": {"skewed": True},
"software_versions": {"skewed": False},
"queues": {"missing": ["calendar"]},
"recovery": {"requires_attention": 0},
}
)
assert check["state"] == "error"
assert check["readiness_critical"] is True
assert check["metrics"]["composition_skewed"] is True
assert check["metrics"]["missing_queues"] == 1
def test_database_capacity_check_enforces_shared_rendered_budget(
monkeypatch,
) -> None:
monkeypatch.setattr(routes.core_settings, "state_profile", "shared")
monkeypatch.setattr(routes.core_settings, "database_connection_limit", 100)
monkeypatch.setattr(routes.core_settings, "database_connection_reserve", 10)
monkeypatch.setattr(routes.core_settings, "database_connection_available", 90)
monkeypatch.setattr(routes.core_settings, "database_connection_peak", 70)
healthy = routes._database_capacity_check()
assert healthy["state"] == "ok"
assert healthy["metrics"]["peak"] == 70
monkeypatch.setattr(routes.core_settings, "database_connection_peak", 91)
overrun = routes._database_capacity_check()
assert overrun["state"] == "error"
assert overrun["readiness_critical"] is True