SILVER/NODEDC_MISSION_CORE
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

fix(k1): stabilize live recovery and media admission

Browse Source
This commit is contained in:
DCCONSTRUCTIONS
2026-08-22 00:03:01 +03:00
parent 7217244886
commit eaad9deda1
29 changed files with 1645 additions and 199 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/test_connection_supervisor.py
+36 -37
View File
@@ -1817,7 +1817,7 @@ def test_same_path_positive_reducer_observation_resets_technical_failure_streak(
asyncio.run(scenario())
def test_unrelated_supervisor_revisions_do_not_reset_technical_failure_streak() -> None:
def test_unrelated_supervisor_revisions_preserve_exact_configured_route() -> None:
async def scenario() -> None:
supervisor, epoch = _configured_unverified_supervisor()
path = _available_path()
@@ -1836,8 +1836,12 @@ def test_unrelated_supervisor_revisions_do_not_reset_technical_failure_streak()
async def host_probe(_target: EndpointTarget) -> HostPathProbeResult:
return technical_timeout
tcp_calls = 0
async def tcp_probe(_target: EndpointTarget) -> bool:
raise AssertionError("technical failure must skip TCP")
nonlocal tcp_calls
tcp_calls += 1
return True
monitor = ReadOnlyConnectionMonitor(
supervisor,
@@ -1847,8 +1851,10 @@ def test_unrelated_supervisor_revisions_do_not_reset_technical_failure_streak()
)
first = await monitor.poll_once()
assert first.revision == initial.revision
assert first.host_path is initial.host_path
assert first.host_path.available is True
assert first.host_path.epoch == initial.host_path.epoch
assert first.endpoint.tcp_state == "reachable"
assert first.authority.control_allowed is False
assert supervisor.observe_endpoint(
target=TARGET,
intent_id="bridge-1",
@@ -1857,11 +1863,14 @@ def test_unrelated_supervisor_revisions_do_not_reset_technical_failure_streak()
)
endpoint_refresh = supervisor.snapshot()
assert endpoint_refresh.revision > first.revision
assert endpoint_refresh.host_path is initial.host_path
assert endpoint_refresh.host_path.epoch == initial.host_path.epoch
assert endpoint_refresh.host_path.fingerprint == initial.host_path.fingerprint
second = await monitor.poll_once()
assert second.revision == endpoint_refresh.revision
assert second.host_path is initial.host_path
assert second.host_path.available is True
assert second.host_path.epoch == initial.host_path.epoch
assert second.endpoint.tcp_state == "reachable"
assert second.authority.control_allowed is False
assert supervisor.observe_endpoint(
target=TARGET,
intent_id="bridge-1",
@@ -1870,15 +1879,17 @@ def test_unrelated_supervisor_revisions_do_not_reset_technical_failure_streak()
)
confirmed = await monitor.poll_once()
assert confirmed.host_path.available is False
assert confirmed.host_path.reason_code == "host-wifi-operation-timeout"
assert confirmed.host_path.available is True
assert confirmed.host_path.epoch == initial.host_path.epoch
assert confirmed.endpoint.tcp_state == "reachable"
assert confirmed.authority.control_allowed is False
assert tcp_calls == 3
await monitor.close()
asyncio.run(scenario())
def test_inflight_third_timeout_cannot_overwrite_a_concurrent_positive_observation() -> None:
def test_inflight_timeout_cannot_overwrite_a_concurrent_positive_observation() -> None:
async def scenario() -> None:
supervisor, initial_epoch = _configured_unverified_supervisor()
path = _available_path()
@@ -1892,20 +1903,20 @@ def test_inflight_third_timeout_cannot_overwrite_a_concurrent_positive_observati
observation_failure_class="association-observer",
kernel_route_fingerprint=path.kernel_route_fingerprint,
)
third_probe_entered = asyncio.Event()
release_third_probe = asyncio.Event()
timeout_probe_entered = asyncio.Event()
release_timeout_probe = asyncio.Event()
host_calls = 0
async def host_probe(_target: EndpointTarget) -> HostPathProbeResult:
nonlocal host_calls
host_calls += 1
if host_calls == 3:
third_probe_entered.set()
await release_third_probe.wait()
if host_calls == 1:
timeout_probe_entered.set()
await release_timeout_probe.wait()
return technical_timeout
async def tcp_probe(_target: EndpointTarget) -> bool:
raise AssertionError("technical failure must skip TCP")
return True
monitor = ReadOnlyConnectionMonitor(
supervisor,
@@ -1913,32 +1924,20 @@ def test_inflight_third_timeout_cannot_overwrite_a_concurrent_positive_observati
tcp_probe=tcp_probe,
target_provider=lambda: TARGET,
)
first = await monitor.poll_once()
second = await monitor.poll_once()
assert first.authority.control_allowed is False
assert second.authority.control_allowed is False
inflight_third = asyncio.create_task(monitor.poll_once())
await third_probe_entered.wait()
inflight_timeout = asyncio.create_task(monitor.poll_once())
await timeout_probe_entered.wait()
refreshed_epoch = supervisor.observe_host_path(path)
assert refreshed_epoch == initial_epoch
external_positive = supervisor.snapshot()
release_third_probe.set()
raced_timeout = await inflight_third
release_timeout_probe.set()
raced_timeout = await inflight_timeout
assert raced_timeout.revision == external_positive.revision
assert raced_timeout.host_path is external_positive.host_path
assert raced_timeout.revision >= external_positive.revision
assert raced_timeout.host_path.available is True
assert raced_timeout.host_path.epoch == external_positive.host_path.epoch
assert raced_timeout.host_path.fingerprint == external_positive.host_path.fingerprint
assert raced_timeout.endpoint.tcp_state == "reachable"
assert raced_timeout.authority.control_allowed is False
second_after_positive = await monitor.poll_once()
assert second_after_positive.revision == external_positive.revision
assert second_after_positive.host_path is external_positive.host_path
assert second_after_positive.authority.control_allowed is False
confirmed = await monitor.poll_once()
assert confirmed.host_path.available is False
assert confirmed.host_path.reason_code == "host-wifi-operation-timeout"
assert confirmed.authority.control_allowed is False
await monitor.close()
asyncio.run(scenario())