fix(perception): require joint clock readiness before source activation
This commit is contained in:
@@ -84,6 +84,10 @@ async def run(args):
|
||||
client = reader = None
|
||||
|
||||
try:
|
||||
# Reject a directory/missing input before establishing an activation clock.
|
||||
first = next(camera_events(args.camera_index, 1)).time_ns
|
||||
if not Path(args.sensor_archive).is_file() or not (root / "init.mp4").is_file():
|
||||
raise ValueError("raw recording inputs are missing")
|
||||
deadline = time.monotonic() + 120
|
||||
startup_path = control / "bootstrap.json" if args.cross_host else grant_path
|
||||
while not startup_path.exists():
|
||||
@@ -104,7 +108,6 @@ async def run(args):
|
||||
}
|
||||
else:
|
||||
clock, _ = read_grant(grant_path)
|
||||
first = next(camera_events(args.camera_index, 1)).time_ns
|
||||
if clock["source_zero_ns"] != first - 500_000_000:
|
||||
raise ValueError("recording prefix does not match admitted source clock")
|
||||
report.update(
|
||||
|
||||
@@ -22,6 +22,8 @@ class SourceControl:
|
||||
)
|
||||
self.source_zero_ns = int(bootstrap["source_zero_ns"])
|
||||
self.anchor = None # Latch once after the initial clock handshake, never on resume.
|
||||
self.proposal = None # Retain an uncertain proposal across a lost ACK.
|
||||
self.peer = None
|
||||
self.access = None
|
||||
self.confirmed = False
|
||||
self.failure = None
|
||||
@@ -31,43 +33,79 @@ class SourceControl:
|
||||
|
||||
async def update(self):
|
||||
probe, access = await self.client.poll()
|
||||
bounds = self.window.add(probe)
|
||||
self.window.add(probe)
|
||||
now = time.monotonic_ns()
|
||||
if self.anchor is None:
|
||||
try:
|
||||
self.window.current(now).require(now)
|
||||
except ClockMappingError:
|
||||
pass # Pre-start clock warmup, not a running source pause or retimestamp.
|
||||
else:
|
||||
self.anchor = SourceAnchor(self.source_zero_ns, now + 2_000_000_000)
|
||||
if self.anchor is not None:
|
||||
await self.client.acknowledge(probe, self.anchor, ended=self.ending)
|
||||
self.confirmed = True
|
||||
if (
|
||||
self.anchor is None
|
||||
and self.proposal is None
|
||||
and not self.ending
|
||||
and self.clocks_ready(now)
|
||||
):
|
||||
# Lead only allows bounded grant delivery/source setup AFTER two-sided readiness.
|
||||
# It is not a substitute for either clock gate or a retimestamp on recovery.
|
||||
self.proposal = SourceAnchor(self.source_zero_ns, now + 1_000_000_000)
|
||||
row = {
|
||||
"t1_source_send_ns": str(probe.local_send_ns),
|
||||
"t2_worker_receive_ns": str(probe.remote_receive_ns),
|
||||
"t3_worker_send_ns": str(probe.remote_send_ns),
|
||||
"t4_source_receive_ns": str(probe.local_receive_ns),
|
||||
"source_report_send_ns": str(now),
|
||||
"nonce": probe.nonce,
|
||||
"acknowledged": False,
|
||||
}
|
||||
if len(self.samples) >= 2048:
|
||||
raise ValueError("bounded clock diagnostic count exceeded")
|
||||
self.samples.append(row)
|
||||
# Every probe is echoed, even before an anchor exists. Worker warmup must
|
||||
# see the same evidence; starting the source requires its explicit ACK.
|
||||
self.peer = await self.client.acknowledge(
|
||||
probe, self.anchor or self.proposal, ended=self.ending
|
||||
)
|
||||
now = time.monotonic_ns()
|
||||
if self.anchor is not None and self.peer.anchor != self.anchor:
|
||||
raise ValueError("accepted source anchor disappeared")
|
||||
if self.peer.anchor is not None:
|
||||
if self.peer.anchor != (self.anchor or self.proposal):
|
||||
raise ValueError("foreign accepted source anchor")
|
||||
self.anchor = self.peer.anchor
|
||||
else:
|
||||
self.proposal = None # Explicit nonacceptance, not a lost response.
|
||||
self.confirmed = True
|
||||
self.access = access
|
||||
now = time.monotonic_ns()
|
||||
self.samples.append(
|
||||
bounds = self.window.current(now)
|
||||
row.update(
|
||||
{
|
||||
"at_ns": now,
|
||||
"t5_worker_report_receive_ns": str(self.peer.bounds.measured_at_ns),
|
||||
"t6_source_ack_receive_ns": str(now),
|
||||
"bounds": bounds.to_dict(),
|
||||
"peer": self.peer.to_dict(),
|
||||
"uncertainty_ns": bounds.uncertainty_ns(now),
|
||||
"ready": self.ready(),
|
||||
"acknowledged": self.anchor is not None,
|
||||
"ready": self.anchor is not None and self.confirmed and self.clocks_ready(now),
|
||||
"acknowledged": True,
|
||||
}
|
||||
)
|
||||
if len(self.samples) > 2048:
|
||||
raise ValueError("bounded clock diagnostic count exceeded")
|
||||
|
||||
def clocks_ready(self, now):
|
||||
if self.peer is None:
|
||||
return False
|
||||
try:
|
||||
own = self.window.current(now)
|
||||
own.require(now)
|
||||
# Conservative latest possible Worker time includes ACK transit and
|
||||
# elapsed time since receipt. A cached peer-ready boolean is unsafe.
|
||||
_, upper = own.offset_at(now)
|
||||
self.peer.bounds.require(now + upper)
|
||||
except ClockMappingError:
|
||||
return False
|
||||
return True
|
||||
|
||||
def ready(self):
|
||||
if self.failure:
|
||||
raise RuntimeError("source clock task failed") from self.failure
|
||||
if not self.confirmed:
|
||||
if not self.confirmed or self.anchor is None:
|
||||
return False
|
||||
try:
|
||||
now = time.monotonic_ns()
|
||||
self.window.current(now).require(now)
|
||||
except ClockMappingError:
|
||||
return False
|
||||
return True
|
||||
return self.clocks_ready(time.monotonic_ns())
|
||||
|
||||
async def loop(self):
|
||||
while not self.stopping:
|
||||
@@ -91,13 +129,17 @@ class SourceControl:
|
||||
|
||||
async def wait_anchor(self):
|
||||
deadline = time.monotonic() + 10
|
||||
while self.anchor is None or not self.confirmed:
|
||||
while True:
|
||||
if self.failure:
|
||||
raise RuntimeError("initial source clock handshake failed") from self.failure
|
||||
if time.monotonic() > deadline:
|
||||
raise TimeoutError("bounded initial source clock handshake")
|
||||
if self.anchor is not None:
|
||||
if time.monotonic_ns() >= self.anchor.local_zero_ns - 100_000_000:
|
||||
raise TimeoutError("joint source admission missed immutable start")
|
||||
if self.ready() and self.access is not None:
|
||||
return self.anchor
|
||||
await asyncio.sleep(0.01)
|
||||
return self.anchor
|
||||
|
||||
async def finish(self):
|
||||
self.ending = True
|
||||
|
||||
@@ -276,7 +276,7 @@ def run(args):
|
||||
"costmap_freshness_mode": args.costmap_freshness,
|
||||
"ddrnet_min_source_interval_ms": args.ddrnet_min_source_interval_ms,
|
||||
"input_transport": args.input_transport,
|
||||
"network_clock_mode": "acknowledged-monotonic/v1"
|
||||
"network_clock_mode": "joint-start-monotonic/v2"
|
||||
if getattr(args, "network_cross_host", False)
|
||||
else "local-only",
|
||||
"network_clock_envelope": {
|
||||
|
||||
@@ -145,6 +145,7 @@ class ClockBounds:
|
||||
or type(self.offset_lower_ns) is not int
|
||||
or type(self.offset_upper_ns) is not int
|
||||
or self.offset_lower_ns > self.offset_upper_ns
|
||||
or not -(2**63) < self.offset_lower_ns <= self.offset_upper_ns < 2**63
|
||||
or type(self.rate_ppm) is not int
|
||||
or not 1 <= self.rate_ppm <= 1000
|
||||
or type(self.timestamp_error_ns) is not int
|
||||
@@ -187,6 +188,46 @@ class ClockBounds:
|
||||
"symmetric_network_assumed": False,
|
||||
}
|
||||
|
||||
@classmethod
|
||||
def from_dict(cls, value: object) -> ClockBounds:
|
||||
"""Closed, exact-integer wire contract, including signed clock offsets."""
|
||||
if not isinstance(value, dict):
|
||||
raise ClockMappingError("invalid clock bounds document")
|
||||
|
||||
def integer(key: str) -> int:
|
||||
raw = value.get(key)
|
||||
if not isinstance(raw, str) or not 1 <= len(raw) <= 20:
|
||||
raise ClockMappingError("clock nanoseconds must be canonical decimal strings")
|
||||
try:
|
||||
parsed = int(raw)
|
||||
except ValueError as exc:
|
||||
raise ClockMappingError("invalid clock nanoseconds") from exc
|
||||
if str(parsed) != raw:
|
||||
raise ClockMappingError("noncanonical clock nanoseconds")
|
||||
return parsed
|
||||
|
||||
try:
|
||||
result = cls(
|
||||
value["local_clock_id"],
|
||||
value["remote_clock_id"],
|
||||
integer("measured_at_ns"),
|
||||
integer("expires_at_ns"),
|
||||
integer("offset_lower_ns"),
|
||||
integer("offset_upper_ns"),
|
||||
value["relative_rate_budget_ppm"],
|
||||
integer("timestamp_error_budget_ns"),
|
||||
value["samples"],
|
||||
)
|
||||
except KeyError as exc:
|
||||
raise ClockMappingError("incomplete clock bounds") from exc
|
||||
if (
|
||||
value != result.to_dict()
|
||||
or value.get("conditional_rate_error_envelope") is not True
|
||||
or value.get("symmetric_network_assumed") is not False
|
||||
):
|
||||
raise ClockMappingError("unsupported clock bounds document")
|
||||
return result
|
||||
|
||||
|
||||
class ClockWindow:
|
||||
def __init__(
|
||||
|
||||
@@ -21,9 +21,9 @@ import grpc # type: ignore[import-untyped]
|
||||
|
||||
from . import streaming_wire as wire
|
||||
from .realtime_contract import StreamStart
|
||||
from .streaming_clock import ClockProbe
|
||||
from .streaming_clock import ClockProbe, ClockReceipt
|
||||
from .streaming_grpc import OPTIONS, StreamAccess
|
||||
from .streaming_source_clock import SourceAnchor
|
||||
from .streaming_source_clock import SourceAnchor, SourceClockState
|
||||
|
||||
SERVICE = "missioncore.perception.v1.StreamControl"
|
||||
METHOD = f"/{SERVICE}/Poll"
|
||||
@@ -79,7 +79,8 @@ class StreamControlEndpoint:
|
||||
*,
|
||||
clock_id: str,
|
||||
clock_ns: Callable[[], int] = time.monotonic_ns,
|
||||
observe: Callable[[ClockProbe, int, SourceAnchor, bool], None] | None = None,
|
||||
observe: Callable[[ClockProbe, int, SourceAnchor | None, bool], SourceClockState]
|
||||
| None = None,
|
||||
) -> None:
|
||||
wire.identifier(clock_id)
|
||||
self.activation, self.pending, self.clock_id, self.clock_ns = (
|
||||
@@ -198,7 +199,11 @@ class StreamControlEndpoint:
|
||||
try:
|
||||
self._authenticate(context)
|
||||
value = _document(raw)
|
||||
if set(value) != {"probe", "anchor", "ended"} or self.observe is None:
|
||||
if (
|
||||
set(value) != {"schema_version", "probe", "anchor", "ended"}
|
||||
or value["schema_version"] != "missioncore.stream-clock-report/v2"
|
||||
or self.observe is None
|
||||
):
|
||||
raise ValueError("clock receipt not admitted")
|
||||
evidence = value["probe"]
|
||||
with self._lock:
|
||||
@@ -225,17 +230,18 @@ class StreamControlEndpoint:
|
||||
)
|
||||
),
|
||||
)
|
||||
anchor = SourceAnchor.from_dict(value["anchor"])
|
||||
self.observe(probe, received, anchor, value["ended"])
|
||||
anchor = None if value["anchor"] is None else SourceAnchor.from_dict(value["anchor"])
|
||||
state = self.observe(probe, received, anchor, value["ended"])
|
||||
except (ValueError, TypeError):
|
||||
self.rejected += 1
|
||||
await context.abort(grpc.StatusCode.INVALID_ARGUMENT, "clock receipt rejected")
|
||||
return b""
|
||||
return wire.canonical(
|
||||
{
|
||||
"schema_version": "missioncore.stream-clock-receipt/v1",
|
||||
"schema_version": "missioncore.stream-clock-receipt/v2",
|
||||
"activation_binding": wire.binding(self.activation),
|
||||
"nonce": probe.nonce,
|
||||
"state": state.to_dict(),
|
||||
}
|
||||
)
|
||||
|
||||
@@ -338,8 +344,8 @@ class StreamControlClient:
|
||||
await self.channel.close()
|
||||
|
||||
async def acknowledge(
|
||||
self, probe: ClockProbe, anchor: SourceAnchor, *, ended: bool = False
|
||||
) -> None:
|
||||
self, probe: ClockProbe, anchor: SourceAnchor | None, *, ended: bool = False
|
||||
) -> SourceClockState:
|
||||
if self.polling or probe.local_clock_id != self.clock_id or type(ended) is not bool:
|
||||
raise ValueError("one scoped control request at a time")
|
||||
self.polling = True
|
||||
@@ -365,17 +371,37 @@ class StreamControlClient:
|
||||
)
|
||||
response = _document(
|
||||
await self.report_call(
|
||||
wire.canonical({"probe": evidence, "anchor": anchor.to_dict(), "ended": ended}),
|
||||
wire.canonical(
|
||||
{
|
||||
"schema_version": "missioncore.stream-clock-report/v2",
|
||||
"probe": evidence,
|
||||
"anchor": None if anchor is None else anchor.to_dict(),
|
||||
"ended": ended,
|
||||
}
|
||||
),
|
||||
metadata=self.ticket.metadata(),
|
||||
timeout=0.5,
|
||||
wait_for_ready=False,
|
||||
)
|
||||
)
|
||||
state = SourceClockState.from_dict(response.get("state"))
|
||||
if response != {
|
||||
"schema_version": "missioncore.stream-clock-receipt/v1",
|
||||
"schema_version": "missioncore.stream-clock-receipt/v2",
|
||||
"activation_binding": wire.binding(self.ticket.activation),
|
||||
"nonce": probe.nonce,
|
||||
"state": state.to_dict(),
|
||||
}:
|
||||
raise ValueError("clock receipt response binding mismatch")
|
||||
if (
|
||||
(state.bounds.local_clock_id, state.bounds.remote_clock_id)
|
||||
!= (probe.remote_clock_id, probe.local_clock_id)
|
||||
or state.bounds.rate_ppm != 500
|
||||
or state.bounds.timestamp_error_ns != 50_000
|
||||
or state.ended != ended
|
||||
or (state.anchor is not None and state.anchor != anchor)
|
||||
):
|
||||
raise ValueError("clock receipt state binding mismatch")
|
||||
ClockReceipt(probe, state.bounds.measured_at_ns)
|
||||
return state
|
||||
finally:
|
||||
self.polling = False
|
||||
|
||||
@@ -6,7 +6,14 @@ import threading
|
||||
from dataclasses import dataclass
|
||||
|
||||
from . import streaming_wire as wire
|
||||
from .streaming_clock import ClockMappingError, ClockProbe, ClockReceipt, ClockWindow, _stamp
|
||||
from .streaming_clock import (
|
||||
ClockBounds,
|
||||
ClockMappingError,
|
||||
ClockProbe,
|
||||
ClockReceipt,
|
||||
ClockWindow,
|
||||
_stamp,
|
||||
)
|
||||
from .streaming_continuity import StreamSuspended
|
||||
|
||||
|
||||
@@ -32,6 +39,36 @@ class SourceAnchor:
|
||||
return cls(wire.uint64(value["source_zero_ns"]), wire.uint64(value["local_zero_ns"]))
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class SourceClockState:
|
||||
bounds: ClockBounds
|
||||
anchor: SourceAnchor | None
|
||||
ended: bool
|
||||
|
||||
def __post_init__(self) -> None:
|
||||
if not isinstance(self.bounds, ClockBounds) or type(self.ended) is not bool:
|
||||
raise ValueError("invalid source clock state")
|
||||
if self.anchor is not None and not isinstance(self.anchor, SourceAnchor):
|
||||
raise ValueError("invalid accepted source anchor")
|
||||
|
||||
def to_dict(self) -> dict[str, object]:
|
||||
return {
|
||||
"bounds": self.bounds.to_dict(),
|
||||
"anchor": None if self.anchor is None else self.anchor.to_dict(),
|
||||
"ended": self.ended,
|
||||
}
|
||||
|
||||
@classmethod
|
||||
def from_dict(cls, value: object) -> SourceClockState:
|
||||
if not isinstance(value, dict) or set(value) != {"bounds", "anchor", "ended"}:
|
||||
raise ValueError("invalid source clock state document")
|
||||
return cls(
|
||||
ClockBounds.from_dict(value["bounds"]),
|
||||
None if value["anchor"] is None else SourceAnchor.from_dict(value["anchor"]),
|
||||
value["ended"],
|
||||
)
|
||||
|
||||
|
||||
class SourceClockMonitor:
|
||||
def __init__(self, worker_clock_id: str, source_zero_ns: int) -> None:
|
||||
wire.identifier(worker_clock_id)
|
||||
@@ -44,30 +81,40 @@ class SourceClockMonitor:
|
||||
self._lock = threading.RLock()
|
||||
|
||||
def observe(
|
||||
self, probe: ClockProbe, received_ns: int, anchor: SourceAnchor, ended: bool
|
||||
) -> None:
|
||||
self, probe: ClockProbe, received_ns: int, anchor: SourceAnchor | None, ended: bool
|
||||
) -> SourceClockState:
|
||||
"""Called only after the transport verifies the one-use issued challenge."""
|
||||
receipt = ClockReceipt(probe, received_ns)
|
||||
if probe.remote_clock_id != self.worker_clock_id or type(ended) is not bool:
|
||||
raise ValueError("foreign clock/status")
|
||||
with self._lock:
|
||||
if self.ended or anchor.source_zero_ns != self.source_zero:
|
||||
if self.ended or (anchor is not None and anchor.source_zero_ns != self.source_zero):
|
||||
raise ValueError("source activation ended or source anchor changed")
|
||||
if self.anchor is None:
|
||||
if not 0 <= anchor.local_zero_ns - probe.local_receive_ns <= 10_000_000_000:
|
||||
if (
|
||||
anchor is not None
|
||||
and not 0 <= anchor.local_zero_ns - probe.local_receive_ns <= 10_000_000_000
|
||||
):
|
||||
raise ValueError("initial source start must be bounded and not in the past")
|
||||
self.anchor = anchor
|
||||
elif anchor != self.anchor:
|
||||
raise ValueError("source timeline is immutable for the activation")
|
||||
if self.window is None:
|
||||
self.window = ClockWindow(
|
||||
self.worker_clock_id,
|
||||
probe.local_clock_id,
|
||||
rate_ppm=500,
|
||||
timestamp_error_ns=50_000,
|
||||
)
|
||||
elif anchor != self.anchor:
|
||||
raise ValueError("source timeline is immutable for the activation")
|
||||
assert self.window is not None
|
||||
self.window.add(receipt)
|
||||
bounds = self.window.add(receipt)
|
||||
if self.anchor is None and anchor is not None:
|
||||
try:
|
||||
bounds.require(received_ns)
|
||||
except ClockMappingError:
|
||||
pass # Explicitly unaccepted proposal; keep warming BOTH clock windows.
|
||||
else:
|
||||
self.anchor = anchor
|
||||
self.ended = ended
|
||||
return SourceClockState(bounds, self.anchor, ended)
|
||||
|
||||
def observed(self, now_ns: int) -> tuple[int, int]:
|
||||
"""Source time midpoint + integer half-width, freshly checked at use."""
|
||||
|
||||
@@ -161,68 +161,6 @@ def test_source_eof_ack_precedes_data_channel_retirement(adapter, has_stream):
|
||||
assert calls == (["ack", "end", "drain"] if has_stream else ["ack"])
|
||||
|
||||
|
||||
def test_source_anchor_latches_only_after_initial_clock_gate_and_never_on_recovery(
|
||||
adapter, monkeypatch
|
||||
):
|
||||
module = importlib.import_module("pilot_source_control")
|
||||
healthy, acknowledgements = [False], []
|
||||
|
||||
class Bounds:
|
||||
def require(self, now):
|
||||
if not healthy[0]:
|
||||
raise module.ClockMappingError("synthetic excessive uncertainty")
|
||||
|
||||
def to_dict(self):
|
||||
return {}
|
||||
|
||||
def uncertainty_ns(self, now):
|
||||
return 1_000_000 if healthy[0] else 6_000_000
|
||||
|
||||
class Client:
|
||||
def __init__(self, *args, **kwargs):
|
||||
pass
|
||||
|
||||
async def poll(self):
|
||||
return object(), None
|
||||
|
||||
async def acknowledge(self, probe, anchor, *, ended):
|
||||
acknowledgements.append(anchor)
|
||||
|
||||
monkeypatch.setattr(module, "StreamControlClient", Client)
|
||||
monkeypatch.setattr(
|
||||
module,
|
||||
"ClockWindow",
|
||||
lambda *a, **kw: SimpleNamespace(add=lambda probe: Bounds(), current=lambda now: Bounds()),
|
||||
)
|
||||
control = module.SourceControl(
|
||||
"unused",
|
||||
b"",
|
||||
{
|
||||
"activation": identity().to_dict(),
|
||||
"token": "b" * 64,
|
||||
"remote_clock_id": "worker",
|
||||
"source_zero_ns": str(10**12),
|
||||
},
|
||||
)
|
||||
|
||||
async def check():
|
||||
await control.update()
|
||||
assert control.anchor is None and not control.ready() and not acknowledgements
|
||||
healthy[0] = True
|
||||
await control.update()
|
||||
anchor = await control.wait_anchor()
|
||||
assert anchor == control.anchor and control.ready()
|
||||
healthy[0] = False
|
||||
await control.update()
|
||||
assert not control.ready() and control.anchor == anchor
|
||||
healthy[0] = True
|
||||
await control.update()
|
||||
assert control.ready() and acknowledgements == [anchor] * 3
|
||||
assert [row["acknowledged"] for row in control.samples] == [False, True, True, True]
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
def test_cross_host_bridge_clock_expiry_retains_owner_and_rotates_epoch(tmp_path, tls, adapter): # noqa: F811
|
||||
module, _ = adapter
|
||||
cert, key = tmp_path / "cert", tmp_path / "key"
|
||||
|
||||
@@ -18,6 +18,37 @@ from k1link.perception.streaming_queue import StreamMailbox
|
||||
from k1link.perception.streaming_source_clock import SourceAnchor, SourceClockMonitor
|
||||
|
||||
|
||||
def test_responder_warms_without_source_admission_and_rejects_wide_start():
|
||||
monitor = SourceClockMonitor("worker", 10**12)
|
||||
wide = sample(outbound=20_000_000, inbound=20_000_000)
|
||||
now = wide.local_receive_ns + 10**12 + 1_000_000
|
||||
anchor = SourceAnchor(10**12, wide.local_receive_ns + 1_000_000_000)
|
||||
response = monitor.observe(wide, now, anchor, False)
|
||||
assert response.anchor is monitor.anchor is None and response.bounds.samples == 1
|
||||
with pytest.raises(StreamSuspended):
|
||||
monitor.observed(now)
|
||||
narrow = sample(2)
|
||||
now = narrow.local_receive_ns + 10**12 + 1_000_000
|
||||
response = monitor.observe(narrow, now, None, False)
|
||||
assert response.anchor is None and response.bounds.samples == 2
|
||||
response.bounds.require(now)
|
||||
with pytest.raises(StreamSuspended):
|
||||
monitor.observed(now) # Valid clock is not permission to start without an anchor.
|
||||
third = sample(3)
|
||||
now = third.local_receive_ns + 10**12 + 1_000_000
|
||||
assert monitor.observe(third, now, anchor, False).anchor == anchor
|
||||
assert monitor.observed(now)[1] < 5_000_000
|
||||
|
||||
|
||||
def test_end_before_start_is_terminal_without_inventing_a_source_timeline():
|
||||
monitor = SourceClockMonitor("worker", 10**12)
|
||||
first = sample()
|
||||
state = monitor.observe(first, first.local_receive_ns + 10**12 + 1_000_000, None, True)
|
||||
assert state.ended and state.anchor is None
|
||||
with pytest.raises(StreamSuspended):
|
||||
monitor.observed(state.bounds.measured_at_ns)
|
||||
|
||||
|
||||
def sample(number=1, outbound=1_000_000, inbound=2_000_000, offset=10**12):
|
||||
t1 = 10**16 + number * 100_000_000
|
||||
return ClockProbe(
|
||||
|
||||
@@ -0,0 +1,231 @@
|
||||
"""Joint pre-start admission, using real interval windows with synthetic time."""
|
||||
|
||||
# ruff: noqa: F811 -- shared probe path fixture
|
||||
import asyncio
|
||||
import importlib
|
||||
from dataclasses import replace
|
||||
from types import SimpleNamespace
|
||||
|
||||
import pytest
|
||||
from test_perception_network_graph_adapter import adapter # noqa: F401
|
||||
from test_perception_streaming_control_grpc import controlled
|
||||
from test_perception_streaming_grpc import identity, tls # noqa: F401
|
||||
|
||||
from k1link.perception.streaming_clock import ClockProbe
|
||||
from k1link.perception.streaming_source_clock import SourceClockMonitor
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
def joint(adapter, monkeypatch):
|
||||
module = importlib.import_module("pilot_source_control")
|
||||
clock = [10**16]
|
||||
monitor = SourceClockMonitor("worker", 10**12)
|
||||
state = SimpleNamespace(lost_ack=False, wide_peer=False, grant=True, proposals=[])
|
||||
|
||||
class Client:
|
||||
def __init__(self, *args, clock_id):
|
||||
self.clock_id = clock_id
|
||||
|
||||
async def poll(self):
|
||||
clock[0] += 100_000_000
|
||||
t1 = clock[0]
|
||||
probe = ClockProbe(
|
||||
self.clock_id,
|
||||
"worker",
|
||||
str(t1),
|
||||
t1,
|
||||
t1 + 10**12 + 1_000_000,
|
||||
t1 + 10**12 + 1_010_000,
|
||||
t1 + 3_010_000,
|
||||
)
|
||||
clock[0] = probe.local_receive_ns
|
||||
return probe, object() if monitor.anchor and state.grant else None
|
||||
|
||||
async def acknowledge(self, probe, anchor, *, ended):
|
||||
state.proposals.append(anchor)
|
||||
peer = monitor.observe(probe, clock[0] + 10**12 + 1_000_000, anchor, ended)
|
||||
clock[0] += 2_000_000
|
||||
if state.lost_ack:
|
||||
state.lost_ack = False
|
||||
raise TimeoutError("lost ACK after Worker accepted anchor")
|
||||
if state.wide_peer:
|
||||
peer = replace(
|
||||
peer,
|
||||
bounds=replace(
|
||||
peer.bounds,
|
||||
offset_lower_ns=-(10**12) - 6_000_000,
|
||||
offset_upper_ns=-(10**12) + 6_000_000,
|
||||
),
|
||||
)
|
||||
return peer
|
||||
|
||||
monkeypatch.setattr(
|
||||
module,
|
||||
"time",
|
||||
SimpleNamespace(monotonic_ns=lambda: clock[0], monotonic=lambda: clock[0] / 1e9),
|
||||
)
|
||||
monkeypatch.setattr(module, "StreamControlClient", Client)
|
||||
control = module.SourceControl(
|
||||
"unused",
|
||||
b"",
|
||||
{
|
||||
"activation": identity().to_dict(),
|
||||
"token": "b" * 64,
|
||||
"remote_clock_id": "worker",
|
||||
"source_zero_ns": str(10**12),
|
||||
},
|
||||
)
|
||||
return control, monitor, state, clock
|
||||
|
||||
|
||||
def test_both_windows_warm_before_anchor_and_grant_and_never_reanchor(joint):
|
||||
control, monitor, state, clock = joint
|
||||
|
||||
async def check():
|
||||
await control.update()
|
||||
assert control.anchor is monitor.anchor is None
|
||||
assert len(monitor.window.samples) == 1 and state.proposals == [None]
|
||||
assert not control.ready()
|
||||
await control.update()
|
||||
anchor = control.anchor
|
||||
assert anchor is not None and anchor == monitor.anchor and control.access is None
|
||||
await control.update()
|
||||
assert await control.wait_anchor() == anchor and control.ready()
|
||||
state.wide_peer = True
|
||||
await control.update()
|
||||
assert not control.ready() and control.anchor == anchor
|
||||
state.wide_peer = False
|
||||
await control.update()
|
||||
assert control.ready() and control.anchor == anchor
|
||||
assert all(p == anchor for p in state.proposals[1:])
|
||||
assert all(row["acknowledged"] for row in control.samples)
|
||||
row = control.samples[-1]
|
||||
assert (
|
||||
int(row["t1_source_send_ns"])
|
||||
< int(row["t4_source_receive_ns"])
|
||||
< int(row["t6_source_ack_receive_ns"])
|
||||
)
|
||||
assert int(row["t3_worker_send_ns"]) < int(row["t5_worker_report_receive_ns"])
|
||||
clock[0] += 2_000_000_000
|
||||
assert not control.ready() and control.anchor == anchor
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
def test_one_sided_readiness_never_proposes_start(joint):
|
||||
control, monitor, state, _ = joint
|
||||
|
||||
async def check():
|
||||
state.wide_peer = True
|
||||
for _ in range(4):
|
||||
await control.update()
|
||||
assert control.anchor is monitor.anchor is None
|
||||
assert state.proposals == [None] * 4
|
||||
state.wide_peer = False
|
||||
await control.update()
|
||||
assert control.anchor is None # Fresh peer evidence arrives in this ACK.
|
||||
await control.update()
|
||||
assert control.anchor == monitor.anchor and control.ready()
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
def test_lost_acceptance_echo_keeps_exact_same_proposal(joint):
|
||||
control, monitor, state, _ = joint
|
||||
|
||||
async def check():
|
||||
await control.update()
|
||||
state.lost_ack = True
|
||||
with pytest.raises(TimeoutError):
|
||||
await control.update()
|
||||
proposal = control.proposal
|
||||
assert proposal is not None and control.anchor is None and monitor.anchor == proposal
|
||||
await control.update()
|
||||
assert await control.wait_anchor() == proposal
|
||||
assert state.proposals[-2:] == [proposal, proposal]
|
||||
assert not control.samples[-2]["acknowledged"]
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
def test_missing_grant_cannot_silently_start_late_or_reanchor(joint):
|
||||
control, _, state, clock = joint
|
||||
|
||||
async def check():
|
||||
state.grant = False
|
||||
await control.update()
|
||||
await control.update()
|
||||
anchor = control.anchor
|
||||
waiting = asyncio.create_task(control.wait_anchor())
|
||||
await asyncio.sleep(0)
|
||||
assert not waiting.done()
|
||||
clock[0] = anchor.local_zero_ns
|
||||
with pytest.raises(TimeoutError, match="immutable start"):
|
||||
await waiting
|
||||
assert control.anchor == anchor and control.access is None
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
def test_peer_ready_does_not_replace_local_gate_or_refresh_peer_expiry(joint, monkeypatch):
|
||||
control, _, _, clock = joint
|
||||
|
||||
async def check():
|
||||
await control.update()
|
||||
await control.update()
|
||||
assert control.ready()
|
||||
original = control.window.current
|
||||
|
||||
def wide(now):
|
||||
return replace(
|
||||
original(now),
|
||||
offset_lower_ns=10**12 - 6_000_000,
|
||||
offset_upper_ns=10**12 + 6_000_000,
|
||||
)
|
||||
|
||||
monkeypatch.setattr(control.window, "current", wide)
|
||||
assert not control.ready()
|
||||
monkeypatch.setattr(control.window, "current", original)
|
||||
assert control.ready()
|
||||
old_peer = control.peer
|
||||
clock[0] += 2_000_000_000
|
||||
await control.update()
|
||||
assert control.ready()
|
||||
control.peer = old_peer
|
||||
assert not control.ready() # Fresh Mac probe cannot revive stale Worker evidence.
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
def test_real_tls_joint_handshake_waits_for_worker_anchor_and_data_grant(tmp_path, tls, adapter):
|
||||
module = importlib.import_module("pilot_source_control")
|
||||
|
||||
async def check():
|
||||
async with controlled(tmp_path, tls) as (run, endpoint, server, client, pending, port):
|
||||
monitor = SourceClockMonitor("worker", 10**12)
|
||||
server.observe = monitor.observe
|
||||
server.pending = lambda: pending[0] if monitor.anchor is not None else None
|
||||
control = module.SourceControl(
|
||||
f"localhost:{port}",
|
||||
tls[0],
|
||||
{
|
||||
"activation": identity().to_dict(),
|
||||
"token": client.ticket.token,
|
||||
"remote_clock_id": "worker",
|
||||
"source_zero_ns": str(10**12),
|
||||
},
|
||||
)
|
||||
control.start()
|
||||
try:
|
||||
anchor = await control.wait_anchor()
|
||||
assert anchor == monitor.anchor
|
||||
assert control.access == pending[0] and control.ready()
|
||||
assert control.samples[0]["peer"]["anchor"] is None
|
||||
assert len(control.samples) >= 3
|
||||
assert endpoint.active is None and run.lease.start == identity()
|
||||
await control.finish()
|
||||
assert monitor.ended
|
||||
finally:
|
||||
await control.close()
|
||||
|
||||
asyncio.run(check())
|
||||
@@ -4,11 +4,62 @@ from dataclasses import replace
|
||||
|
||||
import pytest
|
||||
|
||||
from k1link.perception.streaming_clock import ClockMappingError, ClockProbe, ClockWindow
|
||||
from k1link.perception.streaming_clock import (
|
||||
ClockBounds,
|
||||
ClockMappingError,
|
||||
ClockProbe,
|
||||
ClockWindow,
|
||||
)
|
||||
|
||||
BASE = 10**16 # Deliberately above IEEE754 exact-integer range.
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"fault",
|
||||
[
|
||||
None,
|
||||
"float",
|
||||
"integer",
|
||||
"plus",
|
||||
"leading",
|
||||
"negative_zero",
|
||||
"extra",
|
||||
"missing",
|
||||
"flags",
|
||||
"wide",
|
||||
"schema",
|
||||
],
|
||||
)
|
||||
def test_bounds_wire_exact_closed_and_signed(fault):
|
||||
bounds = window().add(probe())
|
||||
document = bounds.to_dict()
|
||||
if fault is None:
|
||||
assert ClockBounds.from_dict(document) == bounds
|
||||
return
|
||||
if fault == "float":
|
||||
document["measured_at_ns"] = float(bounds.measured_at_ns)
|
||||
elif fault == "integer":
|
||||
document["measured_at_ns"] = bounds.measured_at_ns
|
||||
elif fault == "plus":
|
||||
document["offset_upper_ns"] = "+1"
|
||||
elif fault == "leading":
|
||||
document["measured_at_ns"] = "0" + document["measured_at_ns"]
|
||||
elif fault == "negative_zero":
|
||||
document["offset_upper_ns"] = "-0"
|
||||
elif fault == "extra":
|
||||
document["ready"] = True
|
||||
elif fault == "missing":
|
||||
del document["samples"]
|
||||
elif fault == "flags":
|
||||
document["conditional_rate_error_envelope"] = 1
|
||||
elif fault == "wide":
|
||||
document["offset_upper_ns"] = str(2**63)
|
||||
else:
|
||||
document["schema_version"] = "old"
|
||||
with pytest.raises(ClockMappingError):
|
||||
ClockBounds.from_dict(document)
|
||||
|
||||
|
||||
def probe(number=1, *, outbound=1_000_000, inbound=2_000_000, offset=-(10**12)):
|
||||
sent = BASE + number * 100_000_000
|
||||
return ClockProbe(
|
||||
|
||||
@@ -254,3 +254,39 @@ def test_wrong_clock_receipt_cannot_update_mapping(tmp_path, tls, fault):
|
||||
assert monitor.anchor is None and monitor.window is None
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
|
||||
@pytest.mark.parametrize("fault", ["v1", "nonce", "clock", "anchor", "time", "envelope", "end"])
|
||||
def test_receipt_v2_client_rejects_unbound_or_legacy_state(tmp_path, tls, fault):
|
||||
async def check():
|
||||
async with controlled(tmp_path, tls) as (_, _, control, client, _, _):
|
||||
control.observe = SourceClockMonitor("worker", 10**12).observe
|
||||
probe, _ = await client.poll()
|
||||
original = client.report_call
|
||||
|
||||
async def altered(raw, **kwargs):
|
||||
value = wire.parse_header(bytearray(await original(raw, **kwargs)))
|
||||
if fault == "v1":
|
||||
value["schema_version"] = "missioncore.stream-clock-receipt/v1"
|
||||
elif fault == "nonce":
|
||||
value["nonce"] = "foreign"
|
||||
elif fault == "clock":
|
||||
value["state"]["bounds"]["local_clock_id"] = "foreign"
|
||||
elif fault == "anchor":
|
||||
value["state"]["anchor"] = SourceAnchor(
|
||||
10**12, probe.local_receive_ns
|
||||
).to_dict()
|
||||
elif fault == "time":
|
||||
value["state"]["bounds"]["measured_at_ns"] = str(probe.remote_send_ns - 1)
|
||||
elif fault == "envelope":
|
||||
value["state"]["bounds"]["relative_rate_budget_ppm"] = 1
|
||||
else:
|
||||
value["state"]["ended"] = True
|
||||
return wire.canonical(value)
|
||||
|
||||
client.report_call = altered
|
||||
with pytest.raises(ValueError):
|
||||
await client.acknowledge(probe, None)
|
||||
assert not client.polling
|
||||
|
||||
asyncio.run(check())
|
||||
|
||||
Reference in New Issue
Block a user