feat(sniffer): publish decky.{id}.traffic on the bus (DEBT-031)
SnifferEngine gains an optional publish_fn hook, invoked after the dedup + syslog write for traffic-summary events only (tls_session, tcp_flow_timing, tcp_syn_fingerprint) — intermediate parser artifacts like tls_client_hello stay off the bus. The sniffer worker wires get_bus() + a thread-safe shim that marshals sync calls from the scapy sniff thread back onto the asyncio loop via run_coroutine_threadsafe. Bus failure at startup degrades cleanly to publish-off mode; publish failures at runtime never escape the sniff thread.
This commit is contained in:
@@ -52,6 +52,17 @@ _TCP_ACK: int = 0x10
|
||||
_TCP_FIN: int = 0x01
|
||||
_TCP_RST: int = 0x04
|
||||
|
||||
# Event types that should fan out on the service bus as ``decky.{id}.traffic``.
|
||||
# Intermediate parser artifacts (tls_client_hello, tls_certificate) are
|
||||
# intentionally excluded — tls_session covers the completed handshake and
|
||||
# tcp_flow_timing covers the flow summary; together they're the minimum
|
||||
# interesting signal for downstream consumers.
|
||||
_BUS_TRAFFIC_EVENTS: frozenset[str] = frozenset({
|
||||
"tls_session",
|
||||
"tcp_flow_timing",
|
||||
"tcp_syn_fingerprint",
|
||||
})
|
||||
|
||||
|
||||
# ─── TCP option extraction for passive fingerprinting ───────────────────────
|
||||
|
||||
@@ -692,10 +703,16 @@ class SnifferEngine:
|
||||
ip_to_decky: dict[str, str],
|
||||
write_fn: Callable[[str], None],
|
||||
dedup_ttl: float = 300.0,
|
||||
publish_fn: Callable[[str, str, dict[str, Any]], None] | None = None,
|
||||
):
|
||||
self._ip_to_decky = ip_to_decky
|
||||
self._write_fn = write_fn
|
||||
self._dedup_ttl = dedup_ttl
|
||||
# Optional bus publish hook. Called *after* dedup + syslog write, so
|
||||
# every syslog line we emit has a matching bus event and duplicate
|
||||
# storms are already suppressed upstream. Signature:
|
||||
# ``publish_fn(decky_name, event_type, payload_dict)``.
|
||||
self._publish_fn = publish_fn
|
||||
|
||||
self._sessions: dict[tuple[str, int, str, int], dict[str, Any]] = {}
|
||||
self._session_ts: dict[tuple[str, int, str, int], float] = {}
|
||||
@@ -782,6 +799,15 @@ class SnifferEngine:
|
||||
return
|
||||
line = syslog_line(SERVICE_NAME, node_name, event_type, severity=severity, **fields)
|
||||
self._write_fn(line)
|
||||
# Bus fan-out, fire-and-forget. Only emit for traffic-summary event
|
||||
# types — the ones that represent an observable decky interaction
|
||||
# rather than an intermediate parser artifact. Rate is naturally
|
||||
# bounded by the dedup cache above.
|
||||
if self._publish_fn is not None and event_type in _BUS_TRAFFIC_EVENTS:
|
||||
try:
|
||||
self._publish_fn(node_name, event_type, dict(fields))
|
||||
except Exception: # nosec B110 — bus must never break sniff thread
|
||||
pass
|
||||
|
||||
# ── Flow tracking (per-TCP-4-tuple timing + retransmits) ────────────────
|
||||
|
||||
|
||||
@@ -11,12 +11,18 @@ The API never depends on this worker being alive.
|
||||
"""
|
||||
|
||||
import asyncio
|
||||
import contextlib
|
||||
import os
|
||||
import subprocess # nosec B404 — needed for interface checks
|
||||
import threading
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
from pathlib import Path
|
||||
from typing import Any, Callable
|
||||
|
||||
from decnet.bus import topics as _topics
|
||||
from decnet.bus.base import BaseBus
|
||||
from decnet.bus.factory import get_bus
|
||||
from decnet.bus.publish import publish_safely
|
||||
from decnet.logging import get_logger
|
||||
from decnet.network import HOST_IPVLAN_IFACE, HOST_MACVLAN_IFACE
|
||||
from decnet.sniffer.fingerprint import SnifferEngine
|
||||
@@ -41,6 +47,30 @@ def _load_ip_to_decky() -> dict[str, str]:
|
||||
return mapping
|
||||
|
||||
|
||||
def _make_thread_safe_publisher(
|
||||
bus: BaseBus,
|
||||
loop: asyncio.AbstractEventLoop,
|
||||
) -> Callable[[str, str, dict[str, Any]], None]:
|
||||
"""Build a sync callable that marshals bus publishes back to *loop*.
|
||||
|
||||
The scapy sniff loop runs in a dedicated worker thread and cannot
|
||||
``await`` anything. Every call here schedules the async publish on
|
||||
the event loop and returns immediately; the sniff thread is never
|
||||
blocked waiting for the publish to actually land on the wire.
|
||||
"""
|
||||
def _publish(decky_name: str, event_type: str, payload: dict[str, Any]) -> None:
|
||||
topic = _topics.decky(decky_name, _topics.DECKY_TRAFFIC)
|
||||
try:
|
||||
asyncio.run_coroutine_threadsafe(
|
||||
publish_safely(bus, topic, payload, event_type=event_type),
|
||||
loop,
|
||||
)
|
||||
except Exception as exc: # noqa: BLE001
|
||||
logger.debug("sniffer: cross-thread bus publish failed: %s", exc)
|
||||
|
||||
return _publish
|
||||
|
||||
|
||||
def _interface_exists(iface: str) -> bool:
|
||||
"""Check if a network interface exists on this host."""
|
||||
try:
|
||||
@@ -59,6 +89,7 @@ def _sniff_loop(
|
||||
log_path: Path,
|
||||
json_path: Path,
|
||||
stop_event: threading.Event,
|
||||
publish_fn: Callable[[str, str, dict[str, Any]], None] | None = None,
|
||||
) -> None:
|
||||
"""Blocking sniff loop. Runs in a dedicated thread via asyncio.to_thread."""
|
||||
try:
|
||||
@@ -75,7 +106,9 @@ def _sniff_loop(
|
||||
def _write_fn(line: str) -> None:
|
||||
write_event(line, log_path, json_path)
|
||||
|
||||
engine = SnifferEngine(ip_to_decky=ip_map, write_fn=_write_fn)
|
||||
engine = SnifferEngine(
|
||||
ip_to_decky=ip_map, write_fn=_write_fn, publish_fn=publish_fn,
|
||||
)
|
||||
|
||||
# Periodically refresh IP map in a background daemon thread
|
||||
def _refresh_loop() -> None:
|
||||
@@ -150,6 +183,25 @@ async def sniffer_worker(log_file: str) -> None:
|
||||
|
||||
stop_event = threading.Event()
|
||||
|
||||
loop = asyncio.get_running_loop()
|
||||
|
||||
# Connect to the bus for decky.{id}.traffic fan-out. Failure here
|
||||
# is non-fatal: the sniffer still writes syslog, it just doesn't
|
||||
# push notifications to downstream consumers.
|
||||
bus: BaseBus | None = None
|
||||
try:
|
||||
candidate = get_bus(client_name="sniffer")
|
||||
await candidate.connect()
|
||||
bus = candidate
|
||||
except Exception as exc: # noqa: BLE001
|
||||
logger.warning(
|
||||
"sniffer: bus unavailable, running in publish-off mode: %s", exc,
|
||||
)
|
||||
|
||||
publish_fn: Callable[[str, str, dict[str, Any]], None] | None = None
|
||||
if bus is not None:
|
||||
publish_fn = _make_thread_safe_publisher(bus, loop)
|
||||
|
||||
# Dedicated thread pool so the long-running sniff loop doesn't
|
||||
# occupy a slot in the default asyncio executor.
|
||||
sniffer_pool = ThreadPoolExecutor(
|
||||
@@ -157,10 +209,9 @@ async def sniffer_worker(log_file: str) -> None:
|
||||
)
|
||||
|
||||
try:
|
||||
loop = asyncio.get_running_loop()
|
||||
await loop.run_in_executor(
|
||||
sniffer_pool, _sniff_loop,
|
||||
interface, log_path, json_path, stop_event,
|
||||
interface, log_path, json_path, stop_event, publish_fn,
|
||||
)
|
||||
except asyncio.CancelledError:
|
||||
logger.info("sniffer: shutdown requested")
|
||||
@@ -169,6 +220,9 @@ async def sniffer_worker(log_file: str) -> None:
|
||||
raise
|
||||
finally:
|
||||
sniffer_pool.shutdown(wait=False)
|
||||
if bus is not None:
|
||||
with contextlib.suppress(Exception):
|
||||
await bus.close()
|
||||
|
||||
except asyncio.CancelledError:
|
||||
raise
|
||||
|
||||
Reference in New Issue
Block a user