feat(correlation,profiler): publish attacker.observed on first sighting (DEBT-031 worker 3)

CorrelationEngine gains an optional publish_fn hook fired once per unique
attacker IP.  The profiler worker — sole caller of the engine today —
carries the bus physically, builds a thread-safe publisher, and wraps it
with the attacker.observed topic before handing it in.

Bus stays optional: if get_bus() fails or DECNET_BUS_ENABLED=false, the
engine runs publish_fn=None and the worker degrades to DB-only.  Hook
failures log a warning and never break ingestion.
This commit is contained in:
2026-04-21 16:53:03 -04:00
parent 34d9e37ab0
commit e51b65d7c3
4 changed files with 224 additions and 8 deletions

View File

@@ -24,6 +24,7 @@ from __future__ import annotations
from collections import defaultdict
from pathlib import Path
from typing import Any, Callable
from rich.table import Table
@@ -33,17 +34,35 @@ from decnet.logging.syslog_formatter import (
SEVERITY_WARNING,
format_rfc5424,
)
from decnet.logging import get_logger
from decnet.telemetry import traced as _traced, get_tracer as _get_tracer
log = get_logger("correlation.engine")
# ``publish_fn(event_type, payload_dict)``. Sync to avoid rippling
# ``async`` through every call site of :meth:`CorrelationEngine.ingest`;
# the caller wraps bus-publish via
# :func:`decnet.bus.publish.make_thread_safe_publisher`, which is safe to
# invoke from any thread including the event-loop thread.
CorrelationPublishFn = Callable[[str, dict[str, Any]], None]
class CorrelationEngine:
def __init__(self) -> None:
def __init__(
self,
*,
publish_fn: CorrelationPublishFn | None = None,
) -> None:
# attacker_ip → chronological list of events (only events with an IP)
self._events: dict[str, list[LogEvent]] = defaultdict(list)
# Total lines parsed (including no-IP and non-DECNET lines)
self.lines_parsed: int = 0
# Total events indexed (had an attacker_ip)
self.events_indexed: int = 0
# Optional bus hook — invoked on first-sighting of an attacker IP.
# Always fires exactly once per IP for the lifetime of the engine.
self._publish_fn = publish_fn
# ------------------------------------------------------------------ #
# Ingestion #
@@ -61,8 +80,23 @@ class CorrelationEngine:
if event is None:
return None
if event.attacker_ip:
first_sighting = event.attacker_ip not in self._events
self._events[event.attacker_ip].append(event)
self.events_indexed += 1
if first_sighting and self._publish_fn is not None:
try:
self._publish_fn(
"observed",
{
"attacker_ip": event.attacker_ip,
"decky": event.decky,
"service": event.service,
"event_type": event.event_type,
"first_seen": event.timestamp.isoformat(),
},
)
except Exception as exc:
log.warning("correlation publish hook failed: %s", exc)
return event
@_traced("correlation.ingest_file")

View File

@@ -13,11 +13,15 @@ Complexity per cycle: O(new_logs + affected_ips) instead of O(total_logs²).
from __future__ import annotations
import asyncio
import contextlib
import json
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any
from decnet.bus import topics as _topics
from decnet.bus.factory import get_bus
from decnet.bus.publish import make_thread_safe_publisher
from decnet.correlation.engine import CorrelationEngine
from decnet.correlation.parser import LogEvent
from decnet.logging import get_logger
@@ -50,18 +54,44 @@ class _WorkerState:
async def attacker_profile_worker(repo: BaseRepository, *, interval: int = 30) -> None:
"""Periodically updates the Attacker table incrementally. Designed to run as an asyncio Task."""
logger.info("attacker profile worker started interval=%ds", interval)
state = _WorkerState()
# Optional bus wiring — correlator-family publishes ride on the profiler
# worker because CorrelationEngine lives inside it. If the bus is off or
# unreachable the engine runs with publish_fn=None and downstream degrades
# to DB-only.
bus = None
try:
bus = get_bus(client_name="profiler")
await bus.connect()
except Exception as exc:
logger.warning("profiler: bus unavailable, continuing without publish: %s", exc)
bus = None
loop = asyncio.get_running_loop()
raw_publish = make_thread_safe_publisher(bus, loop) if bus is not None else None
def _publish_attacker(event_type: str, payload: dict[str, Any]) -> None:
if raw_publish is None:
return
raw_publish(_topics.attacker(event_type), payload, event_type)
state = _WorkerState(engine=CorrelationEngine(publish_fn=_publish_attacker))
_saved_cursor = await repo.get_state(_STATE_KEY)
if _saved_cursor:
state.last_log_id = _saved_cursor.get("last_log_id", 0)
state.initialized = True
logger.info("attacker worker: resumed from cursor last_log_id=%d", state.last_log_id)
while True:
await asyncio.sleep(interval)
try:
await _incremental_update(repo, state)
except Exception as exc:
logger.error("attacker worker: update failed: %s", exc)
try:
while True:
await asyncio.sleep(interval)
try:
await _incremental_update(repo, state)
except Exception as exc:
logger.error("attacker worker: update failed: %s", exc)
finally:
if bus is not None:
with contextlib.suppress(Exception):
await bus.close()
@_traced("profiler.incremental_update")