Files
2026-08-17 10:07:19 +02:00

371 lines
15 KiB
Python

from __future__ import annotations
import threading
import time
from datetime import datetime, timezone
from typing import Any, Iterable
from .live import RedisUnavailableError, TrafficHistory
from .store import AlertStore
SUMMARY_WINDOWS = (900, 3600, 18000, 21600, 86400)
class AnalyticsSnapshotCache:
"""Disk-backed dashboard snapshots plus Redis-to-SQLite archive worker.
Redis is intentionally only a short-lived ingestion buffer. This worker moves
normalized EVE records and TZSP rate samples to SQLite, deletes committed
entries from Redis, and calculates dashboard snapshots from SQLite instead of
scanning multi-hour Redis sorted sets.
"""
def __init__(
self,
store: AlertStore,
history: TrafficHistory,
stop_event: threading.Event,
interval_seconds: int = 60,
*,
archive_interval_seconds: int = 5,
archive_lag_seconds: int = 10,
archive_batch_size: int = 1000,
) -> None:
self.store = store
self.history = history
self.stop_event = stop_event
self.interval_seconds = max(15, int(interval_seconds))
self.archive_interval_seconds = max(1, int(archive_interval_seconds))
self.archive_lag_seconds = max(2, int(archive_lag_seconds))
self.archive_batch_size = max(100, min(int(archive_batch_size), 5000))
# Archiving and analytics are intentionally separate. A long 24h SQL
# aggregation must never pause Redis draining and allow the live buffer
# to grow back toward its memory ceiling.
self._archive_thread = threading.Thread(
target=self._run_archive,
name="traffic-archive",
daemon=True,
)
self._snapshot_thread = threading.Thread(
target=self._run_snapshots,
name="analytics-snapshot",
daemon=True,
)
self._archive_wake = threading.Event()
self._snapshot_wake = threading.Event()
self._archive_ready = threading.Event()
self._archive_started = threading.Event()
self._stopping = threading.Event()
self._lock = threading.RLock()
now = time.monotonic()
# Standard UI ranges are kept warm. Arbitrary API ranges are added on demand.
self._requested_at: dict[int, float] = {}
self._last_refresh: dict[int, float] = {}
self._errors = 0
self._refreshes = 0
self._archive_runs = 0
self._archive_moved_events = 0
self._archive_moved_throughput = 0
self._last_archive_run = 0.0
self._last_purge = 0.0
self._active_ttl_seconds = max(120, self.interval_seconds * 3)
def start(self) -> None:
self._stopping.clear()
if not self._archive_thread.is_alive():
self._archive_thread.start()
if not self._snapshot_thread.is_alive():
self._snapshot_thread.start()
self._archive_wake.set()
self._snapshot_wake.set()
def stop(self, timeout: float = 2.0) -> None:
self._stopping.set()
self._archive_wake.set()
self._snapshot_wake.set()
if self._archive_thread.is_alive():
self._archive_thread.join(timeout=timeout)
if self._snapshot_thread.is_alive():
self._snapshot_thread.join(timeout=timeout)
def get(self, window_seconds: int) -> dict[str, Any]:
window = self._normalise_window(window_seconds)
now = time.monotonic()
with self._lock:
self._requested_at[window] = now
cached = self.store.traffic_snapshot(window)
cadence = self._refresh_interval(window)
if cached is not None:
cached = self._decorate(cached, "sqlite-snapshot")
age = float(cached.get("snapshot_age_seconds") or 0)
stale = age > cadence * 1.5
cached["snapshot_stale"] = stale
cached["snapshot_refreshing"] = stale
cached["snapshot_refresh_interval_seconds"] = cadence
self._overlay_current_throughput(cached)
if stale:
self._snapshot_wake.set()
return cached
# The first request for an arbitrary range is calculated by the worker.
self._snapshot_wake.set()
now_ms = int(time.time() * 1000)
bins_count = 60
bin_ms = max(1000, int(window * 1000 / bins_count))
return {
"window_seconds": window,
"events": 0,
"bytes": 0,
"alerts": 0,
"blocked": 0,
"timeline": [
{
"ts_ms": now_ms - window * 1000 + idx * bin_ms,
"events": 0,
"bytes": 0,
"alerts": 0,
"bps": 0,
"in_bps": 0,
"out_bps": 0,
"other_bps": 0,
"pps": 0,
}
for idx in range(bins_count)
],
"snapshot_source": "sqlite-background",
"snapshot_age_seconds": 0,
"snapshot_loading": True,
"snapshot_refreshing": True,
"snapshot_refresh_interval_seconds": cadence,
}
def refresh_all(self) -> None:
self.refresh_windows(SUMMARY_WINDOWS)
def refresh_windows(self, windows: Iterable[int]) -> None:
normalized = sorted({self._normalise_window(window) for window in windows})
if not normalized:
return
try:
snapshots = self.history.analytics_many(normalized)
archive_backlog = not self._archive_ready.is_set()
now_wall = time.time()
for window in normalized:
snapshots[window]["archive_backlog"] = archive_backlog
if archive_backlog:
snapshots[window]["analytics_complete"] = False
self.store.save_traffic_snapshot(window, snapshots[window])
with self._lock:
self._last_refresh[window] = now_wall
self._refreshes += 1
except (RedisUnavailableError, KeyError, ValueError, OSError):
with self._lock:
self._errors += 1
except Exception:
with self._lock:
self._errors += 1
def status(self) -> dict[str, Any]:
persisted = self.store.traffic_snapshot_status().get("windows", [])
now = time.monotonic()
with self._lock:
active = [
window for window, requested in self._requested_at.items()
if window in SUMMARY_WINDOWS or now - requested <= self._active_ttl_seconds
]
refreshes = self._refreshes
errors = self._errors
archive_runs = self._archive_runs
moved_events = self._archive_moved_events
moved_throughput = self._archive_moved_throughput
last_archive_run = self._last_archive_run
return {
"backend": "sqlite-archive",
"archive_worker_running": self._archive_thread.is_alive(),
"snapshot_worker_running": self._snapshot_thread.is_alive(),
"archive_started": self._archive_started.is_set(),
"archive_backlog": not self._archive_ready.is_set(),
"interval_seconds": self.interval_seconds,
"archive_interval_seconds": self.archive_interval_seconds,
"archive_lag_seconds": self.archive_lag_seconds,
"windows": list(SUMMARY_WINDOWS),
"persisted": persisted,
"active_windows": sorted(active),
"cadence_seconds": {str(window): self._refresh_interval(window) for window in sorted(set(active) | set(SUMMARY_WINDOWS))},
"refreshes": refreshes,
"errors": errors,
"archive_runs": archive_runs,
"archive_moved_events": moved_events,
"archive_moved_throughput": moved_throughput,
"last_archive_run": last_archive_run,
}
def _run_archive(self) -> None:
while not self.stop_event.is_set() and not self._stopping.is_set():
self._archive_wake.wait(timeout=float(self.archive_interval_seconds))
self._archive_wake.clear()
if self.stop_event.is_set() or self._stopping.is_set():
break
backlog = self._archive_once()
self._archive_started.set()
if backlog:
self._archive_ready.clear()
elif not self._archive_ready.is_set():
self._archive_ready.set()
self._snapshot_wake.set()
now = time.monotonic()
if now - self._last_purge >= 900:
try:
self.history.purge_archive()
except Exception:
with self._lock:
self._errors += 1
self._last_purge = now
if backlog:
# Continue draining old installations immediately instead of
# sleeping while Redis still holds a large backlog. Yield a
# little CPU so EVE/NDR/web threads remain responsive.
if self.stop_event.wait(0.02):
break
self._archive_wake.set()
def _run_snapshots(self) -> None:
poll_seconds = float(min(max(self.interval_seconds, 5), 15))
while not self.stop_event.is_set() and not self._stopping.is_set():
self._snapshot_wake.wait(timeout=poll_seconds)
self._snapshot_wake.clear()
if self.stop_event.is_set() or self._stopping.is_set():
break
# Do not race the very first archive pass. After that, snapshots may
# use the committed SQLite prefix even while Redis still has a
# backlog. This keeps historical charts moving instead of freezing
# for hours; such snapshots are explicitly marked incomplete.
if not self._archive_started.is_set():
continue
try:
due = self._due_windows()
if due:
self.refresh_windows(due)
except Exception:
with self._lock:
self._errors += 1
def _archive_once(self) -> bool:
cutoff_ms = int((time.time() - self.archive_lag_seconds) * 1000)
try:
result = self.history.archive_redis_to_store(
cutoff_ms,
batch_size=self.archive_batch_size,
max_batches=100,
)
except RedisUnavailableError:
with self._lock:
self._errors += 1
return False
except Exception:
with self._lock:
self._errors += 1
return False
with self._lock:
self._archive_runs += 1
self._archive_moved_events += int(result.get("events") or 0)
self._archive_moved_throughput += int(result.get("throughput_samples") or 0)
self._last_archive_run = time.time()
return int(result.get("batches") or 0) >= 100
def _due_windows(self) -> list[int]:
now_mono = time.monotonic()
now_wall = time.time()
due: list[int] = []
with self._lock:
requests = dict(self._requested_at)
last_refresh = dict(self._last_refresh)
# When nobody requested analytics, do not touch SQLite just to keep
# dashboard ranges warm. Current throughput is delivered independently
# from the Rust telemetry path, so background chart work can stay idle.
if not requests:
return []
persisted = {
int(row["window_seconds"]): row.get("generated_at")
for row in self.store.traffic_snapshot_status().get("windows", [])
}
for window, requested_at in requests.items():
if window not in SUMMARY_WINDOWS and now_mono - requested_at > self._active_ttl_seconds:
continue
cadence = self._refresh_interval(window)
last = last_refresh.get(window, 0.0)
if last <= 0 and window in persisted:
last = self._generated_epoch(persisted[window])
if last <= 0 or now_wall - last >= cadence:
due.append(window)
return sorted(due)
def _refresh_interval(self, window: int) -> int:
base = self.interval_seconds
if window <= 900:
return base
if window <= 3600:
return max(base * 2, 120)
if window <= 21600:
return max(base * 5, 300)
return max(base * 15, 900)
def _overlay_current_throughput(self, payload: dict[str, Any]) -> None:
try:
sample = self.history.latest_throughput()
except RedisUnavailableError:
return
if not sample:
return
now_ms = int(time.time() * 1000)
ts_ms = int(sample.get("ts_ms") or 0)
interval = max(float(sample.get("interval_ms") or 1000) / 1000.0, 0.001)
age_ms = max(0, now_ms - ts_ms)
if age_ms > max(3000, round(interval * 3000)):
current = current_in = current_out = current_pps = 0
else:
current = round(max(int(sample.get("bytes_total") or 0), 0) * 8 / interval)
current_in = round(max(int(sample.get("bytes_in") or 0), 0) * 8 / interval)
current_out = round(max(int(sample.get("bytes_out") or 0), 0) * 8 / interval)
current_pps = round(max(int(sample.get("packets_total") or 0), 0) / interval, 2)
payload["current_bps"] = current
payload["current_in_bps"] = current_in
payload["current_out_bps"] = current_out
payload["current_other_bps"] = max(0, current - current_in - current_out)
payload["current_pps"] = current_pps
@staticmethod
def _generated_epoch(value: Any) -> float:
if not value:
return 0.0
try:
parsed = datetime.fromisoformat(str(value).replace("Z", "+00:00"))
if parsed.tzinfo is None:
parsed = parsed.replace(tzinfo=timezone.utc)
return parsed.timestamp()
except (TypeError, ValueError):
return 0.0
def _normalise_window(self, value: int) -> int:
return min(max(int(value), 60), self.history.retention_hours * 3600)
@staticmethod
def _decorate(payload: dict[str, Any], source: str) -> dict[str, Any]:
result = dict(payload)
generated = result.get("generated_at")
age = 0.0
if generated:
try:
parsed = datetime.fromisoformat(str(generated).replace("Z", "+00:00"))
if parsed.tzinfo is None:
parsed = parsed.replace(tzinfo=timezone.utc)
age = max(0.0, (datetime.now(timezone.utc) - parsed.astimezone(timezone.utc)).total_seconds())
except ValueError:
age = 0.0
result["snapshot_source"] = source
result["snapshot_age_seconds"] = round(age, 1)
return result