security-alert-center/backend/app/services/host_silence_scan.py

"""Периодическое сканирование хостов с устаревшим agent.heartbeat (proactive host_silence)."""

from __future__ import annotations

import logging
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone

from sqlalchemy import select
from sqlalchemy.orm import Session

from app.config import get_settings
from app.models import Host, Problem
from app.services.problem_rules import RuleMatch, build_fingerprint, host_silence_match_for_host
from app.services.problems import _correlation_cutoff

logger = logging.getLogger(__name__)


@dataclass(frozen=True)
class HostSilenceScanResult:
    host_id: int
    hostname: str
    problem: Problem
    created: bool


def open_or_refresh_host_silence_problem(
    db: Session,
    host_id: int,
    match: RuleMatch,
    *,
    now: datetime | None = None,
) -> tuple[Problem | None, bool]:
    """
    Открыть или обновить open Problem host_silence без ingest-события.

    Returns (problem, created). problem is None when suppressed by manual-resolve cooldown.
    """
    ref = now or datetime.now(timezone.utc)
    fingerprint = build_fingerprint(
        host_id,
        match.correlation_type,
        match.rule_id,
        match.fingerprint_suffix,
    )
    cutoff = _correlation_cutoff(ref)

    open_problem = db.scalar(
        select(Problem)
        .where(
            Problem.status == "open",
            Problem.fingerprint == fingerprint,
            Problem.host_id == host_id,
            Problem.last_seen_at >= cutoff,
        )
        .order_by(Problem.last_seen_at.desc())
        .limit(1)
    )
    if open_problem:
        open_problem.summary = match.summary
        open_problem.title = match.title
        open_problem.last_seen_at = ref
        open_problem.updated_at = ref
        db.flush()
        return open_problem, False

    settings = get_settings()
    cooldown_hours = settings.sac_host_silence_manual_resolve_cooldown_hours
    if cooldown_hours > 0:
        cooldown = timedelta(hours=cooldown_hours)
        manual_recent = db.scalar(
            select(Problem)
            .where(
                Problem.host_id == host_id,
                Problem.rule_id == match.rule_id,
                Problem.fingerprint == fingerprint,
                Problem.status == "resolved",
                Problem.resolved_by == "manual",
                Problem.updated_at >= ref - cooldown,
            )
            .order_by(Problem.updated_at.desc())
            .limit(1)
        )
        if manual_recent is not None:
            logger.debug(
                "host_silence suppressed (manual cooldown %sh) host_id=%s problem_id=%s",
                cooldown_hours,
                host_id,
                manual_recent.id,
            )
            return None, False

        manual_expired = db.scalar(
            select(Problem)
            .where(
                Problem.host_id == host_id,
                Problem.rule_id == match.rule_id,
                Problem.fingerprint == fingerprint,
                Problem.status == "resolved",
                Problem.resolved_by == "manual",
                Problem.updated_at < ref - cooldown,
            )
            .order_by(Problem.updated_at.desc())
            .limit(1)
        )
        if manual_expired is not None:
            manual_expired.status = "open"
            manual_expired.resolved_by = None
            manual_expired.summary = match.summary
            manual_expired.title = match.title
            manual_expired.last_seen_at = ref
            manual_expired.updated_at = ref
            db.flush()
            return manual_expired, True

    problem = Problem(
        host_id=host_id,
        title=match.title,
        summary=match.summary,
        severity=match.severity,
        status="open",
        rule_id=match.rule_id,
        fingerprint=fingerprint,
        event_count=0,
        last_seen_at=ref,
    )
    db.add(problem)
    db.flush()
    return problem, True


def run_host_silence_scan(db: Session, *, now: datetime | None = None) -> list[HostSilenceScanResult]:
    """Найти stale-хосты и создать/обновить Problems rule:host_silence."""
    ref = now or datetime.now(timezone.utc)
    hosts = db.scalars(select(Host).order_by(Host.id)).all()
    results: list[HostSilenceScanResult] = []

    for host in hosts:
        match = host_silence_match_for_host(db, host.id, hostname=host.hostname, now=ref)
        if match is None:
            continue
        problem, created = open_or_refresh_host_silence_problem(db, host.id, match, now=ref)
        if problem is None:
            continue
        results.append(
            HostSilenceScanResult(
                host_id=host.id,
                hostname=host.hostname,
                problem=problem,
                created=created,
            )
        )
        if created:
            logger.info(
                "host_silence scan: new problem host=%s id=%s problem_id=%s",
                host.hostname,
                host.id,
                problem.id,
            )

    return results


def dispatch_host_silence_notifications(db: Session, results: list[HostSilenceScanResult]) -> int:
    """Telegram/webhook/email только для вновь созданных Problems."""
    from app.services.notify_dispatch import notify_problem

    sent = 0
    for item in results:
        if not item.created:
            continue
        try:
            notify_problem(item.problem, event=None, db=db)
            sent += 1
        except Exception:
            logger.exception(
                "host_silence notify failed host=%s problem_id=%s",
                item.hostname,
                item.problem.id,
            )
    return sent