feat: Problems correlation fingerprint, count, last_seen (d1-2)

- Migration 003; windowed host+type+rule correlation on ingest - GET /problems filters; GET /problems/{id} with event timeline; ack/resolve 409 guard - Tests and SAC_PROBLEM_CORRELATION_WINDOW_MINUTES config Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-28 09:46:12 +10:00
parent 8eb40cf75d
commit a4a224b284
10 changed files with 365 additions and 126 deletions
@@ -0,0 +1,49 @@
+"""problem correlation fields
+
+Revision ID: 003
+Revises: 002
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "003"
+down_revision: Union[str, None] = "002"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column("problems", sa.Column("fingerprint", sa.String(length=128), nullable=True))
+    op.add_column(
+        "problems",
+        sa.Column("event_count", sa.Integer(), nullable=False, server_default="1"),
+    )
+    op.add_column("problems", sa.Column("last_seen_at", sa.DateTime(timezone=True), nullable=True))
+    op.create_index("ix_problems_fingerprint", "problems", ["fingerprint"])
+    op.create_index("ix_problems_last_seen_at", "problems", ["last_seen_at"])
+
+    op.execute(
+        """
+        UPDATE problems p SET
+            fingerprint = 'h' || COALESCE(p.host_id::text, '0') || ':r' || COALESCE(p.rule_id, 'unknown'),
+            event_count = COALESCE(
+                (SELECT COUNT(*)::int FROM problem_events pe WHERE pe.problem_id = p.id),
+                1
+            ),
+            last_seen_at = COALESCE(p.updated_at, p.created_at)
+        """
+    )
+
+    op.alter_column("problems", "fingerprint", nullable=False)
+    op.alter_column("problems", "last_seen_at", nullable=False)
+
+
+def downgrade() -> None:
+    op.drop_index("ix_problems_last_seen_at", table_name="problems")
+    op.drop_index("ix_problems_fingerprint", table_name="problems")
+    op.drop_column("problems", "last_seen_at")
+    op.drop_column("problems", "event_count")
+    op.drop_column("problems", "fingerprint")
@@ -7,7 +7,7 @@ from sqlalchemy.orm import Session, joinedload
 from pydantic import BaseModel

 from sqlalchemy import func, select
-from app.models import Problem
+
 from sqlalchemy.orm import Session, joinedload


@@ -21,11 +21,12 @@ class ProblemSummary(BaseModel):


 router = APIRouter(prefix="/problems", tags=["problems"])
+
+
+


 class ProblemSummary(BaseModel):
-    model_config = {"from_attributes": True}
-

    id: int

@@ -34,43 +35,112 @@ class ProblemListResponse(BaseModel):
    hostname: str | None

    title: str
+
+    summary: str
+
+    severity: str
+
+    status: str
+
+    rule_id: str | None
+
+    fingerprint: str
+
+    event_count: int
+
+    last_seen_at: datetime
+
+    created_at: datetime
+
+    updated_at: datetime
+
+
+
+
+
+class ProblemListResponse(BaseModel):
+
+    items: list[ProblemSummary]
+
+    total: int
+
+    page: int
+
+    page_size: int





+class ProblemEventItem(BaseModel):
+
+    id: int

    event_id: str

-    stmt = select(Problem).options(joinedload(Problem.host))
-    count_stmt = select(func.count()).select_from(Problem)
+    occurred_at: datetime
+
+    type: str

    severity: str

+    title: str
+
+    summary: str
+
+
+
+
+
+class ProblemDetail(ProblemSummary):
+
    events: list[ProblemEventItem]


-        stmt.order_by(Problem.updated_at.desc()).offset((page - 1) * page_size).limit(page_size)
+


-    items = [
-        ProblemSummary(
-            id=p.id,
-            host_id=p.host_id,
-            hostname=p.host.hostname if p.host else None,
-            title=p.title,
-            summary=p.summary,
-            severity=p.severity,
-            status=p.status,
-            rule_id=p.rule_id,
-            created_at=p.created_at,
-            updated_at=p.updated_at,
-        )
-        for p in rows
-    ]
+def _problem_summary(p: Problem) -> ProblemSummary:

    return ProblemSummary(

+        id=p.id,
+
+        host_id=p.host_id,
+
+        hostname=p.host.hostname if p.host else None,
+
+        title=p.title,
+
+        summary=p.summary,
+
+        severity=p.severity,
+
+        status=p.status,
+
+        rule_id=p.rule_id,
+
+        fingerprint=p.fingerprint,
+
+        event_count=p.event_count,
+
+        last_seen_at=p.last_seen_at,
+
+        created_at=p.created_at,
+
+        updated_at=p.updated_at,
+
+    )
+
+
+
+
+
+@router.get("", response_model=ProblemListResponse)
+
+def list_problems(
+
+    page: int = Query(1, ge=1),

    page_size: int = Query(50, ge=1, le=200),

@@ -80,6 +150,8 @@ def ack_problem(

    host_id: int | None = Query(None),

+    hostname: str | None = Query(None),
+
    db: Session = Depends(get_db),

    _user: str = Depends(get_current_user),
@@ -94,6 +166,9 @@ def resolve_problem(

    if status:

+        stmt = stmt.where(Problem.status == status)
+
        count_stmt = count_stmt.where(Problem.status == status)

    if severity:
+
@@ -53,6 +53,9 @@ class Settings(BaseSettings):
    # Порог «живости» агента: agent.heartbeat раз в ~12 ч (ssh-monitor)
    sac_heartbeat_stale_minutes: int = 780

+    # Окно корреляции Problems: host + type + rule в одном open Problem
+    sac_problem_correlation_window_minutes: int = 60
+

@lru_cache
 def get_settings() -> Settings:
@@ -1,6 +1,6 @@
 from datetime import datetime

-from sqlalchemy import DateTime, ForeignKey, String, Text, func
+from sqlalchemy import DateTime, ForeignKey, Integer, String, Text, func
 from sqlalchemy.orm import Mapped, mapped_column, relationship

 from app.database import Base
@@ -16,6 +16,9 @@ class Problem(Base):
    severity: Mapped[str] = mapped_column(String(16), index=True)
    status: Mapped[str] = mapped_column(String(32), index=True, default="open")
    rule_id: Mapped[str | None] = mapped_column(String(64))
+    fingerprint: Mapped[str] = mapped_column(String(128), index=True)
+    event_count: Mapped[int] = mapped_column(Integer, default=1)
+    last_seen_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), index=True)
    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
    updated_at: Mapped[datetime] = mapped_column(
        DateTime(timezone=True), server_default=func.now(), onupdate=func.now()
@@ -1,11 +1,13 @@
-"""Auto-create Problems from ingested events (MVP rules)."""
+"""Auto-create Problems from ingested events (MVP rules + correlation)."""
+
+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 Event, Problem, ProblemEvent

-# event types that always open a problem
 PROBLEM_TYPES = frozenset(
    {
        "ssh.ip.banned",
@@ -16,41 +18,63 @@ PROBLEM_TYPES = frozenset(
 HIGH_SEVERITIES = frozenset({"high", "critical"})


-def maybe_create_problem(db: Session, event: Event) -> tuple[Problem | None, bool]:
-    if event.severity not in HIGH_SEVERITIES and event.type not in PROBLEM_TYPES:
-        return None, False
+def problem_rule_id(event: Event) -> str | None:
+    if event.severity in HIGH_SEVERITIES:
+        return "high_severity"
+    if event.type in PROBLEM_TYPES:
+        return f"type:{event.type}"
+    return None

-    rule_id = "high_severity" if event.severity in HIGH_SEVERITIES else f"type:{event.type}"

-    existing = db.scalar(
-        select(Problem)
-        .join(ProblemEvent)
-        .where(
-            Problem.status == "open",
-            Problem.rule_id == rule_id,
-            Problem.host_id == event.host_id,
+def problem_fingerprint(host_id: int, event_type: str, rule_id: str) -> str:
+    return f"h{host_id}:t{event_type}:r{rule_id}"
+
+
+def _correlation_cutoff(now: datetime) -> datetime:
+    minutes = get_settings().sac_problem_correlation_window_minutes
+    return now - timedelta(minutes=minutes)
+
+
+def _append_event(db: Session, problem: Problem, event: Event) -> None:
+    linked = db.scalar(
+        select(ProblemEvent).where(
+            ProblemEvent.problem_id == problem.id,
            ProblemEvent.event_id == event.id,
        )
-        .limit(1)
    )
-    if existing:
-        return existing, False
+    if linked is not None:
+        return
+    db.add(ProblemEvent(problem_id=problem.id, event_id=event.id))
+    problem.event_count = (problem.event_count or 0) + 1
+    problem.last_seen_at = event.occurred_at
+    problem.updated_at = datetime.now(timezone.utc)
+    if event.severity in HIGH_SEVERITIES and problem.severity not in HIGH_SEVERITIES:
+        problem.severity = event.severity
+    db.flush()
+
+
+def maybe_create_problem(db: Session, event: Event) -> tuple[Problem | None, bool]:
+    rule_id = problem_rule_id(event)
+    if rule_id is None:
+        return None, False
+
+    fingerprint = problem_fingerprint(event.host_id, event.type, rule_id)
+    now = datetime.now(timezone.utc)
+    cutoff = _correlation_cutoff(now)

-    # dedupe: one open problem per host+rule (append event link)
    open_problem = db.scalar(
        select(Problem)
        .where(
            Problem.status == "open",
-            Problem.rule_id == rule_id,
+            Problem.fingerprint == fingerprint,
            Problem.host_id == event.host_id,
+            Problem.last_seen_at >= cutoff,
        )
-        .order_by(Problem.created_at.desc())
+        .order_by(Problem.last_seen_at.desc())
        .limit(1)
    )
    if open_problem:
-        db.add(ProblemEvent(problem_id=open_problem.id, event_id=event.id))
-        open_problem.updated_at = event.received_at
-        db.flush()
+        _append_event(db, open_problem, event)
        return open_problem, False

    problem = Problem(
@@ -60,6 +84,9 @@ def maybe_create_problem(db: Session, event: Event) -> tuple[Problem | None, boo
        severity=event.severity,
        status="open",
        rule_id=rule_id,
+        fingerprint=fingerprint,
+        event_count=1,
+        last_seen_at=event.occurred_at,
    )
    db.add(problem)
    db.flush()
@@ -14,6 +14,7 @@ from sqlalchemy.orm import sessionmaker
 from sqlalchemy.pool import StaticPool

 from app.auth.api_key import hash_api_key
+from app.auth.jwt_auth import create_access_token
 from app.database import Base, get_db
 from app.main import app as fastapi_app
 import app.models  # noqa: F401 — register all tables on Base.metadata
@@ -80,3 +81,9 @@ def client(db_session, monkeypatch):
@pytest.fixture
 def auth_headers():
    return {"Authorization": f"Bearer {TEST_API_KEY}"}
+
+
+@pytest.fixture
+def jwt_headers():
+    token = create_access_token("test-admin")
+    return {"Authorization": f"Bearer {token}"}
@@ -0,0 +1,69 @@
+"""Problems correlation and API smoke tests."""
+
+import uuid
+from datetime import datetime, timezone
+
+from tests.test_ingest import VALID_EVENT
+
+
+def _event_payload(**overrides):
+    now = datetime.now(timezone.utc).astimezone().isoformat(timespec="seconds")
+    base = {
+        **VALID_EVENT,
+        "event_id": str(uuid.uuid4()),
+        "occurred_at": now,
+        "severity": "high",
+        "type": "ssh.ip.banned",
+        "title": "Ban test",
+        "summary": "test problem correlation",
+    }
+    base.update(overrides)
+    return base
+
+
+def test_problem_correlates_two_events(client, auth_headers, jwt_headers):
+    e1 = _event_payload()
+    e2 = _event_payload(title="Ban test 2")
+    assert client.post("/api/v1/events", json=e1, headers=auth_headers).status_code == 201
+    assert client.post("/api/v1/events", json=e2, headers=auth_headers).status_code == 201
+
+    r = client.get("/api/v1/problems?status=open", headers=jwt_headers)
+    assert r.status_code == 200
+    data = r.json()
+    assert data["total"] == 1
+    problem = data["items"][0]
+    assert problem["event_count"] == 2
+    assert problem["fingerprint"].startswith("h")
+    assert ":tssh.ip.banned:" in problem["fingerprint"]
+
+    detail = client.get(f"/api/v1/problems/{problem['id']}", headers=jwt_headers)
+    assert detail.status_code == 200
+    assert len(detail.json()["events"]) == 2
+
+
+def test_problem_ack_and_resolve(client, auth_headers, jwt_headers):
+    payload = _event_payload()
+    client.post("/api/v1/events", json=payload, headers=auth_headers)
+    pid = client.get("/api/v1/problems?status=open", headers=jwt_headers).json()["items"][0]["id"]
+
+    ack = client.post(f"/api/v1/problems/{pid}/ack", headers=jwt_headers)
+    assert ack.status_code == 200
+    assert ack.json()["status"] == "acknowledged"
+
+    res = client.post(f"/api/v1/problems/{pid}/resolve", headers=jwt_headers)
+    assert res.status_code == 200
+    assert res.json()["status"] == "resolved"
+
+
+def test_new_problem_after_resolve(client, auth_headers, jwt_headers):
+    p1 = _event_payload()
+    client.post("/api/v1/events", json=p1, headers=auth_headers)
+    pid = client.get("/api/v1/problems?status=open", headers=jwt_headers).json()["items"][0]["id"]
+    client.post(f"/api/v1/problems/{pid}/resolve", headers=jwt_headers)
+
+    p2 = _event_payload(summary="after resolve")
+    client.post("/api/v1/events", json=p2, headers=auth_headers)
+    open_items = client.get("/api/v1/problems?status=open", headers=jwt_headers).json()["items"]
+    assert len(open_items) == 1
+    assert open_items[0]["event_count"] == 1
+    assert open_items[0]["id"] != pid
@@ -26,4 +26,7 @@ TELEGRAM_MIN_SEVERITY=high
 # Статус хоста по agent.heartbeat (минуты; ssh-monitor шлёт ~раз в 12 ч)
 SAC_HEARTBEAT_STALE_MINUTES=780

+# Корреляция Problems: host + type + rule в одном open Problem (минуты)
+SAC_PROBLEM_CORRELATION_WINDOW_MINUTES=60
+
 CORS_ORIGINS=*
@@ -82,7 +82,7 @@
 ### День 1

 - [x] `d1-1` Ingest: `event_id` UUID обязателен, UNIQUE индекс, коды duplicate, логи, docs/schema
- [ ] `d1-2` MVP Problems: модель, корреляция `host+type+окно`, API `GET/ack/resolve`
+- [x] `d1-2` MVP Problems: модель, корреляция `host+type+окно`, API `GET/ack/resolve`
 - [ ] `d1-3` Правила v1: `brute-force burst`, `privilege spike`, `host silence`

 ### День 2
@@ -109,9 +109,9 @@
 - [x] `09:00–10:00` миграция `event_id` + UNIQUE, валидация ingest
 - [x] `10:00–11:00` поведение duplicate (`201/409`) и логирование reject
 - [x] `11:00–12:00` обновить `event-schema-v1.json` и `agent-integration.md` (в т.ч. `host.display_name` для агентов)
- [ ] `13:00–14:30` модель Problems + миграция
- [ ] `14:30–16:00` корреляция при ingest (`fingerprint`, `count`, `last_seen`)
- [ ] `16:00–17:30` API `GET /problems`, `ack`, `resolve` + smoke
+- [x] `13:00–14:30` модель Problems + миграция
+- [x] `14:30–16:00` корреляция при ingest (`fingerprint`, `count`, `last_seen`)
+- [x] `16:00–17:30` API `GET /problems`, `ack`, `resolve` + smoke
 - [ ] `18:00–19:00` правила `brute/privilege/silence` + unit tests

 #### День 2
@@ -97,6 +97,9 @@ export interface ProblemSummary {
  severity: string;
  status: string;
  rule_id: string | null;
+  fingerprint?: string;
+  event_count?: number;
+  last_seen_at?: string;
  created_at: string;
  updated_at: string;
 }