fix: dedupe host_silence problems under concurrent scans (0.4.11)

Treat acknowledged as active, dedupe by hostname, add PostgreSQL advisory lock and a unique index. Migration closes existing duplicate open problems.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
2026-07-03 09:05:28 +10:00
parent 3765d4d476
commit 6c43b8637e
9 changed files with 334 additions and 31 deletions
+117
View File
@@ -211,6 +211,123 @@ def test_scan_reopens_after_manual_cooldown_expires(db_session, scan_settings, m
assert later[0].problem.resolved_by is None
def test_scan_does_not_duplicate_acknowledged_problem(db_session, scan_settings):
hb = _ingest(
db_session,
type="agent.heartbeat",
category="agent",
severity="info",
title="hb",
summary="heartbeat",
)
hb.received_at = datetime.now(timezone.utc) - timedelta(hours=2)
db_session.flush()
now = datetime.now(timezone.utc)
first = run_host_silence_scan(db_session, now=now)
assert first[0].created is True
first[0].problem.status = "acknowledged"
db_session.flush()
second = run_host_silence_scan(db_session, now=now + timedelta(minutes=5))
assert len(second) == 1
assert second[0].created is False
assert second[0].problem.id == first[0].problem.id
problems = db_session.scalars(
select(Problem).where(
Problem.host_id == hb.host_id,
Problem.rule_id == RULE_HOST_SILENCE,
Problem.status.in_(("open", "acknowledged")),
)
).all()
assert len(problems) == 1
def test_scan_dedupes_by_hostname_across_duplicate_hosts(db_session, scan_settings):
from app.models import Host
now = datetime.now(timezone.utc)
host_a = Host(
hostname="COMM-PC",
os_family="windows",
product="rdp-login-monitor",
ipv4="192.168.162.33",
)
host_b = Host(
hostname="COMM-PC",
os_family="windows",
product="rdp-login-monitor",
ipv4="192.168.162.33",
)
db_session.add_all([host_a, host_b])
db_session.flush()
hb = _ingest(
db_session,
host={"hostname": host_a.hostname, "os_family": "windows", "ipv4": host_a.ipv4},
type="agent.heartbeat",
category="agent",
severity="info",
title="hb",
summary="heartbeat",
)
hb.host_id = host_a.id
hb.received_at = now - timedelta(hours=2)
db_session.flush()
first = run_host_silence_scan(db_session, now=now)
assert len(first) == 1
assert first[0].created is True
hb_b = _ingest(
db_session,
event_id=str(uuid.uuid4()),
host={"hostname": host_b.hostname, "os_family": "windows", "ipv4": host_b.ipv4},
source={"product": "rdp-login-monitor", "product_version": "2.0.0-SAC"},
type="agent.heartbeat",
category="agent",
severity="info",
title="hb2",
summary="heartbeat",
)
hb_b.host_id = host_b.id
hb_b.received_at = now - timedelta(hours=2)
db_session.flush()
second = run_host_silence_scan(db_session, now=now + timedelta(minutes=5))
assert len(second) == 2
assert all(not item.created for item in second)
assert {item.problem.id for item in second} == {first[0].problem.id}
active = db_session.scalars(
select(Problem).where(
Problem.rule_id == RULE_HOST_SILENCE,
Problem.status.in_(("open", "acknowledged")),
)
).all()
assert len(active) == 1
def test_scan_skipped_when_advisory_lock_not_acquired(db_session, scan_settings, monkeypatch):
monkeypatch.setattr(
"app.services.host_silence_scan._host_silence_scan_lock",
lambda _db: False,
)
hb = _ingest(
db_session,
type="agent.heartbeat",
category="agent",
severity="info",
title="hb",
summary="heartbeat",
)
hb.received_at = datetime.now(timezone.utc) - timedelta(hours=2)
db_session.flush()
assert run_host_silence_scan(db_session) == []
def test_scan_notifies_only_on_create(db_session, scan_settings):
hb = _ingest(
db_session,