151 lines
8.7 KiB
Python
151 lines
8.7 KiB
Python
import uuid
|
|
from datetime import timedelta
|
|
from unittest.mock import patch,Mock
|
|
import pytest
|
|
from django.utils import timezone
|
|
from django.contrib.auth import get_user_model
|
|
from rest_framework.test import APIClient
|
|
from monitoring.models import *
|
|
from monitoring.auth import digest
|
|
from monitoring.ingest import ingest
|
|
from monitoring.tasks import evaluate_alarms,cleanup,deliver_notifications
|
|
from agent.outbox import Outbox
|
|
from agent.main import envelope,safe_event,send_batch
|
|
@pytest.fixture
|
|
def server(db):
|
|
seg=Segment.objects.create(name='Production')
|
|
return Server.objects.create(name='host-01',segment=seg)
|
|
@pytest.fixture
|
|
def client():return APIClient()
|
|
@pytest.fixture
|
|
def auth(client,db):
|
|
user=get_user_model().objects.create_user(username='operator',password='test-secure-password-834!',is_staff=True)
|
|
token=AccessToken.objects.create(user=user,digest=digest('test-token'),expires_at=timezone.now()+timedelta(hours=1))
|
|
client.credentials(HTTP_AUTHORIZATION='Bearer test-token');return client
|
|
|
|
def ev(kind='metric',payload=None,at=None):
|
|
return {'id':uuid.uuid4(),'kind':kind,'occurred_at':at or timezone.now(),'payload':payload or {'cpu':22,'ram':45,'disk':61}}
|
|
@pytest.mark.django_db
|
|
def test_ingest_auth_scope_idempotence(server,client):
|
|
payload=ev();payload['id']=str(payload['id']);payload['occurred_at']=payload['occurred_at'].isoformat()
|
|
assert client.post('/api/ingest/',{'events':[payload]},format='json').status_code==401
|
|
AgentCredential.objects.create(server=server,digest=digest('agent-secret'))
|
|
client.credentials(HTTP_AUTHORIZATION='Agent agent-secret')
|
|
for _ in range(2):assert client.post('/api/ingest/',{'events':[payload]},format='json').status_code==200
|
|
assert Event.objects.count()==1 and Metric.objects.count()==1
|
|
payload['id']=str(uuid.uuid4());payload['payload']['cpu']=120
|
|
assert client.post('/api/ingest/',{'events':[payload]},format='json').status_code==400
|
|
assert Event.objects.count()==1
|
|
@pytest.mark.django_db
|
|
def test_transaction_validation_no_partial_batch(server,client):
|
|
AgentCredential.objects.create(server=server,digest=digest('agent-secret'));client.credentials(HTTP_AUTHORIZATION='Agent agent-secret')
|
|
events=[envelope('metric',{'cpu':30,'ram':40,'disk':50}),envelope('metric',{'cpu':-1,'ram':40,'disk':50})]
|
|
assert client.post('/api/ingest/',{'events':events},format='json').status_code==400
|
|
assert Event.objects.count()==0
|
|
@pytest.mark.django_db
|
|
def test_task_out_of_order_and_terminal_state(server):
|
|
now=timezone.now();task={'uuid':'t1','name':'x'}
|
|
ingest(server.pk,[ev('task-succeeded',dict(task,runtime=2),now),ev('task-started',task,now-timedelta(seconds=4)),ev('task-received',task,now+timedelta(seconds=1))])
|
|
t=Task.objects.get();assert t.state=='SUCCESS' and t.runtime==2
|
|
@pytest.mark.django_db
|
|
def test_stale_metric_replay_not_online(server,auth):
|
|
ingest(server.pk,[ev(at=timezone.now()-timedelta(hours=1))])
|
|
assert auth.get('/api/summary/').json()['servers'][0]['status']=='Offline'
|
|
evaluate_alarms();assert Alarm.objects.filter(code='heartbeat',resolved_at__isnull=True).exists()
|
|
@pytest.mark.django_db
|
|
def test_alarm_recovery_ack_and_permissions(server,auth):
|
|
ingest(server.pk,[ev(payload={'cpu':95,'ram':20,'disk':30})]);evaluate_alarms();evaluate_alarms()
|
|
a=Alarm.objects.get(code='cpu');assert Notification.objects.count()==1
|
|
assert auth.post(f'/api/alarms/{a.pk}/ack/').status_code==200
|
|
assert Audit.objects.filter(action='alarm_ack').count()==1
|
|
assert auth.post(f'/api/alarms/{a.pk}/ack/').status_code==200
|
|
assert Audit.objects.filter(action='alarm_ack').count()==1
|
|
ingest(server.pk,[ev()]);evaluate_alarms();a.refresh_from_db();assert a.resolved_at
|
|
user=get_user_model().objects.get(username='operator');user.is_staff=False;user.save()
|
|
assert auth.put('/api/settings/',{},format='json').status_code==403
|
|
@pytest.mark.django_db
|
|
def test_auth_expiry_and_revocation(auth):
|
|
assert auth.get('/api/summary/').status_code==200
|
|
AccessToken.objects.update(expires_at=timezone.now()-timedelta(seconds=1))
|
|
assert auth.get('/api/summary/').status_code==401
|
|
@pytest.mark.django_db
|
|
def test_retention(server):
|
|
ingest(server.pk,[ev(at=timezone.now()-timedelta(days=100)),ev()]);cleanup()
|
|
assert Event.objects.count()==1 and Metric.objects.count()==1
|
|
@pytest.mark.django_db
|
|
def test_notification_retries(server,settings):
|
|
settings.ALERT_RECIPIENTS=['ops@example.com'];evaluate_alarms()
|
|
with patch('monitoring.tasks.send_mail',side_effect=OSError()):deliver_notifications()
|
|
n=Notification.objects.first();assert n.attempts==1 and not n.sent_at
|
|
Notification.objects.update(next_attempt=timezone.now())
|
|
with patch('monitoring.tasks.send_mail',return_value=1):deliver_notifications()
|
|
n.refresh_from_db();assert n.sent_at
|
|
@pytest.mark.django_db
|
|
def test_report_csv_and_all_views(server,auth):
|
|
ingest(server.pk,[ev()])
|
|
for path in ['summary','servers','tasks','events','alarms','settings','flower','health']:
|
|
assert auth.get('/api/'+path+'/').status_code==200
|
|
assert auth.get(f'/api/servers/{server.pk}/').status_code==200
|
|
r=auth.get('/api/report/');assert r.status_code==200 and r.content.startswith(b'\x89PNG')
|
|
server.name='=EVIL()';server.save()
|
|
assert "'=EVIL()" in auth.get('/api/export/').content.decode()
|
|
@pytest.mark.django_db
|
|
def test_login_rate_limit_and_logout(client):
|
|
from django.core.cache import cache
|
|
cache.clear()
|
|
get_user_model().objects.create_user(username='test-login',password='example-Password!31')
|
|
response=client.post('/api/auth/login/',{'username':'test-login','password':'example-Password!31'},format='json')
|
|
assert response.status_code==200
|
|
client.credentials(HTTP_AUTHORIZATION='Bearer '+response.json()['token'])
|
|
assert client.post('/api/auth/logout/').status_code==200
|
|
assert client.get('/api/summary/').status_code==401
|
|
client.credentials()
|
|
for _ in range(10):client.post('/api/auth/login/',{'username':'bad','password':'bad'},format='json')
|
|
assert client.post('/api/auth/login/',{'username':'bad','password':'bad'},format='json').status_code==429
|
|
|
|
def test_outbox_survives_restart_and_failed_delivery(tmp_path):
|
|
path=tmp_path/'queue.db';out=Outbox(path);item=envelope('metric',{'cpu':1,'ram':2,'disk':3});out.add(item);out.add(item);out.close()
|
|
out=Outbox(path);assert out.count()==1
|
|
session=Mock();session.post.side_effect=OSError('offline')
|
|
with pytest.raises(OSError):send_batch(session,out,'https://example.com','secret')
|
|
assert out.count()==1
|
|
session.post.side_effect=None;session.post.return_value.json.return_value={'received':1}
|
|
assert send_batch(session,out,'https://example.com','secret')==1 and out.count()==0
|
|
|
|
def test_payload_redaction():
|
|
e=safe_event({'type':'task-succeeded','uuid':'a','hostname':'worker','args':'SECRET','result':'SECRET','kwargs':{'password':'SECRET'},'runtime':2})
|
|
assert set(e['payload'])=={'uuid','hostname','runtime'}
|
|
|
|
@pytest.mark.django_db
|
|
def test_ingest_normalizes_numeric_input(server,client):
|
|
AgentCredential.objects.create(server=server,digest=digest('numeric-agent'))
|
|
client.credentials(HTTP_AUTHORIZATION='Agent numeric-agent')
|
|
event=envelope('metric',{'cpu':'20.5','ram':'30','disk':'40','services':{'web':'false'}})
|
|
assert client.post('/api/ingest/',{'events':[event]},format='json').status_code==200
|
|
server.refresh_from_db()
|
|
assert server.latest['cpu']==20.5 and server.latest['services']['web'] is False
|
|
|
|
@pytest.mark.django_db
|
|
def test_batch_cross_agent_collision_rolls_back(server):
|
|
existing=ev();ingest(server.pk,[existing])
|
|
other=Server.objects.create(name='other',segment=server.segment)
|
|
with pytest.raises(ValueError):ingest(other.pk,[ev(),existing])
|
|
assert not Event.objects.filter(server=other).exists()
|
|
|
|
@pytest.mark.django_db
|
|
def test_payload_allowlist_and_future_clock(server,client):
|
|
AgentCredential.objects.create(server=server,digest=digest('agent-secret'));client.credentials(HTTP_AUTHORIZATION='Agent agent-secret')
|
|
item=envelope('task-succeeded',{'uuid':'x','name':'task','args':'secret','result':'secret'})
|
|
assert client.post('/api/ingest/',{'events':[item]},format='json').status_code==200
|
|
assert Event.objects.get().payload=={'uuid':'x','name':'task'}
|
|
item=envelope('metric',{'cpu':3,'ram':4,'disk':5},timezone.now()+timedelta(hours=1))
|
|
assert client.post('/api/ingest/',{'events':[item]},format='json').status_code==400
|
|
|
|
@pytest.mark.django_db
|
|
def test_metrics_survive_new_client_and_flower_restart(server,auth):
|
|
ingest(server.pk,[ev(),ev('flower-status',{'online':False})])
|
|
assert auth.get('/api/summary/').json()['online']==1
|
|
ingest(server.pk,[ev('flower-status',{'online':True})])
|
|
assert Metric.objects.count()==1 and Event.objects.count()==3
|
|
assert auth.get('/api/flower/').json()[0]['online'] is True
|