feat: add secure agent enrollment

This commit is contained in:
lussierj committed 2026-08-26 21:31:09 +00:00
1 parent 82309db79a
commit d9f0da694b
26 files changed
+2942 -76

No files matched your search

@@ -0,0 +1,625 @@
"""PostgreSQL integration tests for secure Agent enrollment."""
import unittest
from concurrent.futures import ThreadPoolExecutor
from uuid import uuid4
from psycopg.rows import dict_row
from app.agent_auth.exceptions import AgentAuthenticationFailed
from app.agent_auth.repository import AgentAuthRepository
from app.agent_auth.service import AgentAuthService, hash_agent_secret
from app.agent_enrollment.exceptions import (
AgentEnrollmentConflict,
AgentEnrollmentNotFound,
AgentEnrollmentTokenInvalid,
)
from app.agent_enrollment.models import (
AgentEnrollmentCreateRequest,
AgentEnrollRequest,
)
from app.agent_enrollment.repository import (
AgentEnrollmentRepository,
default_connection_factory,
)
from app.agent_enrollment.service import AgentEnrollmentService
from app.agent_zones.models import AgentCheckinRequest
from app.agent_zones.repository import AgentZoneRepository
from app.agent_zones.service import AgentZoneService
class AgentEnrollmentPostgresTests(unittest.TestCase):
def setUp(self):
self.site_id = uuid4()
self.zone_id = uuid4()
self.site_name = f"Enrollment Test Site {self.site_id.hex[:8]}"
self.zone_name = f"Enrollment Test Zone {self.zone_id.hex[:8]}"
connection = default_connection_factory()
try:
with connection.cursor() as cur:
cur.execute(
"""
INSERT INTO sites (id, name)
VALUES (%s, %s)
""",
(self.site_id, self.site_name),
)
cur.execute(
"""
INSERT INTO agent_zones (
id,
site_id,
name
)
VALUES (%s, %s, %s)
""",
(
self.zone_id,
self.site_id,
self.zone_name,
),
)
cur.execute(
"""
INSERT INTO agent_zone_network_ranges (
agent_zone_id,
network
)
VALUES (%s, '10.250.0.0/24'::cidr)
""",
(self.zone_id,),
)
connection.commit()
finally:
connection.close()
self.service = AgentEnrollmentService(
AgentEnrollmentRepository()
)
def tearDown(self):
connection = default_connection_factory()
try:
with connection.cursor() as cur:
cur.execute(
"""
DELETE FROM audit_logs
WHERE entity_type = 'agent_enrollment'
AND (
new_value_json ->> 'site_id' = %s
OR entity_id IN (
SELECT id
FROM agent_enrollments
WHERE site_id = %s
)
)
""",
(str(self.site_id), self.site_id),
)
cur.execute(
"""
DELETE FROM agent_enrollments
WHERE site_id = %s
""",
(self.site_id,),
)
cur.execute(
"""
DELETE FROM minerops_site_agents
WHERE site_id = %s
""",
(self.site_id,),
)
cur.execute(
"""
DELETE FROM agent_zone_network_ranges
WHERE agent_zone_id = %s
""",
(self.zone_id,),
)
cur.execute(
"""
DELETE FROM agent_zones
WHERE id = %s
""",
(self.zone_id,),
)
cur.execute(
"""
DELETE FROM sites
WHERE id = %s
""",
(self.site_id,),
)
connection.commit()
finally:
connection.close()
def create_enrollment(self, display_name="Integration Agent"):
# created_by_user_id is nullable at the persistence layer.
# Router-level tests separately prove that the real admin path supplies
# the authenticated user's UUID.
return self.service.create_enrollment(
AgentEnrollmentCreateRequest(
site_id=self.site_id,
agent_zone_id=self.zone_id,
display_name=display_name,
),
actor_user_id=None,
)
def test_replacement_revokes_previous_open_enrollment(self):
first = self.create_enrollment("First Agent")
second = self.create_enrollment("Replacement Agent")
self.assertNotEqual(
first["enrollment_token"],
second["enrollment_token"],
)
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT
token_hash,
used_at,
revoked_at
FROM agent_enrollments
WHERE site_id = %s
AND agent_zone_id = %s
""",
(self.site_id, self.zone_id),
)
rows = cur.fetchall()
finally:
connection.close()
self.assertEqual(len(rows), 2)
by_hash = {
row["token_hash"]: row
for row in rows
}
first_hash = hash_agent_secret(
first["enrollment_token"]
)
second_hash = hash_agent_secret(
second["enrollment_token"]
)
self.assertIsNotNone(by_hash[first_hash]["revoked_at"])
self.assertIsNone(by_hash[first_hash]["used_at"])
self.assertIsNone(by_hash[second_hash]["revoked_at"])
self.assertIsNone(by_hash[second_hash]["used_at"])
with self.assertRaises(AgentEnrollmentTokenInvalid):
self.service.redeem(
AgentEnrollRequest(
token=first["enrollment_token"]
)
)
listed = self.service.list_enrollments(
site_id=self.site_id,
agent_zone_id=self.zone_id,
)
self.assertEqual(len(listed), 2)
for item in listed:
self.assertNotIn("token_hash", item)
self.assertNotIn("agent_secret_hash", item)
self.assertNotIn("enrollment_token", item)
self.assertNotIn("agent_secret", item)
def test_concurrent_creation_leaves_exactly_one_open_enrollment(self):
def create_one(number):
service = AgentEnrollmentService(
AgentEnrollmentRepository()
)
return service.create_enrollment(
AgentEnrollmentCreateRequest(
site_id=self.site_id,
agent_zone_id=self.zone_id,
display_name=f"Concurrent Agent {number}",
),
actor_user_id=None,
)
with ThreadPoolExecutor(max_workers=2) as executor:
results = list(
executor.map(create_one, (1, 2))
)
self.assertEqual(len(results), 2)
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT
count(*) AS total,
count(*) FILTER (
WHERE used_at IS NULL
AND revoked_at IS NULL
) AS open_count,
count(*) FILTER (
WHERE revoked_at IS NOT NULL
) AS revoked_count,
count(*) FILTER (
WHERE revoked_at IS NOT NULL
AND revoked_at < created_at
) AS invalid_revocation_count
FROM agent_enrollments
WHERE site_id = %s
AND agent_zone_id = %s
""",
(self.site_id, self.zone_id),
)
counts = cur.fetchone()
finally:
connection.close()
self.assertEqual(counts["total"], 2)
self.assertEqual(counts["open_count"], 1)
self.assertEqual(counts["revoked_count"], 1)
self.assertEqual(counts["invalid_revocation_count"], 0)
def test_concurrent_redemption_has_exactly_one_winner(self):
enrollment = self.create_enrollment()
token = enrollment["enrollment_token"]
def redeem_once():
service = AgentEnrollmentService(
AgentEnrollmentRepository()
)
try:
return (
"success",
service.redeem(
AgentEnrollRequest(token=token)
),
)
except AgentEnrollmentTokenInvalid:
return ("invalid", None)
with ThreadPoolExecutor(max_workers=2) as executor:
results = list(
executor.map(lambda _: redeem_once(), range(2))
)
successes = [
result
for status, result in results
if status == "success"
]
failures = [
status
for status, _ in results
if status == "invalid"
]
self.assertEqual(len(successes), 1)
self.assertEqual(len(failures), 1)
credentials = successes[0]
agent_id = credentials["agent_id"]
agent_secret = credentials["agent_secret"]
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT
agent_id,
agent_secret_hash,
credential_created_at,
credential_revoked_at,
last_payload::text AS last_payload
FROM minerops_site_agents
WHERE site_id = %s
AND agent_zone_id = %s
""",
(self.site_id, self.zone_id),
)
agents = cur.fetchall()
cur.execute(
"""
SELECT
used_at,
revoked_at,
enrolled_agent_id,
token_hash
FROM agent_enrollments
WHERE id = %s
""",
(enrollment["id"],),
)
enrollment_row = cur.fetchone()
cur.execute(
"""
SELECT new_value_json::text AS payload
FROM audit_logs
WHERE entity_type = 'agent_enrollment'
AND entity_id = %s
ORDER BY created_at
""",
(enrollment["id"],),
)
audit_rows = cur.fetchall()
finally:
connection.close()
self.assertEqual(len(agents), 1)
agent = agents[0]
self.assertEqual(agent["agent_id"], agent_id)
self.assertEqual(
agent["agent_secret_hash"],
hash_agent_secret(agent_secret),
)
self.assertIsNotNone(agent["credential_created_at"])
self.assertIsNone(agent["credential_revoked_at"])
self.assertIsNotNone(enrollment_row["used_at"])
self.assertIsNone(enrollment_row["revoked_at"])
self.assertEqual(
enrollment_row["enrolled_agent_id"],
agent_id,
)
# Neither plaintext credential may persist in operational state.
persisted_text = "\n".join(
[
str(enrollment_row),
str(agent),
*(row["payload"] for row in audit_rows),
]
)
self.assertNotIn(token, persisted_text)
self.assertNotIn(agent_secret, persisted_text)
self.assertNotIn(agent_secret, agent["last_payload"])
# The newly issued per-Agent credential must immediately authenticate.
auth_service = AgentAuthService(
AgentAuthRepository()
)
self.assertEqual(
auth_service.authenticate(
agent_id,
agent_secret,
),
agent_id,
)
with self.assertRaises(AgentAuthenticationFailed):
auth_service.authenticate(
agent_id,
"definitely-not-the-issued-secret",
)
# Normal check-in must never mutate the credential.
secret_hash_before = agent["agent_secret_hash"]
AgentZoneService(
AgentZoneRepository()
).checkin(
agent_id,
AgentCheckinRequest(
site_id=self.site_id,
agent_zone_id=self.zone_id,
site_name=self.site_name,
hostname="integration-test-agent",
version="integration-test",
),
)
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT
agent_secret_hash,
credential_created_at,
credential_revoked_at
FROM minerops_site_agents
WHERE agent_id = %s
""",
(agent_id,),
)
after_checkin = cur.fetchone()
finally:
connection.close()
self.assertEqual(
after_checkin["agent_secret_hash"],
secret_hash_before,
)
self.assertIsNotNone(
after_checkin["credential_created_at"]
)
self.assertIsNone(
after_checkin["credential_revoked_at"]
)
def test_expired_token_is_rejected_without_side_effects(self):
enrollment = self.create_enrollment()
connection = default_connection_factory()
try:
with connection.cursor() as cur:
cur.execute(
"""
UPDATE agent_enrollments
SET created_at = now() - interval '2 hours',
expires_at = now() - interval '1 hour'
WHERE id = %s
""",
(enrollment["id"],),
)
connection.commit()
finally:
connection.close()
with self.assertRaises(AgentEnrollmentTokenInvalid):
self.service.redeem(
AgentEnrollRequest(
token=enrollment["enrollment_token"]
)
)
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT used_at, revoked_at, enrolled_agent_id
FROM agent_enrollments
WHERE id = %s
""",
(enrollment["id"],),
)
row = cur.fetchone()
cur.execute(
"""
SELECT count(*) AS count
FROM minerops_site_agents
WHERE agent_zone_id = %s
""",
(self.zone_id,),
)
agent_count = cur.fetchone()["count"]
finally:
connection.close()
self.assertIsNone(row["used_at"])
self.assertIsNone(row["revoked_at"])
self.assertIsNone(row["enrolled_agent_id"])
self.assertEqual(agent_count, 0)
def test_disabled_and_occupied_zone_do_not_consume_token(self):
enrollment = self.create_enrollment()
token = enrollment["enrollment_token"]
connection = default_connection_factory()
try:
with connection.cursor() as cur:
cur.execute(
"""
UPDATE agent_zones
SET is_enabled = false
WHERE id = %s
""",
(self.zone_id,),
)
connection.commit()
finally:
connection.close()
with self.assertRaises(AgentEnrollmentConflict):
self.service.redeem(
AgentEnrollRequest(token=token)
)
connection = default_connection_factory()
try:
with connection.cursor() as cur:
cur.execute(
"""
UPDATE agent_zones
SET is_enabled = true
WHERE id = %s
""",
(self.zone_id,),
)
cur.execute(
"""
INSERT INTO minerops_site_agents (
agent_id,
site_id,
agent_zone_id,
site_name
)
VALUES (%s, %s, %s, %s)
""",
(
f"legacy-{uuid4().hex}",
self.site_id,
self.zone_id,
self.site_name,
),
)
connection.commit()
finally:
connection.close()
with self.assertRaises(AgentEnrollmentConflict):
self.service.redeem(
AgentEnrollRequest(token=token)
)
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT used_at, revoked_at, enrolled_agent_id
FROM agent_enrollments
WHERE id = %s
""",
(enrollment["id"],),
)
row = cur.fetchone()
finally:
connection.close()
self.assertIsNone(row["used_at"])
self.assertIsNone(row["revoked_at"])
self.assertIsNone(row["enrolled_agent_id"])
def test_site_zone_mismatch_is_rejected_before_creation(self):
wrong_site_id = uuid4()
with self.assertRaises(AgentEnrollmentNotFound):
self.service.create_enrollment(
AgentEnrollmentCreateRequest(
site_id=wrong_site_id,
agent_zone_id=self.zone_id,
display_name="Wrong Site Agent",
),
actor_user_id=None,
)
connection = default_connection_factory()
try:
with connection.cursor(row_factory=dict_row) as cur:
cur.execute(
"""
SELECT count(*) AS count
FROM agent_enrollments
WHERE agent_zone_id = %s
""",
(self.zone_id,),
)
count = cur.fetchone()["count"]
finally:
connection.close()
self.assertEqual(count, 0)
if __name__ == "__main__":
unittest.main()