fix(selfhost): retire deprecated Matrix services during updates
This commit is contained in:
@@ -6,6 +6,7 @@ import os
|
||||
import io
|
||||
from pathlib import Path
|
||||
import subprocess
|
||||
import tarfile
|
||||
import tempfile
|
||||
import unittest
|
||||
from unittest.mock import patch
|
||||
@@ -35,6 +36,7 @@ class Http:
|
||||
self.bad_health = False
|
||||
self.release = {'id': 1, 'tag_name': 'v3.1.0', 'draft': False, 'prerelease': False}
|
||||
self.fail_download = False
|
||||
self.reported_version = '3.1.0'
|
||||
|
||||
def refresh(self):
|
||||
self.files['SHA256SUMS'] = ''.join(hashlib.sha256(blob).hexdigest() + ' ' + name + '\n'
|
||||
@@ -59,7 +61,7 @@ class Http:
|
||||
if url.endswith('/tags/v3.1.0'):
|
||||
return {'commit': {'sha': 'a' * 40}}
|
||||
if url.endswith('/backend/version'):
|
||||
return {'product': 'FEDEO', 'version': '3.1.0'}
|
||||
return {'product': 'FEDEO', 'version': self.reported_version}
|
||||
if url.endswith('/health'):
|
||||
return {'status': 'ok', 'product': 'FEDEO', 'version': '3.0.0' if self.bad_health else '3.1.0', 'revision': 'a' * 40}
|
||||
raise AssertionError(url)
|
||||
@@ -74,6 +76,9 @@ class Runner:
|
||||
self.fail = None
|
||||
self.on_pull = None
|
||||
self.interrupt = None
|
||||
self.matrix_containers = {}
|
||||
self.on_stop = None
|
||||
self.real_tar = False
|
||||
|
||||
def run(self, args, output=None, input_file=None, **kwargs):
|
||||
self.calls.append(args)
|
||||
@@ -91,8 +96,13 @@ class Runner:
|
||||
return 'image-' + args[3].rsplit('/', 1)[-1].split('@')[0]
|
||||
if args[:2] == ['docker', 'inspect']:
|
||||
name = args[2]
|
||||
if name in self.matrix_containers:
|
||||
return json.dumps([self.matrix_containers[name]])
|
||||
return json.dumps([{'Image': 'image-' + name, 'Config': {'Labels': {'com.docker.compose.service': name}},
|
||||
'State': {'Running': name != 'createbuckets', 'Health': {'Status': 'healthy'}}}])
|
||||
if args[:3] == ['docker', 'update', '--restart=no']:
|
||||
self.matrix_containers[args[3]]['HostConfig']['RestartPolicy'] = {'Name': 'no', 'MaximumRetryCount': 0}
|
||||
return args[3]
|
||||
if 'config' in args:
|
||||
result = subprocess.run(args, env={'PATH': os.environ['PATH'], 'HOME': os.environ['HOME']},
|
||||
capture_output=True, text=True)
|
||||
@@ -102,15 +112,25 @@ class Runner:
|
||||
if 'ps' in args:
|
||||
if args[:2] == ['docker', 'ps']:
|
||||
return ''
|
||||
if args[-1] in ('backend', 'frontend', 'db', 'minio'):
|
||||
if args[-1] in ('backend', 'frontend', 'db', 'minio') or args[-1] in self.matrix_containers:
|
||||
return args[-1]
|
||||
return 'backend\nfrontend\ndb\nminio\ntraefik\nnode-exporter\ncreatebuckets\n'
|
||||
return 'backend\nfrontend\ndb\nminio\ntraefik\nnode-exporter\ncreatebuckets\n' + '\n'.join(self.matrix_containers)
|
||||
if 'stop' in args or 'start' in args:
|
||||
for name, container in self.matrix_containers.items():
|
||||
if name in args:
|
||||
container['State']['Running'] = 'start' in args
|
||||
if 'stop' in args and self.on_stop:
|
||||
self.on_stop()
|
||||
if 'pull' in args and self.on_pull:
|
||||
self.on_pull()
|
||||
if any('pg_database_size' in part for part in args):
|
||||
return '1000'
|
||||
if args[0] == 'du':
|
||||
return '1000\tminio'
|
||||
if args[0] == 'tar' and self.real_tar:
|
||||
with open(output, 'wb') if output else open(os.devnull, 'wb') as destination:
|
||||
subprocess.run(args, stdout=destination, check=True)
|
||||
return ''
|
||||
if output:
|
||||
Path(output).write_bytes(b'PGDMP fake dump')
|
||||
if args[0] == 'tar' and '-cpf' in args:
|
||||
@@ -153,6 +173,196 @@ class UpdateTests(unittest.TestCase):
|
||||
self.assertEqual((self.root / 'docker-compose.yml').read_bytes(), self.original_compose)
|
||||
self.assertFalse(self.called('stop'))
|
||||
|
||||
def add_legacy_matrix(self):
|
||||
services = {}
|
||||
for name in sorted(u.LEGACY_MATRIX_SERVICES):
|
||||
services[name] = {'image': 'example.test/' + name + ':legacy', 'restart': 'unless-stopped',
|
||||
'container_name': 'fedeo-' + name, 'networks': ['internal']}
|
||||
self.runner.matrix_containers[name] = {
|
||||
'Id': name, 'Image': 'image-' + name,
|
||||
'Config': {'Labels': {'com.docker.compose.service': name}}, 'Mounts': [],
|
||||
'HostConfig': {'RestartPolicy': {'Name': 'unless-stopped', 'MaximumRetryCount': 0}},
|
||||
'State': {'Running': True, 'Health': {'Status': 'healthy'}}}
|
||||
for name, directory, target in [('matrix-db', 'postgres', '/var/lib/postgresql/data'),
|
||||
('matrix-synapse', 'synapse', '/data')]:
|
||||
source = self.root / 'matrix' / directory
|
||||
source.mkdir(parents=True)
|
||||
(source / 'data').write_text('preserved ' + name)
|
||||
services[name]['volumes'] = [{'type': 'bind', 'source': str(source), 'target': target}]
|
||||
self.runner.matrix_containers[name]['Mounts'] = [{'Type': 'bind', 'Source': str(source), 'Destination': target}]
|
||||
# Use the real Compose parser to model an older base file, including the removed dependency.
|
||||
config = json.loads(subprocess.check_output(['docker', 'compose', '--project-directory', str(self.root),
|
||||
'--env-file', str(self.root / '.env'), '-f', str(self.root / 'docker-compose.yml'),
|
||||
'config', '--format', 'json'], env={'PATH': os.environ['PATH'], 'HOME': os.environ['HOME']}))
|
||||
config['services'].update(services)
|
||||
config['services']['backend']['depends_on']['matrix-synapse'] = {'condition': 'service_healthy'}
|
||||
(self.root / 'docker-compose.yml').write_bytes(data(config))
|
||||
self.original_compose = (self.root / 'docker-compose.yml').read_bytes()
|
||||
|
||||
def test_check_plans_matrix_retirement_without_stopping_or_disabling_it(self):
|
||||
self.add_legacy_matrix()
|
||||
result = self.updater.execute('check', '3.1.0')
|
||||
self.assertEqual(set(result['plan']['retiredServices']), u.LEGACY_MATRIX_SERVICES)
|
||||
self.assertEqual(result['plan']['retainedDataPaths'], ['matrix'])
|
||||
self.assertEqual(result['plan']['estimatedBackupBytes'], 4000 + u.GIB)
|
||||
self.assertFalse(self.called('--restart=no'))
|
||||
self.assert_untouched()
|
||||
|
||||
def test_matrix_is_backed_up_then_retired_without_removing_containers_or_data(self):
|
||||
self.add_legacy_matrix()
|
||||
# An image-created Redis volume must remain attached to the retained container.
|
||||
self.runner.matrix_containers['matrix-redis']['Mounts'] = [
|
||||
{'Type': 'volume', 'Source': '/var/lib/docker/volumes/redis/_data', 'Destination': '/data', 'Name': 'redis'}]
|
||||
result = self.run_update()
|
||||
self.assertEqual(result['status'], 'succeeded')
|
||||
self.assertNotIn(b'matrix-', (self.root / 'docker-compose.yml').read_bytes())
|
||||
self.assertTrue((self.root / 'matrix/postgres/data').exists())
|
||||
self.assertTrue((self.root / 'matrix/synapse/data').exists())
|
||||
backup = Path(result['backup'])
|
||||
self.assertEqual((backup / 'config/docker-compose.yml').read_bytes(), self.original_compose)
|
||||
metadata = u.read_json(backup / 'complete.json')
|
||||
self.assertEqual(metadata['matrixDataPaths'], ['matrix'])
|
||||
self.assertEqual(set(metadata['retainedContainerIds']), u.LEGACY_MATRIX_SERVICES)
|
||||
original = u.read_json(backup / 'containers.json')
|
||||
self.assertEqual(next(c for c in original if c.get('Id') == 'matrix-db')['HostConfig']['RestartPolicy']['Name'], 'unless-stopped')
|
||||
for container in self.runner.matrix_containers.values():
|
||||
self.assertFalse(container['State']['Running'])
|
||||
self.assertEqual(container['HostConfig']['RestartPolicy']['Name'], 'no')
|
||||
calls = self.runner.calls
|
||||
archive = next(i for i, a in enumerate(calls) if a[0] == 'tar' and '-cpf' in a)
|
||||
retire = next(i for i, a in enumerate(calls) if '--restart=no' in a)
|
||||
self.assertIn('matrix', calls[archive])
|
||||
self.assertLess(archive, retire)
|
||||
self.assertFalse(any('--remove-orphans' in a or a[:2] in (['docker', 'rm'], ['docker', 'volume']) for a in calls))
|
||||
# Future updates see stopped Matrix orphans; they do not bring them back.
|
||||
(self.root / '.env').write_bytes(self.original_env)
|
||||
self.assertEqual(self.run_update()['status'], 'succeeded')
|
||||
self.assertFalse(any(c['State']['Running'] for c in self.runner.matrix_containers.values()))
|
||||
|
||||
def test_matrix_archive_failure_preserves_base_and_restarts_original_stack(self):
|
||||
self.add_legacy_matrix()
|
||||
self.runner.fail = lambda a: a[0] == 'tar' and '-cpf' in a
|
||||
with self.assertRaises(u.UpdateError):
|
||||
self.run_update()
|
||||
self.assertFalse(self.called('--restart=no'))
|
||||
self.assertEqual((self.root / 'docker-compose.yml').read_bytes(), self.original_compose)
|
||||
self.assertFalse(self.updater.status()['recoveryRequired'])
|
||||
self.assertTrue(all(c['State']['Running'] for c in self.runner.matrix_containers.values()))
|
||||
|
||||
def test_real_matrix_archive_contains_database_media_and_configuration(self):
|
||||
self.add_legacy_matrix()
|
||||
self.runner.real_tar = True
|
||||
(self.root / 'matrix/livekit.yaml').write_text('preserved config')
|
||||
(self.root / 'matrix/postgres/pg_wal').mkdir()
|
||||
(self.root / 'matrix/postgres/pg_wal/segment').write_bytes(b'preserved WAL')
|
||||
expected = {str(path.relative_to(self.root)): path.read_bytes()
|
||||
for path in (self.root / 'matrix').rglob('*') if path.is_file()}
|
||||
result = self.run_update()
|
||||
backup = Path(result['backup'])
|
||||
with tarfile.open(backup / 'files.tar') as archive:
|
||||
for name, contents in expected.items():
|
||||
self.assertEqual(archive.extractfile(name).read(), contents)
|
||||
self.assertEqual(u.sha((backup / 'files.tar').read_bytes()), u.read_json(backup / 'complete.json')['sha256']['files.tar'])
|
||||
|
||||
def test_matrix_must_be_stopped_before_a_cold_backup(self):
|
||||
self.add_legacy_matrix()
|
||||
self.runner.on_stop = lambda: self.runner.matrix_containers['matrix-db']['State'].update(Running=True)
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Matrix-Container läuft noch'):
|
||||
self.run_update()
|
||||
self.assertFalse(self.called('-cpf'))
|
||||
self.assertFalse(self.called('--restart=no'))
|
||||
|
||||
def test_interrupted_retirement_requires_recovery_and_preserves_backup(self):
|
||||
self.add_legacy_matrix()
|
||||
self.runner.interrupt = lambda a: '--restart=no' in a and a[-1] == 'matrix-element'
|
||||
with self.assertRaises(KeyboardInterrupt):
|
||||
self.run_update()
|
||||
state = self.updater.status()
|
||||
self.assertEqual(state['phase'], 'retiring_matrix')
|
||||
self.assertTrue(state['recoveryRequired'])
|
||||
self.assertTrue((Path(state['backup']) / 'complete.json').exists())
|
||||
self.assertEqual((self.root / 'docker-compose.yml').read_bytes(), self.original_compose)
|
||||
self.assertFalse(any(c['State']['Running'] for c in self.runner.matrix_containers.values()))
|
||||
|
||||
def test_matrix_symlinks_and_nonstandard_mounts_are_rejected_before_stopping(self):
|
||||
self.add_legacy_matrix()
|
||||
link = self.root / 'matrix/postgres/pg_wal'
|
||||
link.symlink_to(self.root / 'postgres', target_is_directory=True)
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Symbolische Links'):
|
||||
self.run_update()
|
||||
self.assert_untouched()
|
||||
link.unlink()
|
||||
self.runner.matrix_containers['matrix-db']['Mounts'][0]['Source'] = '/different/postgres'
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Tatsächliche Matrix-Datenpfade'):
|
||||
self.run_update()
|
||||
self.assert_untouched()
|
||||
|
||||
def test_unknown_additional_services_are_still_rejected(self):
|
||||
self.add_legacy_matrix()
|
||||
config = json.loads(self.original_compose)
|
||||
config['services']['matrix-custom'] = {'image': 'example.test/custom:latest'}
|
||||
(self.root / 'docker-compose.yml').write_bytes(data(config))
|
||||
with self.assertRaisesRegex(u.UpdateError, 'matrix-custom'):
|
||||
self.run_update()
|
||||
self.assertFalse(self.called('stop'))
|
||||
|
||||
def test_matrix_cannot_be_reintroduced_by_an_override(self):
|
||||
self.add_legacy_matrix()
|
||||
(self.root / 'docker-compose.override.yml').write_text('services:\n matrix-element:\n image: example.test/element:legacy\n')
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Zielkonfiguration.*matrix-element'):
|
||||
self.run_update()
|
||||
self.assert_untouched()
|
||||
|
||||
def test_nonstandard_matrix_bind_and_named_volume_are_rejected(self):
|
||||
self.add_legacy_matrix()
|
||||
for volume in [{'type': 'bind', 'source': str(self.root / 'postgres'), 'target': '/data'},
|
||||
{'type': 'volume', 'source': 'matrixdata', 'target': '/data'}]:
|
||||
with self.subTest(volume=volume['type']):
|
||||
config = json.loads(self.original_compose)
|
||||
config['services']['matrix-synapse']['volumes'] = [volume]
|
||||
config['volumes'] = {'matrixdata': {}}
|
||||
(self.root / 'docker-compose.yml').write_bytes(data(config))
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Matrix-Daten müssen'):
|
||||
self.run_update()
|
||||
self.assertFalse(self.called('stop'))
|
||||
|
||||
def test_resolve_target_requires_matrix_to_remain_retired(self):
|
||||
self.add_legacy_matrix()
|
||||
self.runner.fail = lambda a: 'migrate' in a
|
||||
with self.assertRaises(u.UpdateError):
|
||||
self.run_update()
|
||||
self.runner.fail = None
|
||||
self.runner.matrix_containers['matrix-db']['HostConfig']['RestartPolicy']['Name'] = 'always'
|
||||
with self.assertRaisesRegex(u.UpdateError, 'automatisch neu starten'):
|
||||
self.updater.resolve('3.1.0')
|
||||
self.assertTrue(self.updater.status()['recoveryRequired'])
|
||||
self.runner.matrix_containers['matrix-db']['HostConfig']['RestartPolicy']['Name'] = 'no'
|
||||
self.assertEqual(self.updater.resolve('3.1.0')['status'], 'recovered')
|
||||
|
||||
def test_resolve_rollback_requires_original_matrix_images_state_and_restart_policy(self):
|
||||
self.add_legacy_matrix()
|
||||
originals = copy.deepcopy(self.runner.matrix_containers)
|
||||
self.runner.fail = lambda a: 'migrate' in a
|
||||
with self.assertRaises(u.UpdateError):
|
||||
self.run_update()
|
||||
self.runner.fail = None
|
||||
self.http.reported_version = '3.0.0'
|
||||
(self.root / 'docker-compose.yml').write_bytes(self.original_compose)
|
||||
(self.root / '.env').write_bytes(self.original_env)
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Matrix-Container und Neustartrichtlinien'):
|
||||
self.updater.resolve('3.0.0')
|
||||
self.assertTrue(self.updater.status()['recoveryRequired'])
|
||||
self.runner.matrix_containers = copy.deepcopy(originals)
|
||||
for key, value in [('Image', 'wrong-image'), ('HostConfig', {'RestartPolicy': {'Name': 'no'}}),
|
||||
('State', {'Running': True, 'Health': {'Status': 'unhealthy'}})]:
|
||||
with self.subTest(key=key):
|
||||
self.runner.matrix_containers['matrix-db'][key] = value
|
||||
with self.assertRaisesRegex(u.UpdateError, 'Matrix-Container und Neustartrichtlinien'):
|
||||
self.updater.resolve('3.0.0')
|
||||
self.runner.matrix_containers = copy.deepcopy(originals)
|
||||
self.assertEqual(self.updater.resolve('3.0.0')['status'], 'recovered')
|
||||
self.assertFalse(self.updater.status()['recoveryRequired'])
|
||||
|
||||
def test_success_pins_images_preserves_configuration_and_checks_both_public_services(self):
|
||||
result = self.run_update()
|
||||
self.assertEqual(result['status'], 'succeeded')
|
||||
|
||||
Reference in New Issue
Block a user