11 import arvados_testutil as tutil
12 import run_test_server
14 class KeepTestCase(run_test_server.TestCaseWithServers):
20 super(KeepTestCase, cls).setUpClass()
21 run_test_server.authorize_with("admin")
22 cls.api_client = arvados.api('v1')
23 cls.keep_client = arvados.KeepClient(api_client=cls.api_client,
24 proxy='', local_store='')
26 def test_KeepBasicRWTest(self):
27 foo_locator = self.keep_client.put('foo')
28 self.assertRegexpMatches(
30 '^acbd18db4cc2f85cedef654fccc4a4d8\+3',
31 'wrong md5 hash from Keep.put("foo"): ' + foo_locator)
32 self.assertEqual(self.keep_client.get(foo_locator),
34 'wrong content from Keep.get(md5("foo"))')
36 def test_KeepBinaryRWTest(self):
37 blob_str = '\xff\xfe\xf7\x00\x01\x02'
38 blob_locator = self.keep_client.put(blob_str)
39 self.assertRegexpMatches(
41 '^7fc7c53b45e53926ba52821140fef396\+6',
42 ('wrong locator from Keep.put(<binarydata>):' + blob_locator))
43 self.assertEqual(self.keep_client.get(blob_locator),
45 'wrong content from Keep.get(md5(<binarydata>))')
47 def test_KeepLongBinaryRWTest(self):
48 blob_str = '\xff\xfe\xfd\xfc\x00\x01\x02\x03'
50 blob_str = blob_str + blob_str
51 blob_locator = self.keep_client.put(blob_str)
52 self.assertRegexpMatches(
54 '^84d90fc0d8175dd5dcfab04b999bc956\+67108864',
55 ('wrong locator from Keep.put(<binarydata>): ' + blob_locator))
56 self.assertEqual(self.keep_client.get(blob_locator),
58 'wrong content from Keep.get(md5(<binarydata>))')
60 def test_KeepSingleCopyRWTest(self):
61 blob_str = '\xff\xfe\xfd\xfc\x00\x01\x02\x03'
62 blob_locator = self.keep_client.put(blob_str, copies=1)
63 self.assertRegexpMatches(
65 '^c902006bc98a3eb4a3663b65ab4a6fab\+8',
66 ('wrong locator from Keep.put(<binarydata>): ' + blob_locator))
67 self.assertEqual(self.keep_client.get(blob_locator),
69 'wrong content from Keep.get(md5(<binarydata>))')
71 def test_KeepEmptyCollectionTest(self):
72 blob_locator = self.keep_client.put('', copies=1)
73 self.assertRegexpMatches(
75 '^d41d8cd98f00b204e9800998ecf8427e\+0',
76 ('wrong locator from Keep.put(""): ' + blob_locator))
79 class KeepPermissionTestCase(run_test_server.TestCaseWithServers):
81 KEEP_SERVER = {'blob_signing_key': 'abcdefghijk0123456789',
82 'enforce_permissions': True}
84 def test_KeepBasicRWTest(self):
85 run_test_server.authorize_with('active')
86 keep_client = arvados.KeepClient()
87 foo_locator = keep_client.put('foo')
88 self.assertRegexpMatches(
90 r'^acbd18db4cc2f85cedef654fccc4a4d8\+3\+A[a-f0-9]+@[a-f0-9]+$',
91 'invalid locator from Keep.put("foo"): ' + foo_locator)
92 self.assertEqual(keep_client.get(foo_locator),
94 'wrong content from Keep.get(md5("foo"))')
96 # GET with an unsigned locator => NotFound
97 bar_locator = keep_client.put('bar')
98 unsigned_bar_locator = "37b51d194a7513e45b56f6524f2d51f2+3"
99 self.assertRegexpMatches(
101 r'^37b51d194a7513e45b56f6524f2d51f2\+3\+A[a-f0-9]+@[a-f0-9]+$',
102 'invalid locator from Keep.put("bar"): ' + bar_locator)
103 self.assertRaises(arvados.errors.NotFoundError,
105 unsigned_bar_locator)
107 # GET from a different user => NotFound
108 run_test_server.authorize_with('spectator')
109 self.assertRaises(arvados.errors.NotFoundError,
113 # Unauthenticated GET for a signed locator => NotFound
114 # Unauthenticated GET for an unsigned locator => NotFound
115 keep_client.api_token = ''
116 self.assertRaises(arvados.errors.NotFoundError,
119 self.assertRaises(arvados.errors.NotFoundError,
121 unsigned_bar_locator)
124 # KeepOptionalPermission: starts Keep with --permission-key-file
125 # but not --enforce-permissions (i.e. generate signatures on PUT
126 # requests, but do not require them for GET requests)
128 # All of these requests should succeed when permissions are optional:
129 # * authenticated request, signed locator
130 # * authenticated request, unsigned locator
131 # * unauthenticated request, signed locator
132 # * unauthenticated request, unsigned locator
133 class KeepOptionalPermission(run_test_server.TestCaseWithServers):
135 KEEP_SERVER = {'blob_signing_key': 'abcdefghijk0123456789',
136 'enforce_permissions': False}
140 super(KeepOptionalPermission, cls).setUpClass()
141 run_test_server.authorize_with("admin")
142 cls.api_client = arvados.api('v1')
145 super(KeepOptionalPermission, self).setUp()
146 self.keep_client = arvados.KeepClient(api_client=self.api_client,
147 proxy='', local_store='')
149 def _put_foo_and_check(self):
150 signed_locator = self.keep_client.put('foo')
151 self.assertRegexpMatches(
153 r'^acbd18db4cc2f85cedef654fccc4a4d8\+3\+A[a-f0-9]+@[a-f0-9]+$',
154 'invalid locator from Keep.put("foo"): ' + signed_locator)
155 return signed_locator
157 def test_KeepAuthenticatedSignedTest(self):
158 signed_locator = self._put_foo_and_check()
159 self.assertEqual(self.keep_client.get(signed_locator),
161 'wrong content from Keep.get(md5("foo"))')
163 def test_KeepAuthenticatedUnsignedTest(self):
164 signed_locator = self._put_foo_and_check()
165 self.assertEqual(self.keep_client.get("acbd18db4cc2f85cedef654fccc4a4d8"),
167 'wrong content from Keep.get(md5("foo"))')
169 def test_KeepUnauthenticatedSignedTest(self):
170 # Check that signed GET requests work even when permissions
171 # enforcement is off.
172 signed_locator = self._put_foo_and_check()
173 self.keep_client.api_token = ''
174 self.assertEqual(self.keep_client.get(signed_locator),
176 'wrong content from Keep.get(md5("foo"))')
178 def test_KeepUnauthenticatedUnsignedTest(self):
179 # Since --enforce-permissions is not in effect, GET requests
180 # need not be authenticated.
181 signed_locator = self._put_foo_and_check()
182 self.keep_client.api_token = ''
183 self.assertEqual(self.keep_client.get("acbd18db4cc2f85cedef654fccc4a4d8"),
185 'wrong content from Keep.get(md5("foo"))')
188 class KeepProxyTestCase(run_test_server.TestCaseWithServers):
191 KEEP_PROXY_SERVER = {'auth': 'admin'}
195 super(KeepProxyTestCase, cls).setUpClass()
196 cls.api_client = arvados.api('v1')
199 arvados.config.settings().pop('ARVADOS_EXTERNAL_CLIENT', None)
200 super(KeepProxyTestCase, self).tearDown()
202 def test_KeepProxyTest1(self):
203 # Will use ARVADOS_KEEP_PROXY environment variable that is set by
205 keep_client = arvados.KeepClient(api_client=self.api_client,
207 baz_locator = keep_client.put('baz')
208 self.assertRegexpMatches(
210 '^73feffa4b7f6bb68e44cf984c85f6e88\+3',
211 'wrong md5 hash from Keep.put("baz"): ' + baz_locator)
212 self.assertEqual(keep_client.get(baz_locator),
214 'wrong content from Keep.get(md5("baz"))')
215 self.assertTrue(keep_client.using_proxy)
217 def test_KeepProxyTest2(self):
218 # Don't instantiate the proxy directly, but set the X-External-Client
219 # header. The API server should direct us to the proxy.
220 arvados.config.settings()['ARVADOS_EXTERNAL_CLIENT'] = 'true'
221 keep_client = arvados.KeepClient(api_client=self.api_client,
222 proxy='', local_store='')
223 baz_locator = keep_client.put('baz2')
224 self.assertRegexpMatches(
226 '^91f372a266fe2bf2823cb8ec7fda31ce\+4',
227 'wrong md5 hash from Keep.put("baz2"): ' + baz_locator)
228 self.assertEqual(keep_client.get(baz_locator),
230 'wrong content from Keep.get(md5("baz2"))')
231 self.assertTrue(keep_client.using_proxy)
234 class KeepClientServiceTestCase(unittest.TestCase):
235 def mock_keep_services(self, *services):
236 api_client = mock.MagicMock(name='api_client')
237 api_client.keep_services().accessible().execute.return_value = {
238 'items_available': len(services),
240 'uuid': 'zzzzz-bi6l4-{:015x}'.format(index),
241 'owner_uuid': 'zzzzz-tpzed-000000000000000',
242 'service_host': host,
243 'service_port': port,
244 'service_ssl_flag': ssl,
245 'service_type': servtype,
246 } for index, (host, port, ssl, servtype)
247 in enumerate(services)],
251 def mock_n_keep_disks(self, service_count):
252 return self.mock_keep_services(
253 *[("keep0x{:x}".format(index), 80, False, 'disk')
254 for index in range(service_count)])
256 def get_service_roots(self, *services):
257 api_client = self.mock_keep_services(*services)
258 keep_client = arvados.KeepClient(api_client=api_client)
259 services = keep_client.weighted_service_roots('000000')
260 return [urlparse.urlparse(url) for url in sorted(services)]
262 def test_ssl_flag_respected_in_roots(self):
263 services = self.get_service_roots(('keep', 10, False, 'disk'),
264 ('keep', 20, True, 'disk'))
265 self.assertEqual(10, services[0].port)
266 self.assertEqual('http', services[0].scheme)
267 self.assertEqual(20, services[1].port)
268 self.assertEqual('https', services[1].scheme)
270 def test_correct_ports_with_ipv6_addresses(self):
271 service = self.get_service_roots(('100::1', 10, True, 'proxy'))[0]
272 self.assertEqual('100::1', service.hostname)
273 self.assertEqual(10, service.port)
275 # test_get_timeout and test_put_timeout test that
276 # KeepClient.get and KeepClient.put use the appropriate timeouts
277 # when connected directly to a Keep server (i.e. non-proxy timeout)
279 def test_get_timeout(self):
280 api_client = self.mock_keep_services(('keep', 10, False, 'disk'))
281 force_timeout = [socket.timeout("timed out")]
282 with tutil.mock_get(force_timeout) as mock_session:
283 keep_client = arvados.KeepClient(api_client=api_client)
284 with self.assertRaises(arvados.errors.KeepReadError):
285 keep_client.get('ffffffffffffffffffffffffffffffff')
286 self.assertTrue(mock_session.return_value.get.called)
288 arvados.KeepClient.DEFAULT_TIMEOUT,
289 mock_session.return_value.get.call_args[1]['timeout'])
291 def test_put_timeout(self):
292 api_client = self.mock_keep_services(('keep', 10, False, 'disk'))
293 force_timeout = [socket.timeout("timed out")]
294 with tutil.mock_put(force_timeout) as mock_session:
295 keep_client = arvados.KeepClient(api_client=api_client)
296 with self.assertRaises(arvados.errors.KeepWriteError):
297 keep_client.put('foo')
298 self.assertTrue(mock_session.return_value.put.called)
300 arvados.KeepClient.DEFAULT_TIMEOUT,
301 mock_session.return_value.put.call_args[1]['timeout'])
303 def test_proxy_get_timeout(self):
304 # Force a timeout, verifying that the requests.get or
305 # requests.put method was called with the proxy_timeout
306 # setting rather than the default timeout.
307 api_client = self.mock_keep_services(('keep', 10, False, 'proxy'))
308 force_timeout = [socket.timeout("timed out")]
309 with tutil.mock_get(force_timeout) as mock_session:
310 keep_client = arvados.KeepClient(api_client=api_client)
311 with self.assertRaises(arvados.errors.KeepReadError):
312 keep_client.get('ffffffffffffffffffffffffffffffff')
313 self.assertTrue(mock_session.return_value.get.called)
315 arvados.KeepClient.DEFAULT_PROXY_TIMEOUT,
316 mock_session.return_value.get.call_args[1]['timeout'])
318 def test_proxy_put_timeout(self):
319 # Force a timeout, verifying that the requests.get or
320 # requests.put method was called with the proxy_timeout
321 # setting rather than the default timeout.
322 api_client = self.mock_keep_services(('keep', 10, False, 'proxy'))
323 force_timeout = [socket.timeout("timed out")]
324 with tutil.mock_put(force_timeout) as mock_session:
325 keep_client = arvados.KeepClient(api_client=api_client)
326 with self.assertRaises(arvados.errors.KeepWriteError):
327 keep_client.put('foo')
328 self.assertTrue(mock_session.return_value.put.called)
330 arvados.KeepClient.DEFAULT_PROXY_TIMEOUT,
331 mock_session.return_value.put.call_args[1]['timeout'])
333 def test_probe_order_reference_set(self):
334 # expected_order[i] is the probe order for
335 # hash=md5(sprintf("%064x",i)) where there are 16 services
336 # with uuid sprintf("anything-%015x",j) with j in 0..15. E.g.,
337 # the first probe for the block consisting of 64 "0"
338 # characters is the service whose uuid is
339 # "zzzzz-bi6l4-000000000000003", so expected_order[0][0]=='3'.
341 list('3eab2d5fc9681074'),
342 list('097dba52e648f1c3'),
343 list('c5b4e023f8a7d691'),
344 list('9d81c02e76a3bf54'),
347 hashlib.md5("{:064x}".format(x)).hexdigest()
348 for x in range(len(expected_order))]
349 api_client = self.mock_n_keep_disks(16)
350 keep_client = arvados.KeepClient(api_client=api_client)
351 for i, hash in enumerate(hashes):
352 roots = keep_client.weighted_service_roots(hash)
354 re.search(r'//\[?keep0x([0-9a-f]+)', root).group(1)
356 self.assertEqual(expected_order[i], got_order)
358 def test_probe_waste_adding_one_server(self):
360 hashlib.md5("{:064x}".format(x)).hexdigest() for x in range(100)]
361 initial_services = 12
362 api_client = self.mock_n_keep_disks(initial_services)
363 keep_client = arvados.KeepClient(api_client=api_client)
365 keep_client.weighted_service_roots(hash) for hash in hashes]
366 for added_services in range(1, 12):
367 api_client = self.mock_n_keep_disks(initial_services+added_services)
368 keep_client = arvados.KeepClient(api_client=api_client)
370 for hash_index in range(len(hashes)):
371 probe_after = keep_client.weighted_service_roots(
373 penalty = probe_after.index(probes_before[hash_index][0])
374 self.assertLessEqual(penalty, added_services)
375 total_penalty += penalty
376 # Average penalty per block should not exceed
377 # N(added)/N(orig) by more than 20%, and should get closer
378 # to the ideal as we add data points.
381 len(hashes) / initial_services)
384 (120 - added_services)/100)
386 expect_penalty * 8/10)
388 min_penalty <= total_penalty <= max_penalty,
389 "With {}+{} services, {} blocks, penalty {} but expected {}..{}".format(
397 def check_64_zeros_error_order(self, verb, exc_class):
400 data = hashlib.md5(data).hexdigest() + '+1234'
401 api_client = self.mock_n_keep_disks(16)
402 keep_client = arvados.KeepClient(api_client=api_client)
403 with mock.patch('requests.' + verb,
404 side_effect=socket.timeout) as req_mock, \
405 self.assertRaises(exc_class) as err_check:
406 getattr(keep_client, verb)(data)
407 urls = [urlparse.urlparse(url)
408 for url in err_check.exception.service_errors()]
409 self.assertEqual([('keep0x' + c, 80) for c in '3eab2d5fc9681074'],
410 [(url.hostname, url.port) for url in urls])
412 def test_get_error_shows_probe_order(self):
413 self.check_64_zeros_error_order('get', arvados.errors.KeepReadError)
415 def test_put_error_shows_probe_order(self):
416 self.check_64_zeros_error_order('put', arvados.errors.KeepWriteError)
418 def check_no_services_error(self, verb, exc_class):
419 api_client = mock.MagicMock(name='api_client')
420 api_client.keep_services().accessible().execute.side_effect = (
421 arvados.errors.ApiError)
422 keep_client = arvados.KeepClient(api_client=api_client)
423 with self.assertRaises(exc_class) as err_check:
424 getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0')
425 self.assertEqual(0, len(err_check.exception.service_errors()))
427 def test_get_error_with_no_services(self):
428 self.check_no_services_error('get', arvados.errors.KeepReadError)
430 def test_put_error_with_no_services(self):
431 self.check_no_services_error('put', arvados.errors.KeepWriteError)
433 def check_errors_from_last_retry(self, verb, exc_class):
434 api_client = self.mock_n_keep_disks(2)
435 req_mock = getattr(tutil, 'mock_{}_responses'.format(verb))(
436 "retry error reporting test", 500, 500, 403, 403)
437 with req_mock, tutil.skip_sleep, \
438 self.assertRaises(exc_class) as err_check:
439 keep_client = arvados.KeepClient(api_client=api_client)
440 getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0',
442 self.assertEqual([403, 403], [
443 getattr(error, 'status_code', None)
444 for error in err_check.exception.service_errors().itervalues()])
446 def test_get_error_reflects_last_retry(self):
447 self.check_errors_from_last_retry('get', arvados.errors.KeepReadError)
449 def test_put_error_reflects_last_retry(self):
450 self.check_errors_from_last_retry('put', arvados.errors.KeepWriteError)
452 def test_put_error_does_not_include_successful_puts(self):
453 data = 'partial failure test'
454 data_loc = '{}+{}'.format(hashlib.md5(data).hexdigest(), len(data))
455 api_client = self.mock_n_keep_disks(3)
456 with tutil.mock_put_responses(data_loc, 200, 500, 500) as req_mock, \
457 self.assertRaises(arvados.errors.KeepWriteError) as exc_check:
458 keep_client = arvados.KeepClient(api_client=api_client)
459 keep_client.put(data)
460 self.assertEqual(2, len(exc_check.exception.service_errors()))
463 class KeepClientRetryTestMixin(object):
464 # Testing with a local Keep store won't exercise the retry behavior.
465 # Instead, our strategy is:
466 # * Create a client with one proxy specified (pointed at a black
467 # hole), so there's no need to instantiate an API client, and
468 # all HTTP requests come from one place.
469 # * Mock httplib's request method to provide simulated responses.
470 # This lets us test the retry logic extensively without relying on any
471 # supporting servers, and prevents side effects in case something hiccups.
472 # To use this mixin, define DEFAULT_EXPECT, DEFAULT_EXCEPTION, and
475 # Test classes must define TEST_PATCHER to a method that mocks
476 # out appropriate methods in the client.
478 PROXY_ADDR = 'http://[%s]:65535/' % (tutil.TEST_HOST,)
479 TEST_DATA = 'testdata'
480 TEST_LOCATOR = 'ef654c40ab4f1747fc699915d4f70902+8'
483 self.client_kwargs = {'proxy': self.PROXY_ADDR, 'local_store': ''}
485 def new_client(self, **caller_kwargs):
486 kwargs = self.client_kwargs.copy()
487 kwargs.update(caller_kwargs)
488 return arvados.KeepClient(**kwargs)
490 def run_method(self, *args, **kwargs):
491 raise NotImplementedError("test subclasses must define run_method")
493 def check_success(self, expected=None, *args, **kwargs):
495 expected = self.DEFAULT_EXPECT
496 self.assertEqual(expected, self.run_method(*args, **kwargs))
498 def check_exception(self, error_class=None, *args, **kwargs):
499 if error_class is None:
500 error_class = self.DEFAULT_EXCEPTION
501 self.assertRaises(error_class, self.run_method, *args, **kwargs)
503 def test_immediate_success(self):
504 with self.TEST_PATCHER(self.DEFAULT_EXPECT, 200):
507 def test_retry_then_success(self):
508 with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
509 self.check_success(num_retries=3)
511 def test_no_default_retry(self):
512 with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
513 self.check_exception()
515 def test_no_retry_after_permanent_error(self):
516 with self.TEST_PATCHER(self.DEFAULT_EXPECT, 403, 200):
517 self.check_exception(num_retries=3)
519 def test_error_after_retries_exhausted(self):
520 with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 500, 200):
521 self.check_exception(num_retries=1)
523 def test_num_retries_instance_fallback(self):
524 self.client_kwargs['num_retries'] = 3
525 with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
530 class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase):
531 DEFAULT_EXPECT = KeepClientRetryTestMixin.TEST_DATA
532 DEFAULT_EXCEPTION = arvados.errors.KeepReadError
533 HINTED_LOCATOR = KeepClientRetryTestMixin.TEST_LOCATOR + '+K@xyzzy'
534 TEST_PATCHER = staticmethod(tutil.mock_get_responses)
536 def run_method(self, locator=KeepClientRetryTestMixin.TEST_LOCATOR,
538 return self.new_client().get(locator, *args, **kwargs)
540 def test_specific_exception_when_not_found(self):
541 with tutil.mock_get_responses(self.DEFAULT_EXPECT, 404, 200):
542 self.check_exception(arvados.errors.NotFoundError, num_retries=3)
544 def test_general_exception_with_mixed_errors(self):
545 # get should raise a NotFoundError if no server returns the block,
546 # and a high threshold of servers report that it's not found.
547 # This test rigs up 50/50 disagreement between two servers, and
548 # checks that it does not become a NotFoundError.
549 client = self.new_client()
550 with tutil.mock_get_responses(self.DEFAULT_EXPECT, 404, 500):
551 with self.assertRaises(arvados.errors.KeepReadError) as exc_check:
552 client.get(self.HINTED_LOCATOR)
553 self.assertNotIsInstance(
554 exc_check.exception, arvados.errors.NotFoundError,
555 "mixed errors raised NotFoundError")
557 def test_hint_server_can_succeed_without_retries(self):
558 with tutil.mock_get_responses(self.DEFAULT_EXPECT, 404, 200, 500):
559 self.check_success(locator=self.HINTED_LOCATOR)
561 def test_try_next_server_after_timeout(self):
562 with tutil.mock_get([
563 socket.timeout("timed out"),
564 tutil.fake_requests_response(200, self.DEFAULT_EXPECT)]):
565 self.check_success(locator=self.HINTED_LOCATOR)
567 def test_retry_data_with_wrong_checksum(self):
568 with tutil.mock_get((tutil.fake_requests_response(200, s) for s in ['baddata', self.TEST_DATA])):
569 self.check_success(locator=self.HINTED_LOCATOR)
573 class KeepClientRetryPutTestCase(KeepClientRetryTestMixin, unittest.TestCase):
574 DEFAULT_EXPECT = KeepClientRetryTestMixin.TEST_LOCATOR
575 DEFAULT_EXCEPTION = arvados.errors.KeepWriteError
576 TEST_PATCHER = staticmethod(tutil.mock_put_responses)
578 def run_method(self, data=KeepClientRetryTestMixin.TEST_DATA,
579 copies=1, *args, **kwargs):
580 return self.new_client().put(data, copies, *args, **kwargs)
582 def test_do_not_send_multiple_copies_to_same_server(self):
583 with tutil.mock_put_responses(self.DEFAULT_EXPECT, 200):
584 self.check_exception(copies=2, num_retries=3)