X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/47b0cb35b5ee933757c4342bb75fd286c1dac8cb..63b03a39adfd78961c5bbb6a3a2d02ccd8c92e4d:/sdk/python/tests/test_keep_client.py diff --git a/sdk/python/tests/test_keep_client.py b/sdk/python/tests/test_keep_client.py index 72bc741541..baae28e3d7 100644 --- a/sdk/python/tests/test_keep_client.py +++ b/sdk/python/tests/test_keep_client.py @@ -1,6 +1,7 @@ import hashlib import mock import os +import random import re import socket import unittest @@ -11,84 +12,6 @@ import arvados.retry import arvados_testutil as tutil import run_test_server -class KeepRendezvousWeightTestCase(unittest.TestCase): - def setUp(self): - self.keep_client = arvados.KeepClient( - api_client=mock.MagicMock(name='api_client'), - proxy='', local_store='') - self.keep_client._keep_services = [] - self.n_services = 0 - - def addServices(self, n): - for x in range(n): - uuid = "zzzzz-bi6l4-{:015x}".format(self.n_services) - uri = "https://keep0x{:x}.zzzzz.arvadosapi.com:25107/".format(self.n_services) - self.keep_client._keep_services.append( - {'uuid': uuid, '_service_root': uri}) - self.n_services += 1 - - def test_ProbeOrderReferenceSet(self): - # expected_order[i] is the probe order for - # hash=md5(sprintf("%064x",i)) where there are 16 services - # with uuid sprintf("anything-%015x",j) with j in 0..15. E.g., - # the first probe for the block consisting of 64 "0" - # characters is the service whose uuid is - # "zzzzz-bi6l4-000000000000003", so expected_order[0][0]=='3'. - expected_order = [ - list('3eab2d5fc9681074'), - list('097dba52e648f1c3'), - list('c5b4e023f8a7d691'), - list('9d81c02e76a3bf54'), - ] - hashes = [ - hashlib.md5("{:064x}".format(x)).hexdigest() - for x in range(len(expected_order))] - self.addServices(16) - for i, hash in enumerate(hashes): - roots = self.keep_client.weighted_service_roots(hash) - got_order = [ - re.search(r'//keep0x([0-9a-f]+)', root).group(1) - for root in roots] - self.assertEqual(got_order, expected_order[i]) - - def test_ProbeWasteAddingOneServer(self): - hashes = [ - hashlib.md5("{:064x}".format(x)).hexdigest() for x in range(100)] - initial_services = 12 - self.addServices(initial_services) - probes_before = [ - self.keep_client.weighted_service_roots(hash) for hash in hashes] - for added_services in range(1, 12): - self.addServices(1) - total_penalty = 0 - for hash_index in range(len(hashes)): - probe_after = self.keep_client.weighted_service_roots( - hashes[hash_index]) - penalty = probe_after.index(probes_before[hash_index][0]) - self.assertLessEqual(penalty, added_services) - total_penalty += penalty - # Average penalty per block should not exceed - # N(added)/N(orig) by more than 20%, and should get closer - # to the ideal as we add data points. - expect_penalty = ( - added_services * - len(hashes) / initial_services) - max_penalty = ( - expect_penalty * - (120 - added_services)/100) - min_penalty = ( - expect_penalty * 8/10) - self.assertTrue( - min_penalty <= total_penalty <= max_penalty, - "With {}+{} services, {} blocks, penalty {} but expected {}..{}".format( - initial_services, - added_services, - len(hashes), - total_penalty, - min_penalty, - max_penalty)) - - class KeepTestCase(run_test_server.TestCaseWithServers): MAIN_SERVER = {} KEEP_SERVER = {} @@ -153,6 +76,21 @@ class KeepTestCase(run_test_server.TestCaseWithServers): '^d41d8cd98f00b204e9800998ecf8427e\+0', ('wrong locator from Keep.put(""): ' + blob_locator)) + def test_unicode_must_be_ascii(self): + # If unicode type, must only consist of valid ASCII + foo_locator = self.keep_client.put(u'foo') + self.assertRegexpMatches( + foo_locator, + '^acbd18db4cc2f85cedef654fccc4a4d8\+3', + 'wrong md5 hash from Keep.put("foo"): ' + foo_locator) + + with self.assertRaises(UnicodeEncodeError): + # Error if it is not ASCII + self.keep_client.put(u'\xe2') + + with self.assertRaises(arvados.errors.ArgumentError): + # Must be a string type + self.keep_client.put({}) class KeepPermissionTestCase(run_test_server.TestCaseWithServers): MAIN_SERVER = {} @@ -266,11 +204,12 @@ class KeepOptionalPermission(run_test_server.TestCaseWithServers): class KeepProxyTestCase(run_test_server.TestCaseWithServers): MAIN_SERVER = {} KEEP_SERVER = {} - KEEP_PROXY_SERVER = {'auth': 'admin'} + KEEP_PROXY_SERVER = {} @classmethod def setUpClass(cls): super(KeepProxyTestCase, cls).setUpClass() + run_test_server.authorize_with('active') cls.api_client = arvados.api('v1') def tearDown(self): @@ -309,42 +248,214 @@ class KeepProxyTestCase(run_test_server.TestCaseWithServers): self.assertTrue(keep_client.using_proxy) -class KeepClientServiceTestCase(unittest.TestCase): - def mock_keep_services(self, *services): - api_client = mock.MagicMock(name='api_client') - api_client.keep_services().accessible().execute.return_value = { - 'items_available': len(services), - 'items': [{ - 'uuid': 'zzzzz-bi6l4-mockservice{:04x}'.format(index), - 'owner_uuid': 'zzzzz-tpzed-mockownerabcdef', - 'service_host': host, - 'service_port': port, - 'service_ssl_flag': ssl, - 'service_type': servtype, - } for index, (host, port, ssl, servtype) - in enumerate(services)], - } - return api_client - - def get_service_roots(self, *services): - api_client = self.mock_keep_services(*services) +class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock): + def get_service_roots(self, api_client): keep_client = arvados.KeepClient(api_client=api_client) services = keep_client.weighted_service_roots('000000') return [urlparse.urlparse(url) for url in sorted(services)] def test_ssl_flag_respected_in_roots(self): - services = self.get_service_roots(('keep', 10, False, 'disk'), - ('keep', 20, True, 'disk')) - self.assertEqual(10, services[0].port) - self.assertEqual('http', services[0].scheme) - self.assertEqual(20, services[1].port) - self.assertEqual('https', services[1].scheme) + for ssl_flag in [False, True]: + services = self.get_service_roots(self.mock_keep_services( + service_ssl_flag=ssl_flag)) + self.assertEqual( + ('https' if ssl_flag else 'http'), services[0].scheme) def test_correct_ports_with_ipv6_addresses(self): - service = self.get_service_roots(('100::1', 10, True, 'proxy'))[0] + service = self.get_service_roots(self.mock_keep_services( + service_type='proxy', service_host='100::1', service_port=10, count=1))[0] self.assertEqual('100::1', service.hostname) self.assertEqual(10, service.port) + # test_get_timeout and test_put_timeout test that + # KeepClient.get and KeepClient.put use the appropriate timeouts + # when connected directly to a Keep server (i.e. non-proxy timeout) + + def test_get_timeout(self): + api_client = self.mock_keep_services(count=1) + force_timeout = [socket.timeout("timed out")] + with tutil.mock_get(force_timeout) as mock_session: + keep_client = arvados.KeepClient(api_client=api_client) + with self.assertRaises(arvados.errors.KeepReadError): + keep_client.get('ffffffffffffffffffffffffffffffff') + self.assertTrue(mock_session.return_value.get.called) + self.assertEqual( + arvados.KeepClient.DEFAULT_TIMEOUT, + mock_session.return_value.get.call_args[1]['timeout']) + + def test_put_timeout(self): + api_client = self.mock_keep_services(count=1) + force_timeout = [socket.timeout("timed out")] + with tutil.mock_put(force_timeout) as mock_session: + keep_client = arvados.KeepClient(api_client=api_client) + with self.assertRaises(arvados.errors.KeepWriteError): + keep_client.put('foo') + self.assertTrue(mock_session.return_value.put.called) + self.assertEqual( + arvados.KeepClient.DEFAULT_TIMEOUT, + mock_session.return_value.put.call_args[1]['timeout']) + + def test_proxy_get_timeout(self): + # Force a timeout, verifying that the requests.get or + # requests.put method was called with the proxy_timeout + # setting rather than the default timeout. + api_client = self.mock_keep_services(service_type='proxy', count=1) + force_timeout = [socket.timeout("timed out")] + with tutil.mock_get(force_timeout) as mock_session: + keep_client = arvados.KeepClient(api_client=api_client) + with self.assertRaises(arvados.errors.KeepReadError): + keep_client.get('ffffffffffffffffffffffffffffffff') + self.assertTrue(mock_session.return_value.get.called) + self.assertEqual( + arvados.KeepClient.DEFAULT_PROXY_TIMEOUT, + mock_session.return_value.get.call_args[1]['timeout']) + + def test_proxy_put_timeout(self): + # Force a timeout, verifying that the requests.get or + # requests.put method was called with the proxy_timeout + # setting rather than the default timeout. + api_client = self.mock_keep_services(service_type='proxy', count=1) + force_timeout = [socket.timeout("timed out")] + with tutil.mock_put(force_timeout) as mock_session: + keep_client = arvados.KeepClient(api_client=api_client) + with self.assertRaises(arvados.errors.KeepWriteError): + keep_client.put('foo') + self.assertTrue(mock_session.return_value.put.called) + self.assertEqual( + arvados.KeepClient.DEFAULT_PROXY_TIMEOUT, + mock_session.return_value.put.call_args[1]['timeout']) + + def test_probe_order_reference_set(self): + # expected_order[i] is the probe order for + # hash=md5(sprintf("%064x",i)) where there are 16 services + # with uuid sprintf("anything-%015x",j) with j in 0..15. E.g., + # the first probe for the block consisting of 64 "0" + # characters is the service whose uuid is + # "zzzzz-bi6l4-000000000000003", so expected_order[0][0]=='3'. + expected_order = [ + list('3eab2d5fc9681074'), + list('097dba52e648f1c3'), + list('c5b4e023f8a7d691'), + list('9d81c02e76a3bf54'), + ] + hashes = [ + hashlib.md5("{:064x}".format(x)).hexdigest() + for x in range(len(expected_order))] + api_client = self.mock_keep_services(count=16) + keep_client = arvados.KeepClient(api_client=api_client) + for i, hash in enumerate(hashes): + roots = keep_client.weighted_service_roots(hash) + got_order = [ + re.search(r'//\[?keep0x([0-9a-f]+)', root).group(1) + for root in roots] + self.assertEqual(expected_order[i], got_order) + + def test_probe_waste_adding_one_server(self): + hashes = [ + hashlib.md5("{:064x}".format(x)).hexdigest() for x in range(100)] + initial_services = 12 + api_client = self.mock_keep_services(count=initial_services) + keep_client = arvados.KeepClient(api_client=api_client) + probes_before = [ + keep_client.weighted_service_roots(hash) for hash in hashes] + for added_services in range(1, 12): + api_client = self.mock_keep_services(count=initial_services+added_services) + keep_client = arvados.KeepClient(api_client=api_client) + total_penalty = 0 + for hash_index in range(len(hashes)): + probe_after = keep_client.weighted_service_roots( + hashes[hash_index]) + penalty = probe_after.index(probes_before[hash_index][0]) + self.assertLessEqual(penalty, added_services) + total_penalty += penalty + # Average penalty per block should not exceed + # N(added)/N(orig) by more than 20%, and should get closer + # to the ideal as we add data points. + expect_penalty = ( + added_services * + len(hashes) / initial_services) + max_penalty = ( + expect_penalty * + (120 - added_services)/100) + min_penalty = ( + expect_penalty * 8/10) + self.assertTrue( + min_penalty <= total_penalty <= max_penalty, + "With {}+{} services, {} blocks, penalty {} but expected {}..{}".format( + initial_services, + added_services, + len(hashes), + total_penalty, + min_penalty, + max_penalty)) + + def check_64_zeros_error_order(self, verb, exc_class): + data = '0' * 64 + if verb == 'get': + data = hashlib.md5(data).hexdigest() + '+1234' + # Arbitrary port number: + aport = random.randint(1024,65535) + api_client = self.mock_keep_services(service_port=aport, count=16) + keep_client = arvados.KeepClient(api_client=api_client) + with mock.patch('requests.' + verb, + side_effect=socket.timeout) as req_mock, \ + self.assertRaises(exc_class) as err_check: + getattr(keep_client, verb)(data) + urls = [urlparse.urlparse(url) + for url in err_check.exception.request_errors()] + self.assertEqual([('keep0x' + c, aport) for c in '3eab2d5fc9681074'], + [(url.hostname, url.port) for url in urls]) + + def test_get_error_shows_probe_order(self): + self.check_64_zeros_error_order('get', arvados.errors.KeepReadError) + + def test_put_error_shows_probe_order(self): + self.check_64_zeros_error_order('put', arvados.errors.KeepWriteError) + + def check_no_services_error(self, verb, exc_class): + api_client = mock.MagicMock(name='api_client') + api_client.keep_services().accessible().execute.side_effect = ( + arvados.errors.ApiError) + keep_client = arvados.KeepClient(api_client=api_client) + with self.assertRaises(exc_class) as err_check: + getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0') + self.assertEqual(0, len(err_check.exception.request_errors())) + + def test_get_error_with_no_services(self): + self.check_no_services_error('get', arvados.errors.KeepReadError) + + def test_put_error_with_no_services(self): + self.check_no_services_error('put', arvados.errors.KeepWriteError) + + def check_errors_from_last_retry(self, verb, exc_class): + api_client = self.mock_keep_services(count=2) + req_mock = getattr(tutil, 'mock_{}_responses'.format(verb))( + "retry error reporting test", 500, 500, 403, 403) + with req_mock, tutil.skip_sleep, \ + self.assertRaises(exc_class) as err_check: + keep_client = arvados.KeepClient(api_client=api_client) + getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0', + num_retries=3) + self.assertEqual([403, 403], [ + getattr(error, 'status_code', None) + for error in err_check.exception.request_errors().itervalues()]) + + def test_get_error_reflects_last_retry(self): + self.check_errors_from_last_retry('get', arvados.errors.KeepReadError) + + def test_put_error_reflects_last_retry(self): + self.check_errors_from_last_retry('put', arvados.errors.KeepWriteError) + + def test_put_error_does_not_include_successful_puts(self): + data = 'partial failure test' + data_loc = '{}+{}'.format(hashlib.md5(data).hexdigest(), len(data)) + api_client = self.mock_keep_services(count=3) + with tutil.mock_put_responses(data_loc, 200, 500, 500) as req_mock, \ + self.assertRaises(arvados.errors.KeepWriteError) as exc_check: + keep_client = arvados.KeepClient(api_client=api_client) + keep_client.put(data) + self.assertEqual(2, len(exc_check.exception.request_errors())) + class KeepClientRetryTestMixin(object): # Testing with a local Keep store won't exercise the retry behavior. @@ -357,6 +468,10 @@ class KeepClientRetryTestMixin(object): # supporting servers, and prevents side effects in case something hiccups. # To use this mixin, define DEFAULT_EXPECT, DEFAULT_EXCEPTION, and # run_method(). + # + # Test classes must define TEST_PATCHER to a method that mocks + # out appropriate methods in the client. + PROXY_ADDR = 'http://[%s]:65535/' % (tutil.TEST_HOST,) TEST_DATA = 'testdata' TEST_LOCATOR = 'ef654c40ab4f1747fc699915d4f70902+8' @@ -383,28 +498,28 @@ class KeepClientRetryTestMixin(object): self.assertRaises(error_class, self.run_method, *args, **kwargs) def test_immediate_success(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 200): + with self.TEST_PATCHER(self.DEFAULT_EXPECT, 200): self.check_success() def test_retry_then_success(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 500, 200): + with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200): self.check_success(num_retries=3) def test_no_default_retry(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 500, 200): + with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200): self.check_exception() def test_no_retry_after_permanent_error(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 403, 200): + with self.TEST_PATCHER(self.DEFAULT_EXPECT, 403, 200): self.check_exception(num_retries=3) def test_error_after_retries_exhausted(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 500, 500, 200): + with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 500, 200): self.check_exception(num_retries=1) def test_num_retries_instance_fallback(self): self.client_kwargs['num_retries'] = 3 - with tutil.mock_responses(self.DEFAULT_EXPECT, 500, 200): + with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200): self.check_success() @@ -413,13 +528,14 @@ class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase): DEFAULT_EXPECT = KeepClientRetryTestMixin.TEST_DATA DEFAULT_EXCEPTION = arvados.errors.KeepReadError HINTED_LOCATOR = KeepClientRetryTestMixin.TEST_LOCATOR + '+K@xyzzy' + TEST_PATCHER = staticmethod(tutil.mock_get_responses) def run_method(self, locator=KeepClientRetryTestMixin.TEST_LOCATOR, *args, **kwargs): return self.new_client().get(locator, *args, **kwargs) def test_specific_exception_when_not_found(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 404, 200): + with tutil.mock_get_responses(self.DEFAULT_EXPECT, 404, 200): self.check_exception(arvados.errors.NotFoundError, num_retries=3) def test_general_exception_with_mixed_errors(self): @@ -428,7 +544,7 @@ class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase): # This test rigs up 50/50 disagreement between two servers, and # checks that it does not become a NotFoundError. client = self.new_client() - with tutil.mock_responses(self.DEFAULT_EXPECT, 404, 500): + with tutil.mock_get_responses(self.DEFAULT_EXPECT, 404, 500): with self.assertRaises(arvados.errors.KeepReadError) as exc_check: client.get(self.HINTED_LOCATOR) self.assertNotIsInstance( @@ -436,21 +552,17 @@ class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase): "mixed errors raised NotFoundError") def test_hint_server_can_succeed_without_retries(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 404, 200, 500): + with tutil.mock_get_responses(self.DEFAULT_EXPECT, 404, 200, 500): self.check_success(locator=self.HINTED_LOCATOR) def test_try_next_server_after_timeout(self): - side_effects = [ - socket.timeout("timed out"), - (tutil.fake_httplib2_response(200), self.DEFAULT_EXPECT)] - with mock.patch('httplib2.Http.request', - side_effect=iter(side_effects)): + with tutil.mock_get([ + socket.timeout("timed out"), + tutil.fake_requests_response(200, self.DEFAULT_EXPECT)]): self.check_success(locator=self.HINTED_LOCATOR) def test_retry_data_with_wrong_checksum(self): - side_effects = ((tutil.fake_httplib2_response(200), s) - for s in ['baddata', self.TEST_DATA]) - with mock.patch('httplib2.Http.request', side_effect=side_effects): + with tutil.mock_get((tutil.fake_requests_response(200, s) for s in ['baddata', self.TEST_DATA])): self.check_success(locator=self.HINTED_LOCATOR) @@ -458,11 +570,12 @@ class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase): class KeepClientRetryPutTestCase(KeepClientRetryTestMixin, unittest.TestCase): DEFAULT_EXPECT = KeepClientRetryTestMixin.TEST_LOCATOR DEFAULT_EXCEPTION = arvados.errors.KeepWriteError + TEST_PATCHER = staticmethod(tutil.mock_put_responses) def run_method(self, data=KeepClientRetryTestMixin.TEST_DATA, copies=1, *args, **kwargs): return self.new_client().put(data, copies, *args, **kwargs) def test_do_not_send_multiple_copies_to_same_server(self): - with tutil.mock_responses(self.DEFAULT_EXPECT, 200): + with tutil.mock_put_responses(self.DEFAULT_EXPECT, 200): self.check_exception(copies=2, num_retries=3)