def table_metric_mbeans_test(self): """ Test some basic table metric mbeans with simple writes. """ cluster = self.cluster cluster.populate(3) node1, node2, node3 = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start(wait_for_binary_proto=True) version = cluster.version() node1.stress(['write', 'n=10K', 'no-warmup', '-schema', 'replication(factor=3)']) typeName = "ColumnFamily" if version <= '2.2.X' else 'Table' debug('Version {} typeName {}'.format(version, typeName)) # TODO the keyspace and table name are capitalized in 2.0 memtable_size = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='AllMemtablesHeapSize') disk_size = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='LiveDiskSpaceUsed') sstable_count = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='LiveSSTableCount') with JolokiaAgent(node1) as jmx: mem_size = jmx.read_attribute(memtable_size, "Value") self.assertGreater(int(mem_size), 10000) on_disk_size = jmx.read_attribute(disk_size, "Count") self.assertEquals(int(on_disk_size), 0) node1.flush() on_disk_size = jmx.read_attribute(disk_size, "Count") self.assertGreater(int(on_disk_size), 10000) sstables = jmx.read_attribute(sstable_count, "Value") self.assertGreaterEqual(int(sstables), 1)
def query_and_check_repaired_mismatches( self, jmx, session, query, expect_read_repair=True, expect_unconfirmed_inconsistencies=False, expect_confirmed_inconsistencies=False): rr_count = make_mbean('metrics', type='ReadRepair', name='ReconcileRead') unconfirmed_count = make_mbean( 'metrics', type='Table,keyspace=ks', name='RepairedDataInconsistenciesUnconfirmed,scope=tbl') confirmed_count = make_mbean( 'metrics', type='Table,keyspace=ks', name='RepairedDataInconsistenciesConfirmed,scope=tbl') rr_before = self.get_attribute_count(jmx, rr_count) uc_before = self.get_attribute_count(jmx, unconfirmed_count) cc_before = self.get_attribute_count(jmx, confirmed_count) stmt = SimpleStatement(query) stmt.consistency_level = ConsistencyLevel.ALL session.execute(stmt) rr_after = self.get_attribute_count(jmx, rr_count) uc_after = self.get_attribute_count(jmx, unconfirmed_count) cc_after = self.get_attribute_count(jmx, confirmed_count) logger.debug("Read Repair Count: {before}, {after}".format( before=rr_before, after=rr_after)) logger.debug( "Unconfirmed Inconsistency Count: {before}, {after}".format( before=uc_before, after=uc_after)) logger.debug("Confirmed Inconsistency Count: {before}, {after}".format( before=cc_before, after=cc_after)) if expect_read_repair: assert rr_after > rr_before else: assert rr_after == rr_before if expect_unconfirmed_inconsistencies: assert uc_after > uc_before else: assert uc_after == uc_before if expect_confirmed_inconsistencies: assert cc_after > cc_before else: assert cc_after == cc_before
def __init__(self, node, scope): self.node = node self.local_requests_mbean = make_mbean('metrics', type='ClientRequest', scope=scope, name='LocalRequests') self.remote_requests_mbean = make_mbean('metrics', type='ClientRequest', scope=scope, name='RemoteRequests')
def table_metric_mbeans_test(self): """ Test some basic table metric mbeans with simple writes. """ cluster = self.cluster cluster.populate(3) node1, node2, node3 = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start(wait_for_binary_proto=True) version = cluster.version() node1.stress([ 'write', 'n=10K', 'no-warmup', '-schema', 'replication(factor=3)' ]) typeName = "ColumnFamily" if version <= '2.2.X' else 'Table' debug('Version {} typeName {}'.format(version, typeName)) # TODO the keyspace and table name are capitalized in 2.0 memtable_size = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='AllMemtablesHeapSize') disk_size = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='LiveDiskSpaceUsed') sstable_count = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='LiveSSTableCount') with JolokiaAgent(node1) as jmx: mem_size = jmx.read_attribute(memtable_size, "Value") self.assertGreater(int(mem_size), 10000) on_disk_size = jmx.read_attribute(disk_size, "Count") self.assertEquals(int(on_disk_size), 0) node1.flush() on_disk_size = jmx.read_attribute(disk_size, "Count") self.assertGreater(int(on_disk_size), 10000) sstables = jmx.read_attribute(sstable_count, "Value") self.assertGreaterEqual(int(sstables), 1)
def test_table_metric_mbeans(self): """ Test some basic table metric mbeans with simple writes. """ cluster = self.cluster cluster.populate(3) node1, node2, node3 = cluster.nodelist() cluster.start() version = cluster.version() node1.stress([ 'write', 'n=10K', 'no-warmup', '-schema', 'replication(factor=3)' ]) typeName = "ColumnFamily" if version < '3.0' else 'Table' logger.debug('Version {} typeName {}'.format(version, typeName)) # TODO the keyspace and table name are capitalized in 2.0 memtable_size = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='AllMemtablesHeapSize') disk_size = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='LiveDiskSpaceUsed') sstable_count = make_mbean('metrics', type=typeName, keyspace='keyspace1', scope='standard1', name='LiveSSTableCount') with JolokiaAgent(node1) as jmx: mem_size = jmx.read_attribute(memtable_size, "Value") assert int(mem_size) > 10000 on_disk_size = jmx.read_attribute(disk_size, "Count") assert int(on_disk_size) == 0 node1.flush() on_disk_size = jmx.read_attribute(disk_size, "Count") assert int(on_disk_size) > 10000 sstables = jmx.read_attribute(sstable_count, "Value") assert int(sstables) >= 1
def _batchlog_replay_compatibility_test(self, coordinator_idx, current_nodes, previous_version, previous_nodes, protocol_version): session = self.prepare_mixed(coordinator_idx, current_nodes, previous_version, previous_nodes, protocol_version=protocol_version, install_byteman=True) coordinator = self.cluster.nodelist()[coordinator_idx] coordinator.byteman_submit( [mk_bman_path('fail_after_batchlog_write.btm')]) logger.debug( "Injected byteman scripts to enable batchlog replay {}".format( coordinator.name)) query = """ BEGIN BATCH INSERT INTO users (id, firstname, lastname) VALUES (0, 'Jack', 'Sparrow') INSERT INTO users (id, firstname, lastname) VALUES (1, 'Will', 'Turner') APPLY BATCH """ session.execute(query) # batchlog replay skips over all entries that are younger than # 2 * write_request_timeout_in_ms ms: 1x timeout for all mutations to be written, # and another 1x timeout for batch remove mutation to be received. delay = 2 * coordinator.get_conf_option( 'write_request_timeout_in_ms') / 1000.0 + 1 logger.debug( 'Sleeping for {}s for the batches to not be skipped'.format(delay)) time.sleep(delay) total_batches_replayed = 0 blm = make_mbean('db', type='BatchlogManager') for n in self.cluster.nodelist(): if n == coordinator: continue with JolokiaAgent(n) as jmx: logger.debug('Forcing batchlog replay for {}'.format(n.name)) jmx.execute_method(blm, 'forceBatchlogReplay') batches_replayed = jmx.read_attribute(blm, 'TotalBatchesReplayed') logger.debug('{} batches replayed on node {}'.format( batches_replayed, n.name)) total_batches_replayed += batches_replayed assert total_batches_replayed >= 2 for node in self.cluster.nodelist(): session = self.patient_exclusive_cql_connection( node, protocol_version=protocol_version) rows = sorted( session.execute( 'SELECT id, firstname, lastname FROM ks.users')) assert [[0, 'Jack', 'Sparrow'], [1, 'Will', 'Turner']], [list(rows[0]) == list(rows[1])]
def get_table_metric(self, keyspace, table, metric, attr="Count"): mbean = make_mbean("metrics", keyspace=keyspace, scope=table, type="Table", name=metric) return self.jmx.read_attribute(mbean, attr)
def test_blacklisted_directory(self): cluster = self.cluster cluster.set_datadir_count(3) cluster.populate(1) [node] = cluster.nodelist() cluster.start() session = self.patient_cql_connection(node) create_ks(session, 'ks', 1) create_c1c2_table(self, session) insert_c1c2(session, n=10000) node.flush() for k in range(0, 10000): query_c1c2(session, k) node.compact() mbean = make_mbean('db', type='BlacklistedDirectories') with JolokiaAgent(node) as jmx: jmx.execute_method(mbean, 'markUnwritable', [os.path.join(node.get_path(), 'data1')]) for k in range(0, 10000): query_c1c2(session, k) node.nodetool('relocatesstables') for k in range(0, 10000): query_c1c2(session, k)
def test_oversized_mutation(self): """ Test that multi-DC write failures return operation failed rather than a timeout. @jira_ticket CASSANDRA-16334. """ cluster = self.cluster cluster.populate([2, 2]) cluster.set_configuration_options( values={'max_mutation_size_in_kb': 128}) cluster.start() node1 = cluster.nodelist()[0] session = self.patient_exclusive_cql_connection(node1) session.execute( "CREATE KEYSPACE k WITH replication = {'class': 'NetworkTopologyStrategy', 'dc1': 2, 'dc2': 2}" ) session.execute("CREATE TABLE k.t (key int PRIMARY KEY, val blob)") payload = '1' * 1024 * 256 query = "INSERT INTO k.t (key, val) VALUES (1, textAsBlob('{}'))".format( payload) assert_write_failure(session, query, ConsistencyLevel.LOCAL_ONE) assert_write_failure(session, query, ConsistencyLevel.ONE) # verify that no hints are created with JolokiaAgent(node1) as jmx: assert 0 == jmx.read_attribute( make_mbean('metrics', type='Storage', name='TotalHints'), 'Count')
def test_tracing_does_not_interfere_with_digest_calculation(self): """ Test that enabling tracing doesn't interfere with digest responses when using RandomPartitioner. The use of a threadlocal MessageDigest for generating both DigestResponse messages and for calculating tokens meant that the DigestResponse was always incorrect when both RP and tracing were enabled, leading to unnecessary data reads. @jira_ticket CASSANDRA-13964 """ session = self.prepare(random_partitioner=True) self.trace(session) node1 = self.cluster.nodelist()[0] rr_count = make_mbean('metrics', type='ReadRepair', name='RepairedBlocking') with JolokiaAgent(node1) as jmx: # the MBean may not have been initialized, in which case Jolokia agent will return # a HTTP 404 response. If we receive such, we know that no digest mismatch was reported # If we are able to read the MBean attribute, assert that the count is 0 if jmx.has_mbean(rr_count): # expect 0 digest mismatches assert 0 == jmx.read_attribute(rr_count, 'Count') else: pass
def test_tracing_does_not_interfere_with_digest_calculation(self): """ Test that enabling tracing doesn't interfere with digest responses when using RandomPartitioner. The use of a threadlocal MessageDigest for generating both DigestResponse messages and for calculating tokens meant that the DigestResponse was always incorrect when both RP and tracing were enabled, leading to unnecessary data reads. @jira_ticket CASSANDRA-13964 """ cluster = self.cluster cluster.populate(3) cluster.set_configuration_options( values={ 'write_request_timeout_in_ms': 30000, 'read_request_timeout_in_ms': 30000 }) cluster.set_partitioner("org.apache.cassandra.dht.RandomPartitioner") cluster.start( jvm_args=['-Dcassandra.wait_for_tracing_events_timeout_secs=15']) node1 = cluster.nodelist()[0] session = self.patient_cql_connection(node1) create_ks(session, 'ks', 3) session.execute(""" CREATE TABLE ks.users ( userid uuid PRIMARY KEY, firstname text, lastname text, age int ); """) insert = session.prepare( "INSERT INTO ks.users (userid, firstname, lastname, age) " "VALUES (?, 'Frodo', 'Baggins', 32)") insert.consistency_level = ConsistencyLevel.ALL select = session.prepare("SELECT firstname, lastname " "FROM ks.users WHERE userid = ?") select.consistency_level = ConsistencyLevel.ALL for _ in range(10): id = uuid4() session.execute(insert.bind((id, )), timeout=30) res = session.execute(select.bind((id, )), timeout=30, trace=True) assert 1 == len(res.response_future.get_query_trace_ids()) rr_count = make_mbean('metrics', type='ReadRepair', name='RepairedBlocking') with JolokiaAgent(node1) as jmx: # the MBean may not have been initialized, in which case Jolokia agent will return # a HTTP 404 response. If we receive such, we know that no digest mismatch was reported # If we are able to read the MBean attribute, assert that the count is 0 if jmx.has_mbean(rr_count): # expect 0 digest mismatches assert 0 == jmx.read_attribute(rr_count, 'Count') else: pass
def blacklisted_directory_test(self): cluster = self.cluster cluster.set_datadir_count(3) cluster.populate(1) [node] = cluster.nodelist() remove_perf_disable_shared_mem(node) cluster.start(wait_for_binary_proto=True) session = self.patient_cql_connection(node) create_ks(session, 'ks', 1) create_c1c2_table(self, session) insert_c1c2(session, n=10000) node.flush() for k in xrange(0, 10000): query_c1c2(session, k) node.compact() mbean = make_mbean('db', type='BlacklistedDirectories') with JolokiaAgent(node) as jmx: jmx.execute_method(mbean, 'markUnwritable', [os.path.join(node.get_path(), 'data0')]) for k in xrange(0, 10000): query_c1c2(session, k) node.nodetool('relocatesstables') for k in xrange(0, 10000): query_c1c2(session, k)
def overlapping_data_folders(self): """ @jira_ticket CASSANDRA-10902 """ self.cluster.populate(1) node1 = self.cluster.nodelist()[0] default_path = node1.data_directories()[0] node1.set_configuration_options({ 'saved_caches_directory': os.path.join(default_path, 'saved_caches') }) remove_perf_disable_shared_mem(node1) self.cluster.start(wait_for_binary_proto=True) session = self.patient_exclusive_cql_connection(node1) session.execute( "CREATE KEYSPACE ks WITH REPLICATION = {'class': 'SimpleStrategy', 'replication_factor': 1}" ) session.execute("CREATE TABLE ks.tab (key int PRIMARY KEY, a int)") session.execute("INSERT INTO ks.tab (key, a) VALUES (%s, %s)", [0, 0]) session.execute("SELECT * FROM ks.tab WHERE key = %s", [0]) cache_service = make_mbean('db', type="Caches") with JolokiaAgent(node1) as jmx: jmx.execute_method(cache_service, 'saveCaches') self.cluster.stop() self.cluster.start(wait_for_binary_proto=True)
def test_reloadlocalschema(self): """ @jira_ticket CASSANDRA-13954 Test that `nodetool reloadlocalschema` works as intended """ cluster = self.cluster cluster.populate(1) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) # for jmx cluster.start() session = self.patient_cql_connection(node) query = "CREATE KEYSPACE IF NOT EXISTS test WITH replication = {'class': 'NetworkTopologyStrategy', 'datacenter1': 2};" session.execute(query) query = 'CREATE TABLE test.test (pk int, ck int, PRIMARY KEY (pk, ck));' session.execute(query) ss = make_mbean('db', type='StorageService') schema_version = '' # get initial schema version with JolokiaAgent(node) as jmx: schema_version = jmx.read_attribute(ss, 'SchemaVersion') # manually add a regular column 'val' to test.test query = """ INSERT INTO system_schema.columns (keyspace_name, table_name, column_name, clustering_order, column_name_bytes, kind, position, type) VALUES ('test', 'test', 'val', 'none', 0x76616c, 'regular', -1, 'int');""" session.execute(query) # validate that schema version wasn't automatically updated with JolokiaAgent(node) as jmx: self.assertEqual(schema_version, jmx.read_attribute(ss, 'SchemaVersion')) # make sure the new column wasn't automagically picked up assert_invalid( session, 'INSERT INTO test.test (pk, ck, val) VALUES (0, 1, 2);') # force the node to reload schema from disk node.nodetool('reloadlocalschema') # validate that schema version changed with JolokiaAgent(node) as jmx: self.assertNotEqual(schema_version, jmx.read_attribute(ss, 'SchemaVersion')) # try an insert with the new column again and validate it succeeds this time session.execute( 'INSERT INTO test.test (pk, ck, val) VALUES (0, 1, 2);') assert_all(session, 'SELECT pk, ck, val FROM test.test;', [[0, 1, 2]])
def table_metric(node, keyspace, table, name): version = node.get_cassandra_version() typeName = "ColumnFamily" if version < '3.0' else 'Table' with JolokiaAgent(node) as jmx: mbean = make_mbean('metrics', type=typeName, name=name, keyspace=keyspace, scope=table) value = jmx.read_attribute(mbean, 'Value') return value
def __init__(self, node, keyspace, table): assert isinstance(node, Node) self.jmx = JolokiaAgent(node) self.write_latency_mbean = make_mbean("metrics", type="Table", name="WriteLatency", keyspace=keyspace, scope=table) self.speculative_reads_mbean = make_mbean("metrics", type="Table", name="SpeculativeRetries", keyspace=keyspace, scope=table) self.transient_writes_mbean = make_mbean("metrics", type="Table", name="TransientWrites", keyspace=keyspace, scope=table)
def __init__(self, scope, name): self.scope = scope self.name = name self.mbean = make_mbean('metrics', type='ClientRequest', scope=scope, name=name) self.values = {} self.init()
def test_reloadlocalschema(self): """ @jira_ticket CASSANDRA-13954 Test that `nodetool reloadlocalschema` works as intended """ cluster = self.cluster cluster.populate(1) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) # for jmx cluster.start() session = self.patient_cql_connection(node) query = "CREATE KEYSPACE IF NOT EXISTS test WITH replication = {'class': 'NetworkTopologyStrategy', 'datacenter1': 2};" session.execute(query) query = 'CREATE TABLE test.test (pk int, ck int, PRIMARY KEY (pk, ck));' session.execute(query) ss = make_mbean('db', type='StorageService') schema_version = '' # get initial schema version with JolokiaAgent(node) as jmx: schema_version = jmx.read_attribute(ss, 'SchemaVersion') # manually add a regular column 'val' to test.test query = """ INSERT INTO system_schema.columns (keyspace_name, table_name, column_name, clustering_order, column_name_bytes, kind, position, type) VALUES ('test', 'test', 'val', 'none', 0x76616c, 'regular', -1, 'int');""" session.execute(query) # validate that schema version wasn't automatically updated with JolokiaAgent(node) as jmx: self.assertEqual(schema_version, jmx.read_attribute(ss, 'SchemaVersion')) # make sure the new column wasn't automagically picked up assert_invalid(session, 'INSERT INTO test.test (pk, ck, val) VALUES (0, 1, 2);') # force the node to reload schema from disk node.nodetool('reloadlocalschema') # validate that schema version changed with JolokiaAgent(node) as jmx: self.assertNotEqual(schema_version, jmx.read_attribute(ss, 'SchemaVersion')) # try an insert with the new column again and validate it succeeds this time session.execute('INSERT INTO test.test (pk, ck, val) VALUES (0, 1, 2);') assert_all(session, 'SELECT pk, ck, val FROM test.test;', [[0, 1, 2]])
def test_closing_connections(self): """ @jira_ticket CASSANDRA-6546 Test CASSANDRA-6546 - do connections get closed when disabling / renabling thrift service? """ cluster = self.cluster cluster.set_configuration_options(values={ 'start_rpc': 'true', 'rpc_server_type': 'hsha', 'rpc_max_threads': 20 }) cluster.populate(1) (node1, ) = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start() session = self.patient_cql_connection(node1) create_ks(session, 'test', 1) session.execute( "CREATE TABLE \"CF\" (key text PRIMARY KEY, val text) WITH COMPACT STORAGE;" ) def make_connection(): host, port = node1.network_interfaces['thrift'] client = get_thrift_client(host, port) client.transport.open() return client pools = [] connected_thrift_clients = make_mbean('metrics', type='Client', name='connectedThriftClients') for i in range(10): logger.debug("Creating connection pools..") for x in range(3): pools.append(make_connection()) logger.debug( "Disabling/Enabling thrift iteration #{i}".format(i=i)) node1.nodetool('disablethrift') node1.nodetool('enablethrift') logger.debug("Closing connections from the client side..") for client in pools: client.transport.close() with JolokiaAgent(node1) as jmx: num_clients = jmx.read_attribute(connected_thrift_clients, "Value") assert int( num_clients ) == 0, "There are still open Thrift connections after stopping service " + str( num_clients)
def query_and_check_repaired_mismatches(self, jmx, session, query, expect_read_repair=True, expect_unconfirmed_inconsistencies=False, expect_confirmed_inconsistencies=False): rr_count = make_mbean('metrics', type='ReadRepair', name='ReconcileRead') unconfirmed_count = make_mbean('metrics', type='Table,keyspace=ks', name='RepairedDataInconsistenciesUnconfirmed,scope=tbl') confirmed_count = make_mbean('metrics', type='Table,keyspace=ks', name='RepairedDataInconsistenciesConfirmed,scope=tbl') rr_before = self.get_attribute_count(jmx, rr_count) uc_before = self.get_attribute_count(jmx, unconfirmed_count) cc_before = self.get_attribute_count(jmx, confirmed_count) stmt = SimpleStatement(query) stmt.consistency_level = ConsistencyLevel.ALL session.execute(stmt) rr_after = self.get_attribute_count(jmx, rr_count) uc_after = self.get_attribute_count(jmx, unconfirmed_count) cc_after = self.get_attribute_count(jmx, confirmed_count) logger.debug("Read Repair Count: {before}, {after}".format(before=rr_before, after=rr_after)) logger.debug("Unconfirmed Inconsistency Count: {before}, {after}".format(before=uc_before, after=uc_after)) logger.debug("Confirmed Inconsistency Count: {before}, {after}".format(before=cc_before, after=cc_after)) if expect_read_repair: assert rr_after > rr_before else: assert rr_after == rr_before if expect_unconfirmed_inconsistencies: assert uc_after > uc_before else: assert uc_after == uc_before if expect_confirmed_inconsistencies: assert cc_after > cc_before else: assert cc_after == cc_before
def test_closing_connections(self): """ @jira_ticket CASSANDRA-6546 Test CASSANDRA-6546 - do connections get closed when disabling / renabling thrift service? """ cluster = self.cluster cluster.set_configuration_options(values={ 'start_rpc': 'true', 'rpc_server_type': 'hsha', 'rpc_max_threads': 20 }) cluster.populate(1) (node1, ) = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start(wait_for_binary_proto=True) session = self.patient_cql_connection(node1) self.create_ks(session, 'test', 1) session.execute( "CREATE TABLE \"CF\" (key text PRIMARY KEY, val text) WITH COMPACT STORAGE;" ) def make_connection(): pool = pycassa.ConnectionPool('test', timeout=None) cf = pycassa.ColumnFamily(pool, 'CF') return pool pools = [] connected_thrift_clients = make_mbean('metrics', type='Client', name='connectedThriftClients') for i in xrange(10): debug("Creating connection pools..") for x in xrange(3): pools.append(make_connection()) debug("Disabling/Enabling thrift iteration #{i}".format(i=i)) node1.nodetool('disablethrift') node1.nodetool('enablethrift') debug("Closing connections from the client side..") for pool in pools: pool.dispose() with JolokiaAgent(node1) as jmx: num_clients = jmx.read_attribute(connected_thrift_clients, "Value") self.assertEqual( int(num_clients), 0, "There are still open Thrift connections after stopping service" )
def _deprecated_repair_jmx(self, method, arguments): """ * Launch a two node, two DC cluster * Create a keyspace and table * Insert some data * Call the deprecated repair JMX API based on the arguments passed into this method * Check the node log to see if the correct repair was performed based on the jmx args """ cluster = self.cluster logger.debug("Starting cluster..") cluster.populate([1, 1]) node1, node2 = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start() supports_pull_repair = cluster.version() >= LooseVersion('3.10') session = self.patient_cql_connection(node1) create_ks(session, 'ks', 2) create_cf(session, 'cf', read_repair=0.0, columns={'c1': 'text', 'c2': 'text'}) insert_c1c2(session, n=1000, consistency=ConsistencyLevel.ALL) # Run repair mbean = make_mbean('db', 'StorageService') with JolokiaAgent(node1) as jmx: # assert repair runs and returns valid cmd number assert jmx.execute_method(mbean, method, arguments) == 1 # wait for log to start node1.watch_log_for("Starting repair command") # get repair parameters from the log line = node1.grep_log(("Starting repair command #1" + (" \([^\)]+\)" if cluster.version() >= LooseVersion("3.10") else "") + ", repairing keyspace ks with repair options \(parallelism: (?P<parallelism>\w+), primary range: (?P<pr>\w+), " "incremental: (?P<incremental>\w+), job threads: (?P<jobs>\d+), ColumnFamilies: (?P<cfs>.+), dataCenters: (?P<dc>.+), " "hosts: (?P<hosts>.+), # of ranges: (?P<ranges>\d+)(, pull repair: (?P<pullrepair>true|false))?\)")) assert_length_equal(line, 1) line, m = line[0] if supports_pull_repair: assert m.group("pullrepair"), "false" == "Pull repair cannot be enabled through the deprecated API so the pull repair option should always be false." return {"parallelism": m.group("parallelism"), "primary_range": m.group("pr"), "incremental": m.group("incremental"), "job_threads": m.group("jobs"), "column_families": m.group("cfs"), "data_centers": m.group("dc"), "hosts": m.group("hosts"), "ranges": m.group("ranges")}
def _deprecated_repair_jmx(self, method, arguments): """ * Launch a two node, two DC cluster * Create a keyspace and table * Insert some data * Call the deprecated repair JMX API based on the arguments passed into this method * Check the node log to see if the correct repair was performed based on the jmx args """ cluster = self.cluster logger.debug("Starting cluster..") cluster.populate([1, 1]) node1, node2 = cluster.nodelist() cluster.start() supports_pull_repair = cluster.version() >= LooseVersion('3.10') session = self.patient_cql_connection(node1) create_ks(session, 'ks', 2) create_cf(session, 'cf', read_repair=0.0, columns={'c1': 'text', 'c2': 'text'}) insert_c1c2(session, n=1000, consistency=ConsistencyLevel.ALL) # Run repair mbean = make_mbean('db', 'StorageService') with JolokiaAgent(node1) as jmx: # assert repair runs and returns valid cmd number assert jmx.execute_method(mbean, method, arguments) == 1 # wait for log to start node1.watch_log_for("Starting repair command") # get repair parameters from the log line = node1.grep_log((r"Starting repair command #1" + (r" \([^\)]+\)" if cluster.version() >= LooseVersion("3.10") else "") + r", repairing keyspace ks with repair options \(parallelism: (?P<parallelism>\w+), primary range: (?P<pr>\w+), " r"incremental: (?P<incremental>\w+), job threads: (?P<jobs>\d+), ColumnFamilies: (?P<cfs>.+), dataCenters: (?P<dc>.+), " r"hosts: (?P<hosts>.+), # of ranges: (?P<ranges>\d+)(, pull repair: (?P<pullrepair>true|false))?(, ignore unreplicated keyspaces: (?P<ignoreunrepl>true|false))?\)")) assert_length_equal(line, 1) line, m = line[0] if supports_pull_repair: assert m.group("pullrepair"), "false" == "Pull repair cannot be enabled through the deprecated API so the pull repair option should always be false." return {"parallelism": m.group("parallelism"), "primary_range": m.group("pr"), "incremental": m.group("incremental"), "job_threads": m.group("jobs"), "column_families": m.group("cfs"), "data_centers": m.group("dc"), "hosts": m.group("hosts"), "ranges": m.group("ranges")}
def _batchlog_replay_compatibility_test(self, coordinator_idx, current_nodes, previous_version, previous_nodes, protocol_version): session = self.prepare_mixed(coordinator_idx, current_nodes, previous_version, previous_nodes, protocol_version=protocol_version, install_byteman=True) coordinator = self.cluster.nodelist()[coordinator_idx] coordinator.byteman_submit(['./byteman/fail_after_batchlog_write.btm']) debug("Injected byteman scripts to enable batchlog replay {}".format( coordinator.name)) query = """ BEGIN BATCH INSERT INTO users (id, firstname, lastname) VALUES (0, 'Jack', 'Sparrow') INSERT INTO users (id, firstname, lastname) VALUES (1, 'Will', 'Turner') APPLY BATCH """ session.execute(query) total_batches_replayed = 0 blm = make_mbean('db', type='BatchlogManager') for n in self.cluster.nodelist(): if n == coordinator: continue with JolokiaAgent(n) as jmx: debug('Forcing batchlog replay for {}'.format(n.name)) jmx.execute_method(blm, 'forceBatchlogReplay') batches_replayed = jmx.read_attribute(blm, 'TotalBatchesReplayed') debug('{} batches replayed on node {}'.format( batches_replayed, n.name)) total_batches_replayed += batches_replayed assert_greater_equal(total_batches_replayed, 2) for node in self.cluster.nodelist(): session = self.patient_exclusive_cql_connection( node, protocol_version=protocol_version) rows = sorted( session.execute( 'SELECT id, firstname, lastname FROM ks.users')) self.assertEqual([[0, 'Jack', 'Sparrow'], [1, 'Will', 'Turner']], [list(rows[0]), list(rows[1])])
def test_bloom_filter_false_ratio(self): """ Test for CASSANDRA-15834 Verifies if BloomFilterFalseRatio takes into account true negatives. Without this fix, the following scenario (many reads for non-existing rows) would yield BloomFilterFalseRatio=1.0. With the fix we assume it should be less then the default bloom_filter_fp_chance. """ cluster = self.cluster cluster.populate(1) node = cluster.nodelist()[0] cluster.start(wait_for_binary_proto=True) session = self.patient_exclusive_cql_connection(node) keyspace = 'bloom_ratio_test_ks' create_ks(session, keyspace, 1) create_c1c2_table(self, session) insert_c1c2(session, n=10) node.nodetool("flush " + keyspace) for key in range(10000): session.execute("SELECT * from cf where key = '{0}'".format(key)) bloom_filter_false_ratios = [ make_mbean('metrics', type='Table', name='RecentBloomFilterFalseRatio'), make_mbean('metrics', type='Table', keyspace=keyspace, scope='cf', name='BloomFilterFalseRatio'), make_mbean('metrics', type='Table', name='BloomFilterFalseRatio'), make_mbean('metrics', type='Table', keyspace=keyspace, scope='cf', name='RecentBloomFilterFalseRatio'), ] with JolokiaAgent(node) as jmx: for metric in bloom_filter_false_ratios: ratio = jmx.read_attribute(metric, "Value") # Bloom filter false positive ratio should not be greater than the default bloom_filter_fp_chance. assert ratio < 0.01
def test_set_get_batchlog_replay_throttle(self): """ @jira_ticket CASSANDRA-13614 Test that batchlog replay throttle can be set and get through JMX """ cluster = self.cluster cluster.populate(2) node = cluster.nodelist()[0] cluster.start() # Set and get throttle with JMX, ensuring that the rate change is logged with JolokiaAgent(node) as jmx: mbean = make_mbean('db', 'StorageService') jmx.write_attribute(mbean, 'BatchlogReplayThrottleInKB', 4096) assert len(node.grep_log('Updating batchlog replay throttle to 4096 KB/s, 2048 KB/s per endpoint', filename='debug.log')) > 0 assert 4096 == jmx.read_attribute(mbean, 'BatchlogReplayThrottleInKB')
def test_closing_connections(self): """ @jira_ticket CASSANDRA-6546 Test CASSANDRA-6546 - do connections get closed when disabling / renabling thrift service? """ cluster = self.cluster cluster.set_configuration_options(values={ 'start_rpc': 'true', 'rpc_server_type': 'hsha', 'rpc_max_threads': 20 }) cluster.populate(1) (node1,) = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start(wait_for_binary_proto=True) session = self.patient_cql_connection(node1) create_ks(session, 'test', 1) session.execute("CREATE TABLE \"CF\" (key text PRIMARY KEY, val text) WITH COMPACT STORAGE;") def make_connection(): host, port = node1.network_interfaces['thrift'] client = get_thrift_client(host, port) client.transport.open() return client pools = [] connected_thrift_clients = make_mbean('metrics', type='Client', name='connectedThriftClients') for i in range(10): logger.debug("Creating connection pools..") for x in range(3): pools.append(make_connection()) logger.debug("Disabling/Enabling thrift iteration #{i}".format(i=i)) node1.nodetool('disablethrift') node1.nodetool('enablethrift') logger.debug("Closing connections from the client side..") for client in pools: client.transport.close() with JolokiaAgent(node1) as jmx: num_clients = jmx.read_attribute(connected_thrift_clients, "Value") assert int(num_clients) == 0, "There are still open Thrift connections after stopping service " + str(num_clients)
def _batchlog_replay_compatibility_test(self, coordinator_idx, current_nodes, previous_version, previous_nodes, protocol_version): session = self.prepare_mixed(coordinator_idx, current_nodes, previous_version, previous_nodes, protocol_version=protocol_version, install_byteman=True) coordinator = self.cluster.nodelist()[coordinator_idx] coordinator.byteman_submit(['./byteman/fail_after_batchlog_write.btm']) logger.debug("Injected byteman scripts to enable batchlog replay {}".format(coordinator.name)) query = """ BEGIN BATCH INSERT INTO users (id, firstname, lastname) VALUES (0, 'Jack', 'Sparrow') INSERT INTO users (id, firstname, lastname) VALUES (1, 'Will', 'Turner') APPLY BATCH """ session.execute(query) # batchlog replay skips over all entries that are younger than # 2 * write_request_timeout_in_ms ms: 1x timeout for all mutations to be written, # and another 1x timeout for batch remove mutation to be received. delay = 2 * coordinator.get_conf_option('write_request_timeout_in_ms') / 1000.0 + 1 logger.debug('Sleeping for {}s for the batches to not be skipped'.format(delay)) time.sleep(delay) total_batches_replayed = 0 blm = make_mbean('db', type='BatchlogManager') for n in self.cluster.nodelist(): if n == coordinator: continue with JolokiaAgent(n) as jmx: logger.debug('Forcing batchlog replay for {}'.format(n.name)) jmx.execute_method(blm, 'forceBatchlogReplay') batches_replayed = jmx.read_attribute(blm, 'TotalBatchesReplayed') logger.debug('{} batches replayed on node {}'.format(batches_replayed, n.name)) total_batches_replayed += batches_replayed assert total_batches_replayed >= 2 for node in self.cluster.nodelist(): session = self.patient_exclusive_cql_connection(node, protocol_version=protocol_version) rows = sorted(session.execute('SELECT id, firstname, lastname FROM ks.users')) assert [[0, 'Jack', 'Sparrow'], [1, 'Will', 'Turner']], [list(rows[0]) == list(rows[1])]
def test_set_get_batchlog_replay_throttle(self): """ @jira_ticket CASSANDRA-13614 Test that batchlog replay throttle can be set and get through JMX """ cluster = self.cluster cluster.populate(2) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) cluster.start() # Set and get throttle with JMX, ensuring that the rate change is logged with JolokiaAgent(node) as jmx: mbean = make_mbean('db', 'StorageService') jmx.write_attribute(mbean, 'BatchlogReplayThrottleInKB', 4096) assert len(node.grep_log('Updating batchlog replay throttle to 4096 KB/s, 2048 KB/s per endpoint', filename='debug.log')) > 0 assert 4096 == jmx.read_attribute(mbean, 'BatchlogReplayThrottleInKB')
def test_closing_connections(self): """ @jira_ticket CASSANDRA-6546 Test CASSANDRA-6546 - do connections get closed when disabling / renabling thrift service? """ cluster = self.cluster cluster.set_configuration_options(values={ 'start_rpc': 'true', 'rpc_server_type': 'hsha', 'rpc_max_threads': 20 }) cluster.populate(1) (node1,) = cluster.nodelist() remove_perf_disable_shared_mem(node1) cluster.start(wait_for_binary_proto=True) session = self.patient_cql_connection(node1) create_ks(session, 'test', 1) session.execute("CREATE TABLE \"CF\" (key text PRIMARY KEY, val text) WITH COMPACT STORAGE;") def make_connection(): pool = pycassa.ConnectionPool('test', timeout=None) cf = pycassa.ColumnFamily(pool, 'CF') return pool pools = [] connected_thrift_clients = make_mbean('metrics', type='Client', name='connectedThriftClients') for i in xrange(10): debug("Creating connection pools..") for x in xrange(3): pools.append(make_connection()) debug("Disabling/Enabling thrift iteration #{i}".format(i=i)) node1.nodetool('disablethrift') node1.nodetool('enablethrift') debug("Closing connections from the client side..") for pool in pools: pool.dispose() with JolokiaAgent(node1) as jmx: num_clients = jmx.read_attribute(connected_thrift_clients, "Value") self.assertEqual(int(num_clients), 0, "There are still open Thrift connections after stopping service")
def overlapping_data_folders(self): """ @jira_ticket CASSANDRA-10902 """ self.cluster.populate(1) node1 = self.cluster.nodelist()[0] default_path = node1.data_directories()[0] node1.set_configuration_options({'saved_caches_directory': os.path.join(default_path, 'saved_caches')}) remove_perf_disable_shared_mem(node1) self.cluster.start(wait_for_binary_proto=True) session = self.patient_exclusive_cql_connection(node1) session.execute("CREATE KEYSPACE ks WITH REPLICATION = {'class': 'SimpleStrategy', 'replication_factor': 1}") session.execute("CREATE TABLE ks.tab (key int PRIMARY KEY, a int)") session.execute("INSERT INTO ks.tab (key, a) VALUES (%s, %s)", [0, 0]) session.execute("SELECT * FROM ks.tab WHERE key = %s", [0]) cache_service = make_mbean('db', type="Caches") with JolokiaAgent(node1) as jmx: jmx.execute_method(cache_service, 'saveCaches') self.cluster.stop() self.cluster.start(wait_for_binary_proto=True)
def preview_failure_count(node): mbean = make_mbean('metrics', type='Repair', name='PreviewFailures') with JolokiaAgent(node) as jmx: return jmx.read_attribute(mbean, 'Count')
def test_multidatacenter_local_quorum(self): ''' @jira_ticket CASSANDRA-13074 If we do only local datacenters reads in a multidatacenter DES setup, DES should take effect and route around a degraded node ''' def no_cross_dc(scores, cross_dc_nodes): return all('/' + k.address() not in scores for k in cross_dc_nodes) def snitchable(scores_before, scores_after, needed_nodes): return all('/' + k.address() in scores_before and '/' + k.address() in scores_after for k in needed_nodes) cluster = self.cluster cluster.populate([3, 3]) coordinator_node, healthy_node, degraded_node, node4, node5, node6 = cluster.nodelist() # increase DES reset/update interval so we clear any cross-DC startup reads faster cluster.set_configuration_options(values={'dynamic_snitch_reset_interval_in_ms': 10000, 'dynamic_snitch_update_interval_in_ms': 50, 'phi_convict_threshold': 12}) remove_perf_disable_shared_mem(coordinator_node) remove_perf_disable_shared_mem(degraded_node) # Delay reads on the degraded node by 50 milliseconds degraded_node.start(jvm_args=['-Dcassandra.test.read_iteration_delay_ms=50', '-Dcassandra.allow_unsafe_join=true']) cluster.start(wait_for_binary_proto=30, wait_other_notice=True) des = make_mbean('db', type='DynamicEndpointSnitch') read_stage = make_mbean('metrics', type='ThreadPools', path='request', scope='ReadStage', name='CompletedTasks') session = self.patient_exclusive_cql_connection(coordinator_node) session.execute("CREATE KEYSPACE snitchtestks WITH replication = {'class': 'NetworkTopologyStrategy', 'dc1': 3, 'dc2': 3}") session.execute("CREATE TABLE snitchtestks.tbl1 (key int PRIMARY KEY) WITH speculative_retry = 'NONE' AND dclocal_read_repair_chance = 0.0") read_stmt = session.prepare("SELECT * FROM snitchtestks.tbl1 where key = ?") read_stmt.consistency_level = ConsistencyLevel.LOCAL_QUORUM insert_stmt = session.prepare("INSERT INTO snitchtestks.tbl1 (key) VALUES (?)") insert_stmt.consistency_level = ConsistencyLevel.ALL with JolokiaAgent(coordinator_node) as jmx: with JolokiaAgent(degraded_node) as bad_jmx: for x in range(0, 300): session.execute(insert_stmt, [x]) cleared = False # Wait for a snitch reset in case any earlier # startup process populated cross-DC read timings while not cleared: scores = jmx.read_attribute(des, 'Scores') cleared = ('/127.0.0.1' in scores and (len(scores) == 1)) or not scores snitchable_count = 0 for x in range(0, 300): degraded_reads_before = bad_jmx.read_attribute(read_stage, 'Value') scores_before = jmx.read_attribute(des, 'Scores') assert no_cross_dc(scores_before, [node4, node5, node6]), "Cross DC scores were present: " + str(scores_before) future = session.execute_async(read_stmt, [x]) future.result() scores_after = jmx.read_attribute(des, 'Scores') assert no_cross_dc(scores_after, [node4, node5, node6]), "Cross DC scores were present: " + str(scores_after) if snitchable(scores_before, scores_after, [coordinator_node, healthy_node, degraded_node]): snitchable_count = snitchable_count + 1 # If the DES correctly routed the read around the degraded node, # it shouldn't have another completed read request in metrics assert (degraded_reads_before == bad_jmx.read_attribute(read_stage, 'Value')) else: # sleep to give dynamic snitch time to recalculate scores time.sleep(.1) # check that most reads were snitchable, with some # room allowed in case score recalculation is slow assert snitchable_count >= 250
def test_compactionstats(self): """ @jira_ticket CASSANDRA-10504 @jira_ticket CASSANDRA-10427 Test that jmx MBean used by nodetool compactionstats properly updates the progress of a compaction """ cluster = self.cluster cluster.populate(1) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) cluster.start(wait_for_binary_proto=True) # Run a quick stress command to create the keyspace and table node.stress(['write', 'n=1', 'no-warmup']) # Disable compaction on the table node.nodetool('disableautocompaction keyspace1 standard1') node.nodetool('setcompactionthroughput 1') node.stress(['write', 'n=150K', 'no-warmup']) node.flush() # Run a major compaction. This will be the compaction whose # progress we track. node.nodetool_process('compact') # We need to sleep here to give compaction time to start # Why not do something smarter? Because if the bug regresses, # we can't rely on jmx to tell us that compaction started. time.sleep(5) compaction_manager = make_mbean('db', type='CompactionManager') with JolokiaAgent(node) as jmx: progress_string = jmx.read_attribute(compaction_manager, 'CompactionSummary')[0] # Pause in between reads # to allow compaction to move forward time.sleep(2) updated_progress_string = jmx.read_attribute(compaction_manager, 'CompactionSummary')[0] var = 'Compaction@{uuid}(keyspace1, standard1, {progress}/{total})bytes' progress = int(parse.search(var, progress_string).named['progress']) updated_progress = int(parse.search(var, updated_progress_string).named['progress']) logger.debug(progress_string) logger.debug(updated_progress_string) # We want to make sure that the progress is increasing, # and that values other than zero are displayed. assert updated_progress > progress assert progress >= 0 assert updated_progress > 0 # Block until the major compaction is complete # Otherwise nodetool will throw an exception # Give a timeout, in case compaction is broken # and never ends. start = time.time() max_query_timeout = 600 logger.debug("Waiting for compaction to finish:") while (len(jmx.read_attribute(compaction_manager, 'CompactionSummary')) > 0) and ( time.time() - start < max_query_timeout): logger.debug(jmx.read_attribute(compaction_manager, 'CompactionSummary')) time.sleep(2)
def _get_metric(self, metric): mbean = make_mbean("metrics", type="ReadRepair", name=metric) return self.jmx.read_attribute(mbean, "Count")
def commitlog_size(node): commitlog_size_mbean = make_mbean('metrics', type='CommitLog', name='TotalCommitLogSize') with JolokiaAgent(node) as jmx: return jmx.read_attribute(commitlog_size_mbean, 'Value')
def test_mv_metric_mbeans_release(self): """ Test that the right mbeans are created and released when creating mvs """ cluster = self.cluster cluster.set_configuration_options({'enable_materialized_views': 'true'}) cluster.populate(1) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) cluster.start(wait_for_binary_proto=True) node.run_cqlsh(cmds=""" CREATE KEYSPACE mvtest WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor': 1 }; CREATE TABLE mvtest.testtable ( foo int, bar text, baz text, PRIMARY KEY (foo, bar) ); CREATE MATERIALIZED VIEW mvtest.testmv AS SELECT foo, bar, baz FROM mvtest.testtable WHERE foo IS NOT NULL AND bar IS NOT NULL AND baz IS NOT NULL PRIMARY KEY (foo, bar, baz);""") table_memtable_size = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='AllMemtablesHeapSize') table_view_read_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='ViewReadTime') table_view_lock_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='ViewLockAcquireTime') mv_memtable_size = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='AllMemtablesHeapSize') mv_view_read_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='ViewReadTime') mv_view_lock_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='ViewLockAcquireTime') missing_metric_message = "Table metric %s should have been registered after creating table %s" \ "but wasn't!" with JolokiaAgent(node) as jmx: assert jmx.read_attribute(table_memtable_size, "Value") is not None, \ missing_metric_message.format("AllMemtablesHeapSize", "testtable") assert jmx.read_attribute(table_view_read_time, "Count") is not None, \ missing_metric_message.format("ViewReadTime", "testtable") assert jmx.read_attribute(table_view_lock_time, "Count") is not None, \ missing_metric_message.format("ViewLockAcquireTime", "testtable") assert jmx.read_attribute(mv_memtable_size, "Value") is not None, \ missing_metric_message.format("AllMemtablesHeapSize", "testmv") with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_read_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_lock_time, attribute="Count", verbose=False) node.run_cqlsh(cmds="DROP KEYSPACE mvtest;") with JolokiaAgent(node) as jmx: with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=table_memtable_size, attribute="Value", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=table_view_lock_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=table_view_read_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_memtable_size, attribute="Value", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_lock_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_read_time, attribute="Count", verbose=False)
def test_multidatacenter_local_quorum(self): ''' @jira_ticket CASSANDRA-13074 If we do only local datacenters reads in a multidatacenter DES setup, DES should take effect and route around a degraded node ''' def no_cross_dc(scores, cross_dc_nodes): return all('/' + k.address() not in scores for k in cross_dc_nodes) def snitchable(scores_before, scores_after, needed_nodes): return all('/' + k.address() in scores_before and '/' + k.address() in scores_after for k in needed_nodes) cluster = self.cluster cluster.populate([3, 3]) coordinator_node, healthy_node, degraded_node, node4, node5, node6 = cluster.nodelist( ) # increase DES reset/update interval so we clear any cross-DC startup reads faster cluster.set_configuration_options( values={ 'dynamic_snitch_reset_interval_in_ms': 10000, 'dynamic_snitch_update_interval_in_ms': 50, 'phi_convict_threshold': 12 }) remove_perf_disable_shared_mem(coordinator_node) remove_perf_disable_shared_mem(degraded_node) # Delay reads on the degraded node by 50 milliseconds degraded_node.start(jvm_args=[ '-Dcassandra.test.read_iteration_delay_ms=50', '-Dcassandra.allow_unsafe_join=true' ]) cluster.start(wait_for_binary_proto=30, wait_other_notice=True) des = make_mbean('db', type='DynamicEndpointSnitch') read_stage = make_mbean('metrics', type='ThreadPools', path='request', scope='ReadStage', name='CompletedTasks') session = self.patient_exclusive_cql_connection(coordinator_node) session.execute( "CREATE KEYSPACE snitchtestks WITH replication = {'class': 'NetworkTopologyStrategy', 'dc1': 3, 'dc2': 3}" ) session.execute( "CREATE TABLE snitchtestks.tbl1 (key int PRIMARY KEY) WITH speculative_retry = 'NONE' AND dclocal_read_repair_chance = 0.0" ) read_stmt = session.prepare( "SELECT * FROM snitchtestks.tbl1 where key = ?") read_stmt.consistency_level = ConsistencyLevel.LOCAL_QUORUM insert_stmt = session.prepare( "INSERT INTO snitchtestks.tbl1 (key) VALUES (?)") insert_stmt.consistency_level = ConsistencyLevel.ALL with JolokiaAgent(coordinator_node) as jmx: with JolokiaAgent(degraded_node) as bad_jmx: for x in range(0, 300): session.execute(insert_stmt, [x]) cleared = False # Wait for a snitch reset in case any earlier # startup process populated cross-DC read timings while not cleared: scores = jmx.read_attribute(des, 'Scores') cleared = ('/127.0.0.1' in scores and (len(scores) == 1)) or not scores snitchable_count = 0 for x in range(0, 300): degraded_reads_before = bad_jmx.read_attribute( read_stage, 'Value') scores_before = jmx.read_attribute(des, 'Scores') assert_true( no_cross_dc(scores_before, [node4, node5, node6]), "Cross DC scores were present: " + str(scores_before)) future = session.execute_async(read_stmt, [x]) future.result() scores_after = jmx.read_attribute(des, 'Scores') assert_true( no_cross_dc(scores_after, [node4, node5, node6]), "Cross DC scores were present: " + str(scores_after)) if snitchable( scores_before, scores_after, [coordinator_node, healthy_node, degraded_node]): snitchable_count = snitchable_count + 1 # If the DES correctly routed the read around the degraded node, # it shouldn't have another completed read request in metrics assert_equal( degraded_reads_before, bad_jmx.read_attribute(read_stage, 'Value')) else: # sleep to give dynamic snitch time to recalculate scores time.sleep(.1) # check that most reads were snitchable, with some # room allowed in case score recalculation is slow assert_greater_equal(snitchable_count, 250)
def __init__(self, node): assert isinstance(node, Node) self.node = node self.jmx = JolokiaAgent(node) self.mbean = make_mbean("db", type="StorageProxy")
def compact_sstable(self, node, sstable): mbean = make_mbean('db', type='CompactionManager') with JolokiaAgent(node) as jmx: jmx.execute_method(mbean, 'forceUserDefinedCompaction', [sstable])
def test_compactionstats(self): """ @jira_ticket CASSANDRA-10504 @jira_ticket CASSANDRA-10427 Test that jmx MBean used by nodetool compactionstats properly updates the progress of a compaction """ cluster = self.cluster cluster.populate(1) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) cluster.start(wait_for_binary_proto=True) # Run a quick stress command to create the keyspace and table node.stress(['write', 'n=1', 'no-warmup']) # Disable compaction on the table node.nodetool('disableautocompaction keyspace1 standard1') node.nodetool('setcompactionthroughput 1') node.stress(['write', 'n=150K', 'no-warmup']) node.flush() # Run a major compaction. This will be the compaction whose # progress we track. node.nodetool_process('compact') # We need to sleep here to give compaction time to start # Why not do something smarter? Because if the bug regresses, # we can't rely on jmx to tell us that compaction started. time.sleep(5) compaction_manager = make_mbean('db', type='CompactionManager') with JolokiaAgent(node) as jmx: progress_string = jmx.read_attribute(compaction_manager, 'CompactionSummary')[0] # Pause in between reads # to allow compaction to move forward time.sleep(2) updated_progress_string = jmx.read_attribute( compaction_manager, 'CompactionSummary')[0] var = 'Compaction@{uuid}(keyspace1, standard1, {progress}/{total})bytes' progress = int( parse.search(var, progress_string).named['progress']) updated_progress = int( parse.search(var, updated_progress_string).named['progress']) debug(progress_string) debug(updated_progress_string) # We want to make sure that the progress is increasing, # and that values other than zero are displayed. self.assertGreater(updated_progress, progress) self.assertGreaterEqual(progress, 0) self.assertGreater(updated_progress, 0) # Block until the major compaction is complete # Otherwise nodetool will throw an exception # Give a timeout, in case compaction is broken # and never ends. start = time.time() max_query_timeout = 600 debug("Waiting for compaction to finish:") while (len( jmx.read_attribute(compaction_manager, 'CompactionSummary')) > 0) and (time.time() - start < max_query_timeout): debug( jmx.read_attribute(compaction_manager, 'CompactionSummary')) time.sleep(2)
def mv_metric_mbeans_release_test(self): """ Test that the right mbeans are created and released when creating mvs """ cluster = self.cluster cluster.populate(1) node = cluster.nodelist()[0] remove_perf_disable_shared_mem(node) cluster.start(wait_for_binary_proto=True) node.run_cqlsh(cmds=""" CREATE KEYSPACE mvtest WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor': 1 }; CREATE TABLE mvtest.testtable ( foo int, bar text, baz text, PRIMARY KEY (foo, bar) ); CREATE MATERIALIZED VIEW mvtest.testmv AS SELECT foo, bar, baz FROM mvtest.testtable WHERE foo IS NOT NULL AND bar IS NOT NULL AND baz IS NOT NULL PRIMARY KEY (foo, bar, baz);""") table_memtable_size = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='AllMemtablesHeapSize') table_view_read_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='ViewReadTime') table_view_lock_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='ViewLockAcquireTime') mv_memtable_size = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='AllMemtablesHeapSize') mv_view_read_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='ViewReadTime') mv_view_lock_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='ViewLockAcquireTime') missing_metric_message = "Table metric %s should have been registered after creating table %s" \ "but wasn't!" with JolokiaAgent(node) as jmx: self.assertIsNotNone( jmx.read_attribute(table_memtable_size, "Value"), missing_metric_message.format("AllMemtablesHeapSize", "testtable")) self.assertIsNotNone( jmx.read_attribute(table_view_read_time, "Count"), missing_metric_message.format("ViewReadTime", "testtable")) self.assertIsNotNone( jmx.read_attribute(table_view_lock_time, "Count"), missing_metric_message.format("ViewLockAcquireTime", "testtable")) self.assertIsNotNone( jmx.read_attribute(mv_memtable_size, "Value"), missing_metric_message.format("AllMemtablesHeapSize", "testmv")) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=mv_view_read_time, attribute="Count", verbose=False) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=mv_view_lock_time, attribute="Count", verbose=False) node.run_cqlsh(cmds="DROP KEYSPACE mvtest;") with JolokiaAgent(node) as jmx: self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=table_memtable_size, attribute="Value", verbose=False) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=table_view_lock_time, attribute="Count", verbose=False) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=table_view_read_time, attribute="Count", verbose=False) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=mv_memtable_size, attribute="Value", verbose=False) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=mv_view_lock_time, attribute="Count", verbose=False) self.assertRaisesRegexp(Exception, ".*InstanceNotFoundException.*", jmx.read_attribute, mbean=mv_view_read_time, attribute="Count", verbose=False)
def test_mv_metric_mbeans_release(self): """ Test that the right mbeans are created and released when creating mvs """ cluster = self.cluster cluster.set_configuration_options({'enable_materialized_views': 'true'}) cluster.populate(1) node = cluster.nodelist()[0] cluster.start() node.run_cqlsh(cmds=""" CREATE KEYSPACE mvtest WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor': 1 }; CREATE TABLE mvtest.testtable ( foo int, bar text, baz text, PRIMARY KEY (foo, bar) ); CREATE MATERIALIZED VIEW mvtest.testmv AS SELECT foo, bar, baz FROM mvtest.testtable WHERE foo IS NOT NULL AND bar IS NOT NULL AND baz IS NOT NULL PRIMARY KEY (foo, bar, baz);""") table_memtable_size = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='AllMemtablesHeapSize') table_view_read_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='ViewReadTime') table_view_lock_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testtable', name='ViewLockAcquireTime') mv_memtable_size = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='AllMemtablesHeapSize') mv_view_read_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='ViewReadTime') mv_view_lock_time = make_mbean('metrics', type='Table', keyspace='mvtest', scope='testmv', name='ViewLockAcquireTime') missing_metric_message = "Table metric %s should have been registered after creating table %s" \ "but wasn't!" with JolokiaAgent(node) as jmx: assert jmx.read_attribute(table_memtable_size, "Value") is not None, \ missing_metric_message.format("AllMemtablesHeapSize", "testtable") assert jmx.read_attribute(table_view_read_time, "Count") is not None, \ missing_metric_message.format("ViewReadTime", "testtable") assert jmx.read_attribute(table_view_lock_time, "Count") is not None, \ missing_metric_message.format("ViewLockAcquireTime", "testtable") assert jmx.read_attribute(mv_memtable_size, "Value") is not None, \ missing_metric_message.format("AllMemtablesHeapSize", "testmv") with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_read_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_lock_time, attribute="Count", verbose=False) node.run_cqlsh(cmds="DROP KEYSPACE mvtest;") with JolokiaAgent(node) as jmx: with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=table_memtable_size, attribute="Value", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=table_view_lock_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=table_view_read_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_memtable_size, attribute="Value", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_lock_time, attribute="Count", verbose=False) with pytest.raises(Exception, match=".*InstanceNotFoundException.*"): jmx.read_attribute(mbean=mv_view_read_time, attribute="Count", verbose=False)