topology_test.py

import re
import time
import pytest
import logging

from threading import Thread

from cassandra import ConsistencyLevel
from ccmlib.node import TimeoutError, ToolError

from dtest import Tester, create_ks, create_cf, mk_bman_path
from tools.assertions import assert_almost_equal, assert_all, assert_none
from tools.data import insert_c1c2, query_c1c2

since = pytest.mark.since
logger = logging.getLogger(__name__)


class TestTopology(Tester):

    def test_do_not_join_ring(self):
        """
        @jira_ticket CASSANDRA-9034
        Check that AssertionError is not thrown on SizeEstimatesRecorder before node joins ring
        """
        cluster = self.cluster.populate(1)
        node1, = cluster.nodelist()

        node1.start(wait_for_binary_proto=True, join_ring=False,
                    jvm_args=["-Dcassandra.size_recorder_interval=1"])

        # initial delay is 30s
        time.sleep(40)

        node1.stop(gently=False)

    @since('3.0.11')
    def test_size_estimates_multidc(self):
        """
        Test that primary ranges are correctly generated on
        system.size_estimates for multi-dc, multi-ks scenario
        @jira_ticket CASSANDRA-9639
        """
        logger.debug("Creating cluster")
        cluster = self.cluster
        cluster.set_configuration_options(values={'num_tokens': 2})
        cluster.populate([2, 1])
        node1_1, node1_2, node2_1 = cluster.nodelist()

        logger.debug("Setting tokens")
        node1_tokens, node2_tokens, node3_tokens = ['-6639341390736545756,-2688160409776496397',
                                                    '-2506475074448728501,8473270337963525440',
                                                    '-3736333188524231709,8673615181726552074']
        node1_1.set_configuration_options(values={'initial_token': node1_tokens})
        node1_2.set_configuration_options(values={'initial_token': node2_tokens})
        node2_1.set_configuration_options(values={'initial_token': node3_tokens})
        cluster.set_configuration_options(values={'num_tokens': 2})

        logger.debug("Starting cluster")
        cluster.start()

        out, _, _ = node1_1.nodetool('ring')
        logger.debug("Nodetool ring output {}".format(out))

        logger.debug("Creating keyspaces")
        session = self.patient_cql_connection(node1_1)
        create_ks(session, 'ks1', 3)
        create_ks(session, 'ks2', {'dc1': 2})
        create_cf(session, 'ks1.cf1', columns={'c1': 'text', 'c2': 'text'})
        create_cf(session, 'ks2.cf2', columns={'c1': 'text', 'c2': 'text'})

        logger.debug("Refreshing size estimates")
        node1_1.nodetool('refreshsizeestimates')
        node1_2.nodetool('refreshsizeestimates')
        node2_1.nodetool('refreshsizeestimates')

        """
        CREATE KEYSPACE ks1 WITH replication =
            {'class': 'SimpleStrategy', 'replication_factor': '3'}
        CREATE KEYSPACE ks2 WITH replication =
            {'class': 'NetworkTopologyStrategy', 'dc1': '2'}  AND durable_writes = true;

        Datacenter: dc1
        ==========
        Address     Token
                    8473270337963525440
        127.0.0.1   -6639341390736545756
        127.0.0.1   -2688160409776496397
        127.0.0.2   -2506475074448728501
        127.0.0.2   8473270337963525440

        Datacenter: dc2
        ==========
        Address     Token
                    8673615181726552074
        127.0.0.3   -3736333188524231709
        127.0.0.3   8673615181726552074
        """

        logger.debug("Checking node1_1 size_estimates primary ranges")
        session = self.patient_exclusive_cql_connection(node1_1)
        assert_all(session, "SELECT range_start, range_end FROM system.size_estimates "
                            "WHERE keyspace_name = 'ks1'", [['-3736333188524231709', '-2688160409776496397'],
                                                            ['-9223372036854775808', '-6639341390736545756'],
                                                            ['8673615181726552074', '-9223372036854775808']])
        assert_all(session, "SELECT range_start, range_end FROM system.size_estimates "
                            "WHERE keyspace_name = 'ks2'", [['-3736333188524231709', '-2688160409776496397'],
                                                            ['-6639341390736545756', '-3736333188524231709'],
                                                            ['-9223372036854775808', '-6639341390736545756'],
                                                            ['8473270337963525440', '8673615181726552074'],
                                                            ['8673615181726552074', '-9223372036854775808']])

        logger.debug("Checking node1_2 size_estimates primary ranges")
        session = self.patient_exclusive_cql_connection(node1_2)
        assert_all(session, "SELECT range_start, range_end FROM system.size_estimates "
                            "WHERE keyspace_name = 'ks1'", [['-2506475074448728501', '8473270337963525440'],
                                                            ['-2688160409776496397', '-2506475074448728501']])
        assert_all(session, "SELECT range_start, range_end FROM system.size_estimates "
                            "WHERE keyspace_name = 'ks2'", [['-2506475074448728501', '8473270337963525440'],
                                                            ['-2688160409776496397', '-2506475074448728501']])

        logger.debug("Checking node2_1 size_estimates primary ranges")
        session = self.patient_exclusive_cql_connection(node2_1)
        assert_all(session, "SELECT range_start, range_end FROM system.size_estimates "
                            "WHERE keyspace_name = 'ks1'", [['-6639341390736545756', '-3736333188524231709'],
                                                            ['8473270337963525440', '8673615181726552074']])
        assert_none(session, "SELECT range_start, range_end FROM system.size_estimates "
                             "WHERE keyspace_name = 'ks2'")

    def test_simple_removenode(self):
        """ test removenode """
        cluster = self.cluster
        cluster.populate(3)
        cluster.start(jvm_args=["-Dcassandra.size_recorder_interval=1"])
        node1, node2, node3 = cluster.nodelist()

        node1.stress(['write', 'n=10K', 'no-warmup', '-rate', 'threads=8', '-schema', 'replication(factor=2)'])

        node3_id = node3.nodetool('info').stdout[25:61]
        node3.stop(wait_other_notice=True)
        node1.nodetool('removenode ' + node3_id)

        node1.stress(['read', 'n=10K', 'no-warmup', '-rate', 'threads=8'])

    def test_simple_decommission(self):
        """
        @jira_ticket CASSANDRA-9912
        Check that AssertionError is not thrown on SizeEstimatesRecorder after node is decommissioned
        """
        cluster = self.cluster
        cluster.populate(3)
        cluster.start(jvm_args=["-Dcassandra.size_recorder_interval=1"])
        node1, node2, node3 = cluster.nodelist()

        session = self.patient_cql_connection(node1)

        if cluster.version() >= '2.2':
            # reduce system_distributed RF to 2 so we don't require forceful decommission
            session.execute("ALTER KEYSPACE system_distributed WITH REPLICATION = {'class':'SimpleStrategy', 'replication_factor':'2'};")

        # write some data
        node1.stress(['write', 'n=10K', 'no-warmup', '-rate', 'threads=8'])

        # Decommission node and wipe its data
        node2.decommission()
        node2.stop()

        # This sleep is here to give the cluster time to hit the AssertionError
        # described in 9912. Do not remove it.
        time.sleep(10)

    @pytest.mark.skip(reason='Hangs on CI for 2.1')
    def test_concurrent_decommission_not_allowed(self):
        """
        Test concurrent decommission is not allowed
        """
        cluster = self.cluster
        cluster.set_configuration_options(values={'stream_throughput_outbound_megabits_per_sec': 1})
        cluster.populate(2).start()
        node1, node2 = cluster.nodelist()

        session = self.patient_cql_connection(node2)
        create_ks(session, 'ks', 1)
        create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
        insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ALL)

        mark = node2.mark_log()

        def decommission():
            node2.nodetool('decommission')

        # Launch first decommission in a external thread
        t = Thread(target=decommission)
        t.start()

        # Make sure first decommission is initialized before second decommission
        node2.watch_log_for('DECOMMISSIONING', filename='debug.log')

        # Launch a second decommission, should fail
        with pytest.raises(ToolError):
            node2.nodetool('decommission')

        # Check data is correctly forwarded to node1 after node2 is decommissioned
        t.join()
        node2.watch_log_for('DECOMMISSIONED', from_mark=mark)
        session = self.patient_cql_connection(node1)
        session.execute('USE ks')
        for n in range(0, 10000):
            query_c1c2(session, n, ConsistencyLevel.ONE)

    @since('3.10')
    def test_resumable_decommission(self):
        """
        @jira_ticket CASSANDRA-12008

        Test decommission operation is resumable
        """
        self.fixture_dtest_setup.ignore_log_patterns = [r'Streaming error occurred',
                                                        r'Error while decommissioning node',
                                                        r'Remote peer 127.0.0.2 failed stream session',
                                                        r'Remote peer \/?127.0.0.2:7000 failed stream session']
        cluster = self.cluster
        cluster.set_configuration_options(values={'stream_throughput_outbound_megabits_per_sec': 1})
        cluster.populate(3, install_byteman=True).start()
        node1, node2, node3 = cluster.nodelist()

        session = self.patient_cql_connection(node2)
        # reduce system_distributed RF to 2 so we don't require forceful decommission
        session.execute("ALTER KEYSPACE system_distributed WITH REPLICATION = {'class':'SimpleStrategy', 'replication_factor':'2'};")
        create_ks(session, 'ks', 2)
        create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
        insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ALL)

        # Execute first rebuild, should fail
        with pytest.raises(ToolError):
            if cluster.version() >= '4.0':
                script = [mk_bman_path('4.0/decommission_failure_inject.btm')]
            else:
                script = [mk_bman_path('pre4.0/decommission_failure_inject.btm')]
            node2.byteman_submit(script)
            node2.nodetool('decommission')

        # Make sure previous ToolError is due to decommission
        node2.watch_log_for('Error while decommissioning node')

        # Decommission again
        mark = node2.mark_log()
        node2.nodetool('decommission')

        # Check decommision is done and we skipped transfereed ranges
        node2.watch_log_for('DECOMMISSIONED', from_mark=mark)
        node2.grep_log("Skipping transferred range .* of keyspace ks, endpoint {}".format(node2.address_for_current_version_slashy()), filename='debug.log')

        # Check data is correctly forwarded to node1 and node3
        cluster.remove(node2)
        node3.stop(gently=False)
        session = self.patient_exclusive_cql_connection(node1)
        session.execute('USE ks')
        for i in range(0, 10000):
            query_c1c2(session, i, ConsistencyLevel.ONE)
        node1.stop(gently=False)
        node3.start()
        session.shutdown()
        mark = node3.mark_log()
        node3.watch_log_for('Starting listening for CQL clients', from_mark=mark)
        session = self.patient_exclusive_cql_connection(node3)
        session.execute('USE ks')
        for i in range(0, 10000):
            query_c1c2(session, i, ConsistencyLevel.ONE)

    @pytest.mark.no_vnodes
    def test_movement(self):
        cluster = self.cluster

        # Create an unbalanced ring
        cluster.populate(3, tokens=[0, 2**48, 2**62]).start()
        node1, node2, node3 = cluster.nodelist()

        session = self.patient_cql_connection(node1)
        create_ks(session, 'ks', 1)
        create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})

        insert_c1c2(session, n=30000, consistency=ConsistencyLevel.ONE)

        cluster.flush()

        # Move nodes to balance the cluster
        def move_node(node, token):
            mark = node.mark_log()
            node.move(token)  # can't assume 0 is balanced with m3p
            node.watch_log_for('{} state jump to NORMAL'.format(node.address_for_current_version()), from_mark=mark, timeout=180)
            time.sleep(3)

        balancing_tokens = cluster.balanced_tokens(3)

        move_node(node1, balancing_tokens[0])
        move_node(node2, balancing_tokens[1])
        move_node(node3, balancing_tokens[2])

        time.sleep(1)
        cluster.cleanup()
        for node in cluster.nodelist():
            # after moving nodes we need to relocate any tokens in the wrong places, and after doing that
            # we might have overlapping tokens on the disks, so run a major compaction to get balance even
            if cluster.version() >= '3.2':
                node.nodetool("relocatesstables")
            node.nodetool("compact")

        # Check we can get all the keys
        for n in range(0, 30000):
            query_c1c2(session, n, ConsistencyLevel.ONE)

        # Now the load should be basically even
        sizes = [node.data_size() for node in [node1, node2, node3]]

        assert_almost_equal(sizes[0], sizes[1], error=0.05)
        assert_almost_equal(sizes[0], sizes[2], error=0.05)
        assert_almost_equal(sizes[1], sizes[2], error=0.05)

    @pytest.mark.no_vnodes
    def test_decommission(self):
        cluster = self.cluster

        tokens = cluster.balanced_tokens(4)
        cluster.populate(4, tokens=tokens).start()
        node1, node2, node3, node4 = cluster.nodelist()

        session = self.patient_cql_connection(node1)
        create_ks(session, 'ks', 2)
        create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})

        insert_c1c2(session, n=30000, consistency=ConsistencyLevel.QUORUM)

        cluster.flush()
        sizes = [node.data_size() for node in cluster.nodelist() if node.is_running()]
        init_size = sizes[0]
        assert_almost_equal(*sizes)

        time.sleep(.5)
        node4.decommission()
        node4.stop()
        cluster.cleanup()
        time.sleep(.5)

        # Check we can get all the keys
        for n in range(0, 30000):
            query_c1c2(session, n, ConsistencyLevel.QUORUM)

        sizes = [node.data_size() for node in cluster.nodelist() if node.is_running()]
        logger.debug(sizes)
        assert_almost_equal(sizes[0], sizes[1])
        assert_almost_equal((2.0 / 3.0) * sizes[0], sizes[2])
        assert_almost_equal(sizes[2], init_size)

    @pytest.mark.no_vnodes
    def test_move_single_node(self):
        """ Test moving a node in a single-node cluster (#4200) """
        cluster = self.cluster

        # Create an unbalanced ring
        cluster.populate(1, tokens=[0]).start()
        node1 = cluster.nodelist()[0]
        time.sleep(0.2)

        session = self.patient_cql_connection(node1)
        create_ks(session, 'ks', 1)
        create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})

        insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ONE)

        cluster.flush()

        node1.move(2**25)
        time.sleep(1)

        cluster.cleanup()

        # Check we can get all the keys
        for n in range(0, 10000):
            query_c1c2(session, n, ConsistencyLevel.ONE)

    @since('3.0')
    def test_decommissioned_node_cant_rejoin(self):
        """
        @jira_ticket CASSANDRA-8801

        Test that a decommissioned node can't rejoin the cluster by:

        - creating a cluster,
        - decommissioning a node, and
        - asserting that the "decommissioned node won't rejoin" error is in the
        logs for that node and
        - asserting that the node is not running.
        """
        rejoin_err = 'This node was decommissioned and will not rejoin the ring'
        self.fixture_dtest_setup.ignore_log_patterns = list(self.fixture_dtest_setup.ignore_log_patterns) + [
            rejoin_err]

        self.cluster.populate(3).start()
        node1, node2, node3 = self.cluster.nodelist()

        logger.debug('decommissioning...')
        node3.decommission(force=self.cluster.version() >= '4.0')
        logger.debug('stopping...')
        node3.stop()
        logger.debug('attempting restart...')
        node3.start(wait_other_notice=False)
        timedout = False
        try:
            # usually takes 3 seconds, so give it a generous 15
            node3.watch_log_for(rejoin_err, timeout=15)
        except TimeoutError:
            # TimeoutError is not very helpful to the reader of the test output;
            # let that pass and move on to string assertion below
            timedout = True

        n3errors = node3.grep_log_for_errors()
        if len(n3errors) == 0 and timedout:
            raise TimeoutError("timed out and did not find log entry: " + rejoin_err)

        assert re.search(rejoin_err,
                         '\n'.join(['\n'.join(err_list) for err_list in n3errors]), re.MULTILINE)

        # Give the node some time to shut down once it has detected
        # its invalid state. If it doesn't shut down in the 30 seconds,
        # consider filing a bug. It shouldn't take more than 10, in most cases.
        start = time.time()
        while start + 30 > time.time() and node3.is_running():
            time.sleep(1)

        assert not node3.is_running()

    @since('3.0')
    def test_crash_during_decommission(self):
        """
        If a node crashes whilst another node is being decommissioned,
        upon restarting the crashed node should not have invalid entries
        for the decommissioned node
        @jira_ticket CASSANDRA-10231
        """
        cluster = self.cluster
        self.fixture_dtest_setup.ignore_log_patterns = [r'Streaming error occurred', 'Stream failed']
        cluster.populate(3).start()

        node1, node2 = cluster.nodelist()[0:2]
        session = self.patient_cql_connection(node1)
        session.execute("ALTER KEYSPACE system_distributed WITH REPLICATION = {'class':'SimpleStrategy', 'replication_factor':'2'};")

        t = DecommissionInParallel(node1)
        t.start()

        node1.watch_log_for("DECOMMISSIONING", filename='debug.log')
        null_status_pattern = re.compile(r".N(?:\s*)127\.0\.0\.1(?:.*)null(?:\s*)rack1")
        while t.is_alive():
            out = self.show_status(node2)
            if null_status_pattern.search(out):
                logger.debug("Matched null status entry")
                break
            logger.debug("Restarting node2")
            node2.stop(gently=False)
            node2.start(wait_for_binary_proto=True, wait_other_notice=False)

        logger.debug("Waiting for decommission to complete")
        t.join()
        self.show_status(node2)

        logger.debug("Sleeping for 30 seconds to allow gossip updates")
        time.sleep(30)
        out = self.show_status(node2)
        assert not null_status_pattern.search(out)

    @since('3.12')
    @pytest.mark.resource_intensive
    def test_stop_decommission_too_few_replicas_multi_dc(self):
        """
        Decommission should fail when it would result in the number of live replicas being less than
        the replication factor. --force should bypass this requirement.
        @jira_ticket CASSANDRA-12510
        @expected_errors ToolError when # nodes will drop below configured replicas in NTS/SimpleStrategy
        """

        # we need to ignore this error log message which is emitted during the test 
        # because dtest framework which reads the logs would evaluate the node is errorneous and 
        # it would terminate it prematurely
        self.fixture_dtest_setup.ignore_log_patterns = (r'.*Not enough live nodes to maintain replication factor*')

        cluster = self.cluster
        cluster.populate([2, 2]).start()
        node1, node2, node3, node4 = self.cluster.nodelist()
        session = self.patient_cql_connection(node2)
        session.execute("ALTER KEYSPACE system_distributed WITH REPLICATION = {'class':'SimpleStrategy', 'replication_factor':'2'};")
        session.cluster.control_connection.wait_for_schema_agreement()
        create_ks(session, 'ks', {'dc1': 2, 'dc2': 2})
        with pytest.raises(ToolError):
            node4.nodetool('decommission')

        session.execute('DROP KEYSPACE ks')
        session.cluster.control_connection.wait_for_schema_agreement()
        create_ks(session, 'ks2', 4)
        with pytest.raises(ToolError):
            node4.nodetool('decommission')

        self.fixture_dtest_setup.ignore_log_patterns = ()

        node4.nodetool('decommission --force')
        decommissioned = node4.watch_log_for("DECOMMISSIONED", timeout=120)
        assert decommissioned, "Node failed to decommission when passed --force"

    def show_status(self, node):
        out, _, _ = node.nodetool('status')
        logger.debug("Status as reported by node {}".format(node.address()))
        logger.debug(out)
        return out


class DecommissionInParallel(Thread):

    def __init__(self, node):
        Thread.__init__(self)
        self.node = node

    def run(self):
        node = self.node
        mark = node.mark_log()
        try:
            out, err, _ = node.nodetool("decommission")
            node.watch_log_for("DECOMMISSIONED", from_mark=mark)
            logger.debug(out)
            logger.debug(err)
        except ToolError as e:
            logger.debug("Decommission failed with exception: " + str(e))
            pass