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 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 _do_hinted_handoff(self, node1, node2, enabled, keyspace='ks'):
"""
Test that if we stop one node the other one
will store hints only when hinted handoff is enabled
"""
session = self.patient_exclusive_cql_connection(node1)
create_ks(session, keyspace, 2)
create_c1c2_table(self, session)
node2.stop(wait_other_notice=True)
insert_c1c2(session, n=100, consistency=ConsistencyLevel.ONE)
log_mark = node1.mark_log()
node2.start(wait_other_notice=True)
if enabled:
node1.watch_log_for(["Finished hinted"], from_mark=log_mark, timeout=120)
node1.stop(wait_other_notice=True)
# Check node2 for all the keys that should have been delivered via HH if enabled or not if not enabled
session = self.patient_exclusive_cql_connection(node2, keyspace=keyspace)
for n in xrange(0, 100):
if enabled:
query_c1c2(session, n, ConsistencyLevel.ONE)
else:
query_c1c2(session, n, ConsistencyLevel.ONE, tolerate_missing=True, must_be_missing=True)
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)
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)
def quorum_available_during_failure_test(self):
CL = ConsistencyLevel.QUORUM
RF = 3
debug("Creating a ring")
cluster = self.cluster
if DISABLE_VNODES:
cluster.populate(3).start()
else:
tokens = cluster.balanced_tokens(3)
cluster.populate(3, tokens=tokens).start()
node1, node2, node3 = cluster.nodelist()
debug("Set to talk to node 2")
session = self.patient_cql_connection(node2)
self.create_ks(session, 'ks', RF)
create_c1c2_table(self, session)
debug("Generating some data")
insert_c1c2(session, n=100, consistency=CL)
debug("Taking down node1")
node1.stop(wait_other_notice=True)
debug("Reading back data.")
for n in xrange(100):
query_c1c2(session, n, CL)
def _do_hinted_handoff(self, node1, node2, enabled, keyspace='ks'):
"""
Test that if we stop one node the other one
will store hints only when hinted handoff is enabled
"""
session = self.patient_exclusive_cql_connection(node1)
create_ks(session, keyspace, 2)
create_c1c2_table(self, session)
node2.stop(wait_other_notice=True)
insert_c1c2(session, n=100, consistency=ConsistencyLevel.ONE)
log_mark = node1.mark_log()
node2.start()
if enabled:
node1.watch_log_for(["Finished hinted"],
from_mark=log_mark,
timeout=120)
node1.stop(wait_other_notice=True)
# Check node2 for all the keys that should have been delivered via HH if enabled or not if not enabled
session = self.patient_exclusive_cql_connection(node2,
keyspace=keyspace)
for n in range(0, 100):
if enabled:
query_c1c2(session, n, ConsistencyLevel.ONE)
else:
query_c1c2(session,
n,
ConsistencyLevel.ONE,
tolerate_missing=True,
must_be_missing=True)
def readrepair_test(self):
cluster = self.cluster
cluster.set_configuration_options(
values={'hinted_handoff_enabled': False})
if DISABLE_VNODES:
cluster.populate(2).start()
else:
tokens = cluster.balanced_tokens(2)
cluster.populate(2, tokens=tokens).start()
node1, node2 = cluster.nodelist()
session = self.patient_cql_connection(node1)
self.create_ks(session, 'ks', 2)
create_c1c2_table(self, session, read_repair=1.0)
node2.stop(wait_other_notice=True)
insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ONE)
node2.start(wait_other_notice=True)
# query everything to cause RR
for n in xrange(0, 10000):
query_c1c2(session, n, ConsistencyLevel.QUORUM)
node1.stop(wait_other_notice=True)
# Check node2 for all the keys that should have been repaired
session = self.patient_cql_connection(node2, keyspace='ks')
for n in xrange(0, 10000):
query_c1c2(session, n, ConsistencyLevel.ONE)
def quorum_available_during_failure_test(self):
CL = ConsistencyLevel.QUORUM
RF = 3
debug("Creating a ring")
cluster = self.cluster
if DISABLE_VNODES:
cluster.populate(3).start()
else:
tokens = cluster.balanced_tokens(3)
cluster.populate(3, tokens=tokens).start()
node1, node2, node3 = cluster.nodelist()
debug("Set to talk to node 2")
session = self.patient_cql_connection(node2)
create_ks(session, 'ks', RF)
create_c1c2_table(self, session)
debug("Generating some data")
insert_c1c2(session, n=100, consistency=CL)
debug("Taking down node1")
node1.stop(wait_other_notice=True)
debug("Reading back data.")
for n in xrange(100):
query_c1c2(session, n, CL)
def readrepair_test(self):
cluster = self.cluster
cluster.set_configuration_options(values={'hinted_handoff_enabled': False})
if DISABLE_VNODES:
cluster.populate(2).start()
else:
tokens = cluster.balanced_tokens(2)
cluster.populate(2, tokens=tokens).start()
node1, node2 = cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 2)
create_c1c2_table(self, session, read_repair=1.0)
node2.stop(wait_other_notice=True)
insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ONE)
node2.start(wait_for_binary_proto=True, wait_other_notice=True)
# query everything to cause RR
for n in xrange(0, 10000):
query_c1c2(session, n, ConsistencyLevel.QUORUM)
node1.stop(wait_other_notice=True)
# Check node2 for all the keys that should have been repaired
session = self.patient_cql_connection(node2, keyspace='ks')
for n in xrange(0, 10000):
query_c1c2(session, n, ConsistencyLevel.ONE)
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(wait_other_notice=True)
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 = ['./byteman/4.0/decommission_failure_inject.btm']
else:
script = ['./byteman/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)
def resumable_decommission_test(self):
"""
@jira_ticket CASSANDRA-12008
Test decommission operation is resumable
"""
self.ignore_log_patterns = [
r'Streaming error occurred', r'Error while decommissioning node',
r'Remote peer 127.0.0.2 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(wait_other_notice=True)
node1, node2, node3 = cluster.nodelist()
session = self.patient_cql_connection(node2)
self.create_ks(session, 'ks', 2)
self.create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ALL)
# Execute first rebuild, should fail
with self.assertRaises(ToolError):
script = ['./byteman/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 /127.0.0.3",
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 xrange(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 xrange(0, 10000):
query_c1c2(session, i, ConsistencyLevel.ONE)
def test_consistent_reads_after_bootstrap(self):
logger.debug("Creating a ring")
cluster = self.cluster
cluster.set_configuration_options(
values={
'hinted_handoff_enabled': False,
'write_request_timeout_in_ms': 60000,
'read_request_timeout_in_ms': 60000,
'dynamic_snitch_badness_threshold': 0.0
})
cluster.set_batch_commitlog(enabled=True)
cluster.populate(2)
node1, node2 = cluster.nodelist()
cluster.start(wait_for_binary_proto=True, wait_other_notice=True)
logger.debug("Set to talk to node 2")
n2session = self.patient_cql_connection(node2)
create_ks(n2session, 'ks', 2)
create_c1c2_table(self, n2session)
logger.debug("Generating some data for all nodes")
insert_c1c2(n2session,
keys=list(range(10, 20)),
consistency=ConsistencyLevel.ALL)
node1.flush()
logger.debug("Taking down node1")
node1.stop(wait_other_notice=True)
logger.debug("Writing data to only node2")
insert_c1c2(n2session,
keys=list(range(30, 1000)),
consistency=ConsistencyLevel.ONE)
node2.flush()
logger.debug("Restart node1")
node1.start(wait_other_notice=True)
logger.debug("Bootstraping node3")
node3 = new_node(cluster)
node3.start(wait_for_binary_proto=True)
n3session = self.patient_cql_connection(node3)
n3session.execute("USE ks")
logger.debug("Checking that no data was lost")
for n in range(10, 20):
query_c1c2(n3session, n, ConsistencyLevel.ALL)
for n in range(30, 1000):
query_c1c2(n3session, n, ConsistencyLevel.ALL)
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 resumable_decommission_test(self):
"""
@jira_ticket CASSANDRA-12008
Test decommission operation is resumable
"""
self.ignore_log_patterns = [r'Streaming error occurred', r'Error while decommissioning node', r'Remote peer 127.0.0.2 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(wait_other_notice=True)
node1, node2, node3 = cluster.nodelist()
session = self.patient_cql_connection(node2)
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 self.assertRaises(ToolError):
script = ['./byteman/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 /127.0.0.3", 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 xrange(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 xrange(0, 10000):
query_c1c2(session, i, ConsistencyLevel.ONE)
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)
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 test_non_local_read(self):
""" This test reads from a coordinator we know has no copy of the data """
cluster = self.cluster
cluster.populate(3).start()
node1, node2, node3 = cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 2)
create_c1c2_table(self, session)
# insert and get at CL.QUORUM (since RF=2, node1 won't have all key locally)
insert_c1c2(session, n=1000, consistency=ConsistencyLevel.QUORUM)
for n in range(0, 1000):
query_c1c2(session, n, ConsistencyLevel.QUORUM)
def hintedhandoff_decom_test(self):
self.cluster.populate(4).start(wait_for_binary_proto=True)
[node1, node2, node3, node4] = self.cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 2)
create_c1c2_table(self, session)
node4.stop(wait_other_notice=True)
insert_c1c2(session, n=100, consistency=ConsistencyLevel.ONE)
node1.decommission()
node4.start(wait_for_binary_proto=True)
node2.decommission()
node3.decommission()
time.sleep(5)
for x in xrange(0, 100):
query_c1c2(session, x, ConsistencyLevel.ONE)
def hintedhandoff_decom_test(self):
self.cluster.populate(4).start(wait_for_binary_proto=True)
[node1, node2, node3, node4] = self.cluster.nodelist()
session = self.patient_cql_connection(node1)
self.create_ks(session, 'ks', 2)
create_c1c2_table(self, session)
node4.stop(wait_other_notice=True)
insert_c1c2(session, n=100, consistency=ConsistencyLevel.ONE)
node1.decommission()
node4.start(wait_for_binary_proto=True)
node2.decommission()
node3.decommission()
time.sleep(5)
for x in xrange(0, 100):
query_c1c2(session, x, ConsistencyLevel.ONE)
def test_non_local_read(self):
""" This test reads from a coordinator we know has no copy of the data """
cluster = self.cluster
cluster.populate(3).start()
node1, node2, node3 = cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 2)
create_c1c2_table(self, session)
# insert and get at CL.QUORUM (since RF=2, node1 won't have all key locally)
insert_c1c2(session, n=1000, consistency=ConsistencyLevel.QUORUM)
for n in range(0, 1000):
query_c1c2(session, n, ConsistencyLevel.QUORUM)
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)
def test_consistent_reads_after_move(self):
logger.debug("Creating a ring")
cluster = self.cluster
cluster.set_configuration_options(
values={
'hinted_handoff_enabled': False,
'write_request_timeout_in_ms': 60000,
'read_request_timeout_in_ms': 60000,
'dynamic_snitch_badness_threshold': 0.0
})
cluster.set_batch_commitlog(enabled=True)
cluster.populate(3, tokens=[0, 2**48, 2**62]).start()
node1, node2, node3 = cluster.nodelist()
logger.debug("Set to talk to node 2")
n2session = self.patient_cql_connection(node2)
create_ks(n2session, 'ks', 2)
create_c1c2_table(self, n2session)
logger.debug("Generating some data for all nodes")
insert_c1c2(n2session,
keys=list(range(10, 20)),
consistency=ConsistencyLevel.ALL)
node1.flush()
logger.debug("Taking down node1")
node1.stop(wait_other_notice=True)
logger.debug("Writing data to node2")
insert_c1c2(n2session,
keys=list(range(30, 1000)),
consistency=ConsistencyLevel.ONE)
node2.flush()
logger.debug("Restart node1")
node1.start()
logger.debug("Move token on node3")
node3.move(2)
logger.debug("Checking that no data was lost")
for n in range(10, 20):
query_c1c2(n2session, n, ConsistencyLevel.ALL, max_attempts=3)
for n in range(30, 1000):
query_c1c2(n2session, n, ConsistencyLevel.ALL, max_attempts=3)
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()
# 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])
assert_almost_equal(sizes[0], sizes[2])
assert_almost_equal(sizes[1], sizes[2])
def test_consistent_reads_after_bootstrap(self):
logger.debug("Creating a ring")
cluster = self.cluster
cluster.set_configuration_options(values={'hinted_handoff_enabled': False,
'write_request_timeout_in_ms': 60000,
'read_request_timeout_in_ms': 60000,
'dynamic_snitch_badness_threshold': 0.0})
cluster.set_batch_commitlog(enabled=True)
cluster.populate(2)
node1, node2 = cluster.nodelist()
cluster.start(wait_for_binary_proto=True, wait_other_notice=True)
logger.debug("Set to talk to node 2")
n2session = self.patient_cql_connection(node2)
create_ks(n2session, 'ks', 2)
create_c1c2_table(self, n2session)
logger.debug("Generating some data for all nodes")
insert_c1c2(n2session, keys=list(range(10, 20)), consistency=ConsistencyLevel.ALL)
node1.flush()
logger.debug("Taking down node1")
node1.stop(wait_other_notice=True)
logger.debug("Writing data to only node2")
insert_c1c2(n2session, keys=list(range(30, 1000)), consistency=ConsistencyLevel.ONE)
node2.flush()
logger.debug("Restart node1")
node1.start(wait_other_notice=True)
logger.debug("Bootstraping node3")
node3 = new_node(cluster)
node3.start(wait_for_binary_proto=True)
n3session = self.patient_cql_connection(node3)
n3session.execute("USE ks")
logger.debug("Checking that no data was lost")
for n in range(10, 20):
query_c1c2(n3session, n, ConsistencyLevel.ALL)
for n in range(30, 1000):
query_c1c2(n3session, n, ConsistencyLevel.ALL)
def _test_streaming(self, op_zerocopy, op_partial, num_partial, num_zerocopy,
compaction_strategy='LeveledCompactionStrategy', num_keys=1000, rf=3, num_nodes=3):
keys = num_keys
cluster = self.cluster
tokens = cluster.balanced_tokens(num_nodes)
cluster.set_configuration_options(values={'endpoint_snitch': 'org.apache.cassandra.locator.PropertyFileSnitch'})
cluster.set_configuration_options(values={'num_tokens': 1})
cluster.populate(num_nodes)
nodes = cluster.nodelist()
for i in range(0, len(nodes)):
nodes[i].set_configuration_options(values={'initial_token': tokens[i]})
cluster.start(wait_for_binary_proto=True)
session = self.patient_cql_connection(nodes[0])
create_ks(session, name='ks2', rf=rf)
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'},
compaction_strategy=compaction_strategy)
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
session_n2 = self.patient_exclusive_cql_connection(nodes[1])
session_n2.execute("TRUNCATE system.available_ranges;")
mark = nodes[1].mark_log()
nodes[1].nodetool('rebuild -ks ks2')
nodes[1].watch_log_for('Completed submission of build tasks', filename='debug.log', timeout=120)
zerocopy_streamed_sstable = len(
nodes[1].grep_log('.*CassandraEntireSSTableStreamReader.*?Finished receiving Data.*', filename='debug.log',
from_mark=mark))
partial_streamed_sstable = len(
nodes[1].grep_log('.*CassandraStreamReader.*?Finished receiving file.*', filename='debug.log',
from_mark=mark))
assert op_zerocopy(zerocopy_streamed_sstable, num_zerocopy), "%s %s %s" % (num_zerocopy, opmap.get(op_zerocopy),
zerocopy_streamed_sstable)
assert op_partial(partial_streamed_sstable, num_partial), "%s %s %s" % (num_partial, op_partial,
partial_streamed_sstable)
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()
# 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])
assert_almost_equal(sizes[0], sizes[2])
assert_almost_equal(sizes[1], sizes[2])
def concurrent_decommission_not_allowed_test(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(wait_other_notice=True)
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 self.assertRaises(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 xrange(0, 10000):
query_c1c2(session, n, ConsistencyLevel.ONE)
def consistent_reads_after_move_test(self):
debug("Creating a ring")
cluster = self.cluster
cluster.set_configuration_options(values={'hinted_handoff_enabled': False, 'write_request_timeout_in_ms': 60000,
'read_request_timeout_in_ms': 60000, 'dynamic_snitch_badness_threshold': 0.0})
cluster.set_batch_commitlog(enabled=True)
cluster.populate(3, tokens=[0, 2**48, 2**62]).start()
node1, node2, node3 = cluster.nodelist()
debug("Set to talk to node 2")
n2session = self.patient_cql_connection(node2)
create_ks(n2session, 'ks', 2)
create_c1c2_table(self, n2session)
debug("Generating some data for all nodes")
insert_c1c2(n2session, keys=range(10, 20), consistency=ConsistencyLevel.ALL)
node1.flush()
debug("Taking down node1")
node1.stop(wait_other_notice=True)
debug("Writing data to node2")
insert_c1c2(n2session, keys=range(30, 1000), consistency=ConsistencyLevel.ONE)
node2.flush()
debug("Restart node1")
node1.start(wait_other_notice=True)
debug("Move token on node3")
node3.move(2)
debug("Checking that no data was lost")
for n in xrange(10, 20):
query_c1c2(n2session, n, ConsistencyLevel.ALL)
for n in xrange(30, 1000):
query_c1c2(n2session, n, ConsistencyLevel.ALL)
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(wait_other_notice=True)
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)
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_multiple_repair(self):
"""
* Launch a three node cluster
* Create a keyspace with RF 3 and a table
* Insert 49 rows
* Stop node3
* Insert 50 more rows
* Restart node3
* Issue an incremental repair on node3
* Stop node2
* Insert a final50 rows
* Restart node2
* Issue an incremental repair on node2
* Replace node3 with a new node
* Verify data integrity
# TODO: Several more verifications of data need to be interspersed throughout the test. The final assertion is insufficient.
@jira_ticket CASSANDRA-10644
"""
cluster = self.cluster
cluster.populate(3).start()
node1, node2, node3 = cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 3)
if cluster.version() < '4.0':
create_cf(session,
'cf',
read_repair=0.0,
columns={
'c1': 'text',
'c2': 'text'
})
else:
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
logger.debug("insert data")
insert_c1c2(session,
keys=list(range(1, 50)),
consistency=ConsistencyLevel.ALL)
node1.flush()
logger.debug("bringing down node 3")
node3.flush()
node3.stop(gently=False)
logger.debug("inserting additional data into node 1 and 2")
insert_c1c2(session,
keys=list(range(50, 100)),
consistency=ConsistencyLevel.TWO)
node1.flush()
node2.flush()
logger.debug("restarting and repairing node 3")
node3.start(wait_for_binary_proto=True)
if cluster.version() >= "2.2":
node3.repair()
else:
node3.nodetool("repair -par -inc")
# wait stream handlers to be closed on windows
# after session is finished (See CASSANDRA-10644)
if is_win:
time.sleep(2)
logger.debug("stopping node 2")
node2.stop(gently=False)
logger.debug("inserting data in nodes 1 and 3")
insert_c1c2(session,
keys=list(range(100, 150)),
consistency=ConsistencyLevel.TWO)
node1.flush()
node3.flush()
logger.debug("start and repair node 2")
node2.start(wait_for_binary_proto=True)
if cluster.version() >= "2.2":
node2.repair()
else:
node2.nodetool("repair -par -inc")
logger.debug("replace node and check data integrity")
node3.stop(gently=False)
node5 = Node('node5', cluster, True, ('127.0.0.5', 9160),
('127.0.0.5', 7000), '7500', '0', None,
('127.0.0.5', 9042))
cluster.add(node5, False, data_center="dc1")
node5.start(replace_address='127.0.0.3')
assert_one(session, "SELECT COUNT(*) FROM ks.cf LIMIT 200", [149])
def rebuild_with_specific_sources_test(self):
"""
@jira_ticket CASSANDRA-9875
Verifies that an operator can specify specific sources to use
when rebuilding.
1. Set up a 2 node cluster across dc1 and dc2
2. Create new keyspaces with replication factor 2 (one replica in each datacenter)
4. Populate nodes with data
5. Create a new node in dc3 and update the keyspace replication
6. Run rebuild on the new node with a specific source in dc2
7. Assert that streaming only occurred between the new node and the specified source
8. Assert that the rebuild was successful by checking the data
"""
keys = 1000
cluster = self.cluster
tokens = cluster.balanced_tokens_across_dcs(['dc1', 'dc2', 'dc3'])
cluster.set_configuration_options(values={
'endpoint_snitch':
'org.apache.cassandra.locator.PropertyFileSnitch'
})
cluster.set_configuration_options(values={'num_tokens': 1})
cluster.populate([1, 1], tokens=tokens[:2])
node1, node2 = cluster.nodelist()
cluster.start(wait_for_binary_proto=True)
# populate data in dc1, dc2
session = self.patient_exclusive_cql_connection(node1)
# ks1 will be rebuilt in node3
create_ks(session, 'ks1', {'dc1': 1, 'dc2': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
# ks2 will not be rebuilt in node3
create_ks(session, 'ks2', {'dc1': 1, 'dc2': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
session.shutdown()
# bootstrap a new node in dc3 with auto_bootstrap: false
node3 = cluster.create_node('node3',
False, ('127.0.0.3', 9160),
('127.0.0.3', 7000),
'7300',
'2002',
tokens[2],
binary_interface=('127.0.0.3', 9042))
cluster.add(node3, False, data_center='dc3')
node3.start(wait_other_notice=True, wait_for_binary_proto=True)
# wait for snitch to reload
time.sleep(60)
# alter keyspace to replicate to dc3
session = self.patient_exclusive_cql_connection(node3)
session.execute(
"ALTER KEYSPACE ks1 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1, 'dc3':1};"
)
session.execute(
"ALTER KEYSPACE ks2 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1, 'dc3':1};"
)
session.execute('USE ks1')
node2_address = node2.network_interfaces['binary'][0]
node3_address = node3.network_interfaces['binary'][0]
# rebuild only ks1, restricting the source to node2
node3.nodetool('rebuild -ks ks1 -ts (%s,%s] -s %s' %
(tokens[2], str(pow(2, 63) - 1), node2_address))
# verify that node2 streamed to node3
log_matches = node2.grep_log('Session with /%s is complete' %
node3_address)
self.assertTrue(len(log_matches) > 0)
# verify that node1 did not participate
log_matches = node1.grep_log('streaming plan for Rebuild')
self.assertEqual(len(log_matches), 0)
# check data is sent by stopping node1, node2
node1.stop()
node2.stop()
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.ONE)
# ks2 should not be streamed
session.execute('USE ks2')
for i in xrange(0, keys):
query_c1c2(session,
i,
ConsistencyLevel.ONE,
tolerate_missing=True,
must_be_missing=True)
def rebuild_ranges_test(self):
"""
@jira_ticket CASSANDRA-10406
"""
keys = 1000
cluster = self.cluster
tokens = cluster.balanced_tokens_across_dcs(['dc1', 'dc2'])
cluster.set_configuration_options(values={
'endpoint_snitch':
'org.apache.cassandra.locator.PropertyFileSnitch'
})
cluster.set_configuration_options(values={'num_tokens': 1})
node1 = cluster.create_node('node1',
False, ('127.0.0.1', 9160),
('127.0.0.1', 7000),
'7100',
'2000',
tokens[0],
binary_interface=('127.0.0.1', 9042))
node1.set_configuration_options(values={'initial_token': tokens[0]})
cluster.add(node1, True, data_center='dc1')
node1 = cluster.nodelist()[0]
# start node in dc1
node1.start(wait_for_binary_proto=True)
# populate data in dc1
session = self.patient_exclusive_cql_connection(node1)
# ks1 will be rebuilt in node2
create_ks(session, 'ks1', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
# ks2 will not be rebuilt in node2
create_ks(session, 'ks2', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
session.shutdown()
# Bootstraping a new node in dc2 with auto_bootstrap: false
node2 = cluster.create_node('node2',
False, ('127.0.0.2', 9160),
('127.0.0.2', 7000),
'7200',
'2001',
tokens[1],
binary_interface=('127.0.0.2', 9042))
node2.set_configuration_options(values={'initial_token': tokens[1]})
cluster.add(node2, False, data_center='dc2')
node2.start(wait_other_notice=True, wait_for_binary_proto=True)
# wait for snitch to reload
time.sleep(60)
# alter keyspace to replicate to dc2
session = self.patient_exclusive_cql_connection(node2)
session.execute(
"ALTER KEYSPACE ks1 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};"
)
session.execute(
"ALTER KEYSPACE ks2 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};"
)
session.execute('USE ks1')
# rebuild only ks1 with range that is node1's replica
node2.nodetool('rebuild -ks ks1 -ts (%s,%s] dc1' %
(tokens[1], str(pow(2, 63) - 1)))
# check data is sent by stopping node1
node1.stop()
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.ONE)
# ks2 should not be streamed
session.execute('USE ks2')
for i in xrange(0, keys):
query_c1c2(session,
i,
ConsistencyLevel.ONE,
tolerate_missing=True,
must_be_missing=True)
def simple_rebuild_test(self):
"""
@jira_ticket CASSANDRA-9119
Test rebuild from other dc works as expected.
"""
keys = 1000
cluster = self.cluster
cluster.set_configuration_options(values={
'endpoint_snitch':
'org.apache.cassandra.locator.PropertyFileSnitch'
})
node1 = cluster.create_node('node1',
False,
None, ('127.0.0.1', 7000),
'7100',
'2000',
None,
binary_interface=('127.0.0.1', 9042))
cluster.add(node1, True, data_center='dc1')
# start node in dc1
node1.start(wait_for_binary_proto=True)
# populate data in dc1
session = self.patient_exclusive_cql_connection(node1)
create_ks(session, 'ks', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.LOCAL_ONE)
# check data
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
session.shutdown()
# Bootstrapping a new node in dc2 with auto_bootstrap: false
node2 = cluster.create_node('node2',
False, ('127.0.0.2', 9160),
('127.0.0.2', 7000),
'7200',
'2001',
None,
binary_interface=('127.0.0.2', 9042))
cluster.add(node2, False, data_center='dc2')
node2.start(wait_other_notice=True, wait_for_binary_proto=True)
# wait for snitch to reload
time.sleep(60)
# alter keyspace to replicate to dc2
session = self.patient_exclusive_cql_connection(node2)
session.execute(
"ALTER KEYSPACE ks WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};"
)
if self.cluster.version() >= '2.2':
# alter system_auth -- rebuilding it no longer possible after
# CASSANDRA-11848 prevented local node from being considered a source
# Only do this on 2.2+, because on 2.1, this keyspace only
# exists if auth is enabled, which it isn't in this test
session.execute(
"ALTER KEYSPACE system_auth WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};"
)
session.execute('USE ks')
self.rebuild_errors = 0
# rebuild dc2 from dc1
def rebuild():
try:
node2.nodetool('rebuild dc1')
except ToolError as e:
if 'Node is still rebuilding' in e.stdout:
self.rebuild_errors += 1
else:
raise e
class Runner(Thread):
def __init__(self, func):
Thread.__init__(self)
self.func = func
self.thread_exc_info = None
def run(self):
"""
Closes over self to catch any exceptions raised by func and
register them at self.thread_exc_info
Based on http://stackoverflow.com/a/1854263
"""
try:
self.func()
except Exception:
import sys
self.thread_exc_info = sys.exc_info()
cmd1 = Runner(rebuild)
cmd1.start()
# concurrent rebuild should not be allowed (CASSANDRA-9119)
# (following sleep is needed to avoid conflict in 'nodetool()' method setting up env.)
time.sleep(.1)
# we don't need to manually raise exeptions here -- already handled
rebuild()
cmd1.join()
# manually raise exception from cmd1 thread
# see http://stackoverflow.com/a/1854263
if cmd1.thread_exc_info is not None:
raise cmd1.thread_exc_info[1], None, cmd1.thread_exc_info[2]
# exactly 1 of the two nodetool calls should fail
# usually it will be the one in the main thread,
# but occasionally it wins the race with the one in the secondary thread,
# so we check that one succeeded and the other failed
self.assertEqual(
self.rebuild_errors,
1,
msg=
'rebuild errors should be 1, but found {}. Concurrent rebuild should not be allowed, but one rebuild command should have succeeded.'
.format(self.rebuild_errors))
# check data
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
def resumable_rebuild_test(self):
"""
@jira_ticket CASSANDRA-10810
Test rebuild operation is resumable
"""
self.ignore_log_patterns = list(self.ignore_log_patterns) + [
r'Error while rebuilding node',
r'Streaming error occurred on session with peer 127.0.0.3',
r'Remote peer 127.0.0.3 failed stream session'
]
cluster = self.cluster
cluster.set_configuration_options(values={
'endpoint_snitch':
'org.apache.cassandra.locator.PropertyFileSnitch'
})
# Create 2 nodes on dc1
node1 = cluster.create_node('node1',
False, ('127.0.0.1', 9160),
('127.0.0.1', 7000),
'7100',
'2000',
None,
binary_interface=('127.0.0.1', 9042))
node2 = cluster.create_node('node2',
False, ('127.0.0.2', 9160),
('127.0.0.2', 7000),
'7200',
'2001',
None,
binary_interface=('127.0.0.2', 9042))
cluster.add(node1, True, data_center='dc1')
cluster.add(node2, True, data_center='dc1')
node1.start(wait_for_binary_proto=True)
node2.start(wait_for_binary_proto=True)
# Insert data into node1 and node2
session = self.patient_exclusive_cql_connection(node1)
create_ks(session, 'ks', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ALL)
key = list(range(10000, 20000))
session = self.patient_exclusive_cql_connection(node2)
session.execute('USE ks')
insert_c1c2(session, keys=key, consistency=ConsistencyLevel.ALL)
session.shutdown()
# Create a new node3 on dc2
node3 = cluster.create_node('node3',
False, ('127.0.0.3', 9160),
('127.0.0.3', 7000),
'7300',
'2002',
None,
binary_interface=('127.0.0.3', 9042),
byteman_port='8300')
cluster.add(node3, False, data_center='dc2')
node3.start(wait_other_notice=False, wait_for_binary_proto=True)
# Wait for snitch to be refreshed
time.sleep(5)
# Alter necessary keyspace for rebuild operation
session = self.patient_exclusive_cql_connection(node3)
session.execute(
"ALTER KEYSPACE ks WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};"
)
session.execute(
"ALTER KEYSPACE system_auth WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};"
)
# Path to byteman script which makes the streaming to node2 throw an exception, making rebuild fail
if cluster.version() < '4.0':
script = ['./byteman/pre4.0/inject_failure_streaming_to_node2.btm']
else:
script = ['./byteman/4.0/inject_failure_streaming_to_node2.btm']
node3.byteman_submit(script)
# First rebuild must fail and data must be incomplete
with self.assertRaises(ToolError, msg='Unexpected: SUCCEED'):
debug('Executing first rebuild -> '),
node3.nodetool('rebuild dc1')
debug('Expected: FAILED')
session.execute('USE ks')
with self.assertRaises(AssertionError, msg='Unexpected: COMPLETE'):
debug('Checking data is complete -> '),
for i in xrange(0, 20000):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
debug('Expected: INCOMPLETE')
debug('Executing second rebuild -> '),
node3.nodetool('rebuild dc1')
debug('Expected: SUCCEED')
# Check all streaming sessions completed, streamed ranges are skipped and verify streamed data
node3.watch_log_for('All sessions completed')
node3.watch_log_for('Skipping streaming those ranges.')
debug('Checking data is complete -> '),
for i in xrange(0, 20000):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
debug('Expected: COMPLETE')
def rebuild_with_specific_sources_test(self):
"""
@jira_ticket CASSANDRA-9875
Verifies that an operator can specify specific sources to use
when rebuilding.
1. Set up a 2 node cluster across dc1 and dc2
2. Create new keyspaces with replication factor 2 (one replica in each datacenter)
4. Populate nodes with data
5. Create a new node in dc3 and update the keyspace replication
6. Run rebuild on the new node with a specific source in dc2
7. Assert that streaming only occurred between the new node and the specified source
8. Assert that the rebuild was successful by checking the data
"""
keys = 1000
cluster = self.cluster
tokens = cluster.balanced_tokens_across_dcs(['dc1', 'dc2', 'dc3'])
cluster.set_configuration_options(values={'endpoint_snitch': 'org.apache.cassandra.locator.PropertyFileSnitch'})
cluster.set_configuration_options(values={'num_tokens': 1})
cluster.populate([1, 1], tokens=tokens[:2])
node1, node2 = cluster.nodelist()
cluster.start(wait_for_binary_proto=True)
# populate data in dc1, dc2
session = self.patient_exclusive_cql_connection(node1)
# ks1 will be rebuilt in node3
create_ks(session, 'ks1', {'dc1': 1, 'dc2': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
# ks2 will not be rebuilt in node3
create_ks(session, 'ks2', {'dc1': 1, 'dc2': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
session.shutdown()
# bootstrap a new node in dc3 with auto_bootstrap: false
node3 = cluster.create_node('node3', False,
('127.0.0.3', 9160),
('127.0.0.3', 7000),
'7300', '2002', tokens[2],
binary_interface=('127.0.0.3', 9042))
cluster.add(node3, False, data_center='dc3')
node3.start(wait_other_notice=True, wait_for_binary_proto=True)
# wait for snitch to reload
time.sleep(60)
# alter keyspace to replicate to dc3
session = self.patient_exclusive_cql_connection(node3)
session.execute("ALTER KEYSPACE ks1 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1, 'dc3':1};")
session.execute("ALTER KEYSPACE ks2 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1, 'dc3':1};")
session.execute('USE ks1')
node2_address = node2.network_interfaces['binary'][0]
node3_address = node3.network_interfaces['binary'][0]
# rebuild only ks1, restricting the source to node2
node3.nodetool('rebuild -ks ks1 -ts (%s,%s] -s %s' % (tokens[2], str(pow(2, 63) - 1), node2_address))
# verify that node2 streamed to node3
log_matches = node2.grep_log('Session with /%s is complete' % node3_address)
self.assertTrue(len(log_matches) > 0)
# verify that node1 did not participate
log_matches = node1.grep_log('streaming plan for Rebuild')
self.assertEqual(len(log_matches), 0)
# check data is sent by stopping node1, node2
node1.stop()
node2.stop()
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.ONE)
# ks2 should not be streamed
session.execute('USE ks2')
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.ONE, tolerate_missing=True, must_be_missing=True)
def multiple_repair_test(self):
"""
* Launch a three node cluster
* Create a keyspace with RF 3 and a table
* Insert 49 rows
* Stop node3
* Insert 50 more rows
* Restart node3
* Issue an incremental repair on node3
* Stop node2
* Insert a final50 rows
* Restart node2
* Issue an incremental repair on node2
* Replace node3 with a new node
* Verify data integrity
# TODO: Several more verifications of data need to be interspersed throughout the test. The final assertion is insufficient.
@jira_ticket CASSANDRA-10644
"""
cluster = self.cluster
cluster.populate(3).start()
node1, node2, node3 = cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 3)
create_cf(session, 'cf', read_repair=0.0, columns={'c1': 'text', 'c2': 'text'})
debug("insert data")
insert_c1c2(session, keys=range(1, 50), consistency=ConsistencyLevel.ALL)
node1.flush()
debug("bringing down node 3")
node3.flush()
node3.stop(gently=False)
debug("inserting additional data into node 1 and 2")
insert_c1c2(session, keys=range(50, 100), consistency=ConsistencyLevel.TWO)
node1.flush()
node2.flush()
debug("restarting and repairing node 3")
node3.start(wait_for_binary_proto=True)
if cluster.version() >= "2.2":
node3.repair()
else:
node3.nodetool("repair -par -inc")
# wait stream handlers to be closed on windows
# after session is finished (See CASSANDRA-10644)
if is_win:
time.sleep(2)
debug("stopping node 2")
node2.stop(gently=False)
debug("inserting data in nodes 1 and 3")
insert_c1c2(session, keys=range(100, 150), consistency=ConsistencyLevel.TWO)
node1.flush()
node3.flush()
debug("start and repair node 2")
node2.start(wait_for_binary_proto=True)
if cluster.version() >= "2.2":
node2.repair()
else:
node2.nodetool("repair -par -inc")
debug("replace node and check data integrity")
node3.stop(gently=False)
node5 = Node('node5', cluster, True, ('127.0.0.5', 9160), ('127.0.0.5', 7000), '7500', '0', None, ('127.0.0.5', 9042))
cluster.add(node5, False)
node5.start(replace_address='127.0.0.3', wait_other_notice=True)
assert_one(session, "SELECT COUNT(*) FROM ks.cf LIMIT 200", [149])
def simple_rebuild_test(self):
"""
@jira_ticket CASSANDRA-9119
Test rebuild from other dc works as expected.
"""
keys = 1000
cluster = self.cluster
cluster.set_configuration_options(values={'endpoint_snitch': 'org.apache.cassandra.locator.PropertyFileSnitch'})
node1 = cluster.create_node('node1', False,
('127.0.0.1', 9160),
('127.0.0.1', 7000),
'7100', '2000', None,
binary_interface=('127.0.0.1', 9042))
cluster.add(node1, True, data_center='dc1')
# start node in dc1
node1.start(wait_for_binary_proto=True)
# populate data in dc1
session = self.patient_exclusive_cql_connection(node1)
create_ks(session, 'ks', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.LOCAL_ONE)
# check data
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
session.shutdown()
# Bootstrapping a new node in dc2 with auto_bootstrap: false
node2 = cluster.create_node('node2', False,
('127.0.0.2', 9160),
('127.0.0.2', 7000),
'7200', '2001', None,
binary_interface=('127.0.0.2', 9042))
cluster.add(node2, False, data_center='dc2')
node2.start(wait_other_notice=True, wait_for_binary_proto=True)
# wait for snitch to reload
time.sleep(60)
# alter keyspace to replicate to dc2
session = self.patient_exclusive_cql_connection(node2)
session.execute("ALTER KEYSPACE ks WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};")
if self.cluster.version() >= '2.2':
# alter system_auth -- rebuilding it no longer possible after
# CASSANDRA-11848 prevented local node from being considered a source
# Only do this on 2.2+, because on 2.1, this keyspace only
# exists if auth is enabled, which it isn't in this test
session.execute("ALTER KEYSPACE system_auth WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};")
session.execute('USE ks')
self.rebuild_errors = 0
# rebuild dc2 from dc1
def rebuild():
try:
node2.nodetool('rebuild dc1')
except ToolError as e:
if 'Node is still rebuilding' in e.stdout:
self.rebuild_errors += 1
else:
raise e
class Runner(Thread):
def __init__(self, func):
Thread.__init__(self)
self.func = func
self.thread_exc_info = None
def run(self):
"""
Closes over self to catch any exceptions raised by func and
register them at self.thread_exc_info
Based on http://stackoverflow.com/a/1854263
"""
try:
self.func()
except Exception:
import sys
self.thread_exc_info = sys.exc_info()
cmd1 = Runner(rebuild)
cmd1.start()
# concurrent rebuild should not be allowed (CASSANDRA-9119)
# (following sleep is needed to avoid conflict in 'nodetool()' method setting up env.)
time.sleep(.1)
# we don't need to manually raise exeptions here -- already handled
rebuild()
cmd1.join()
# manually raise exception from cmd1 thread
# see http://stackoverflow.com/a/1854263
if cmd1.thread_exc_info is not None:
raise cmd1.thread_exc_info[1], None, cmd1.thread_exc_info[2]
# exactly 1 of the two nodetool calls should fail
# usually it will be the one in the main thread,
# but occasionally it wins the race with the one in the secondary thread,
# so we check that one succeeded and the other failed
self.assertEqual(self.rebuild_errors, 1,
msg='rebuild errors should be 1, but found {}. Concurrent rebuild should not be allowed, but one rebuild command should have succeeded.'.format(self.rebuild_errors))
# check data
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
def test_hintedhandoff_window(self):
"""
Test that we only store at a maximum the hint window worth of hints.
Prior to CASSANDRA-14309 we would store another window worth of hints
if the down node was brought up and then taken back down immediately.
We would also store another window of hints on a live node if the live
node was restarted.
@jira_ticket CASSANDRA-14309
"""
# hint_window_persistent_enabled is set to true by default
self.cluster.set_configuration_options({
'max_hint_window_in_ms': 5000,
'hinted_handoff_enabled': True,
'max_hints_delivery_threads': 1,
'hints_flush_period_in_ms': 100,
})
self.cluster.populate(2).start()
node1, node2 = self.cluster.nodelist()
session = self.patient_cql_connection(node1)
create_ks(session, 'ks', 2)
create_c1c2_table(self, session)
# Stop handoff until very end and take node2 down for first round of hints
node1.nodetool('pausehandoff')
node2.nodetool('disablebinary')
node2.nodetool('disablegossip')
# First round of hints. We expect these to be replayed and the only
# hints within the window
insert_c1c2(session, n=(0, 100), consistency=ConsistencyLevel.ONE)
# Let hint window pass
time.sleep(10)
# Re-enable and disable the node. Prior to CASSANDRA-14215 this should make the hint window on node1 reset.
node2.nodetool('enablegossip')
node2.nodetool('disablegossip')
# Second round of inserts. We do not expect hints to be stored.
insert_c1c2(session, n=(100, 200), consistency=ConsistencyLevel.ONE)
# Restart node1. Prior to CASSANDRA-14215 this would reset node1's hint window.
node1.stop()
node1.start(wait_for_binary_proto=True, wait_other_notice=False)
session = self.patient_exclusive_cql_connection(node1)
session.execute('USE ks')
# Third round of inserts. We do not expect hints to be stored.
insert_c1c2(session, n=(200, 300), consistency=ConsistencyLevel.ONE)
# Enable node2 and wait for hints to be replayed
node2.nodetool('enablegossip')
node2.nodetool('enablebinary')
node1.nodetool('resumehandoff')
node1.watch_log_for('Finished hinted handoff')
# Stop node1 so that we only query node2
node1.stop()
session = self.patient_exclusive_cql_connection(node2)
session.execute('USE ks')
# Ensure first dataset is present (through replayed hints)
for x in range(0, 100):
query_c1c2(session, x, ConsistencyLevel.ONE)
# Ensure second and third datasets are not present
for x in range(100, 300):
query_c1c2(session,
x,
ConsistencyLevel.ONE,
tolerate_missing=True,
must_be_missing=True)
def rebuild_ranges_test(self):
"""
@jira_ticket CASSANDRA-10406
"""
keys = 1000
cluster = self.cluster
tokens = cluster.balanced_tokens_across_dcs(['dc1', 'dc2'])
cluster.set_configuration_options(values={'endpoint_snitch': 'org.apache.cassandra.locator.PropertyFileSnitch'})
cluster.set_configuration_options(values={'num_tokens': 1})
node1 = cluster.create_node('node1', False,
('127.0.0.1', 9160),
('127.0.0.1', 7000),
'7100', '2000', tokens[0],
binary_interface=('127.0.0.1', 9042))
node1.set_configuration_options(values={'initial_token': tokens[0]})
cluster.add(node1, True, data_center='dc1')
node1 = cluster.nodelist()[0]
# start node in dc1
node1.start(wait_for_binary_proto=True)
# populate data in dc1
session = self.patient_exclusive_cql_connection(node1)
# ks1 will be rebuilt in node2
create_ks(session, 'ks1', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
# ks2 will not be rebuilt in node2
create_ks(session, 'ks2', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=keys, consistency=ConsistencyLevel.ALL)
session.shutdown()
# Bootstraping a new node in dc2 with auto_bootstrap: false
node2 = cluster.create_node('node2', False,
('127.0.0.2', 9160),
('127.0.0.2', 7000),
'7200', '2001', tokens[1],
binary_interface=('127.0.0.2', 9042))
node2.set_configuration_options(values={'initial_token': tokens[1]})
cluster.add(node2, False, data_center='dc2')
node2.start(wait_other_notice=True, wait_for_binary_proto=True)
# wait for snitch to reload
time.sleep(60)
# alter keyspace to replicate to dc2
session = self.patient_exclusive_cql_connection(node2)
session.execute("ALTER KEYSPACE ks1 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};")
session.execute("ALTER KEYSPACE ks2 WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};")
session.execute('USE ks1')
# rebuild only ks1 with range that is node1's replica
node2.nodetool('rebuild -ks ks1 -ts (%s,%s] dc1' % (tokens[1], str(pow(2, 63) - 1)))
# check data is sent by stopping node1
node1.stop()
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.ONE)
# ks2 should not be streamed
session.execute('USE ks2')
for i in xrange(0, keys):
query_c1c2(session, i, ConsistencyLevel.ONE, tolerate_missing=True, must_be_missing=True)
def resumable_rebuild_test(self):
"""
@jira_ticket CASSANDRA-10810
Test rebuild operation is resumable
"""
self.ignore_log_patterns = list(self.ignore_log_patterns) + [r'Error while rebuilding node',
r'Streaming error occurred on session with peer 127.0.0.3',
r'Remote peer 127.0.0.3 failed stream session']
cluster = self.cluster
cluster.set_configuration_options(values={'endpoint_snitch': 'org.apache.cassandra.locator.PropertyFileSnitch'})
# Create 2 nodes on dc1
node1 = cluster.create_node('node1', False,
('127.0.0.1', 9160),
('127.0.0.1', 7000),
'7100', '2000', None,
binary_interface=('127.0.0.1', 9042))
node2 = cluster.create_node('node2', False,
('127.0.0.2', 9160),
('127.0.0.2', 7000),
'7200', '2001', None,
binary_interface=('127.0.0.2', 9042))
cluster.add(node1, True, data_center='dc1')
cluster.add(node2, True, data_center='dc1')
node1.start(wait_for_binary_proto=True)
node2.start(wait_for_binary_proto=True)
# Insert data into node1 and node2
session = self.patient_exclusive_cql_connection(node1)
create_ks(session, 'ks', {'dc1': 1})
create_cf(session, 'cf', columns={'c1': 'text', 'c2': 'text'})
insert_c1c2(session, n=10000, consistency=ConsistencyLevel.ALL)
key = list(range(10000, 20000))
session = self.patient_exclusive_cql_connection(node2)
session.execute('USE ks')
insert_c1c2(session, keys=key, consistency=ConsistencyLevel.ALL)
session.shutdown()
# Create a new node3 on dc2
node3 = cluster.create_node('node3', False,
('127.0.0.3', 9160),
('127.0.0.3', 7000),
'7300', '2002', None,
binary_interface=('127.0.0.3', 9042),
byteman_port='8300')
cluster.add(node3, False, data_center='dc2')
node3.start(wait_other_notice=False, wait_for_binary_proto=True)
# Wait for snitch to be refreshed
time.sleep(5)
# Alter necessary keyspace for rebuild operation
session = self.patient_exclusive_cql_connection(node3)
session.execute("ALTER KEYSPACE ks WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};")
session.execute("ALTER KEYSPACE system_auth WITH REPLICATION = {'class':'NetworkTopologyStrategy', 'dc1':1, 'dc2':1};")
# Path to byteman script which makes the streaming to node2 throw an exception, making rebuild fail
script = ['./byteman/inject_failure_streaming_to_node2.btm']
node3.byteman_submit(script)
# First rebuild must fail and data must be incomplete
with self.assertRaises(ToolError, msg='Unexpected: SUCCEED'):
debug('Executing first rebuild -> '),
node3.nodetool('rebuild dc1')
debug('Expected: FAILED')
session.execute('USE ks')
with self.assertRaises(AssertionError, msg='Unexpected: COMPLETE'):
debug('Checking data is complete -> '),
for i in xrange(0, 20000):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
debug('Expected: INCOMPLETE')
debug('Executing second rebuild -> '),
node3.nodetool('rebuild dc1')
debug('Expected: SUCCEED')
# Check all streaming sessions completed, streamed ranges are skipped and verify streamed data
node3.watch_log_for('All sessions completed')
node3.watch_log_for('Skipping streaming those ranges.')
debug('Checking data is complete -> '),
for i in xrange(0, 20000):
query_c1c2(session, i, ConsistencyLevel.LOCAL_ONE)
debug('Expected: COMPLETE')
