def test_config_from_second_generation_snapshot(cluster):
"""
A regression test for #44: confirm that if we load a snapshot
on startup, do nothing, then re-create a snapshot we don't end
up with a messed up nodes config.
"""
cluster.create(3)
# Bump the log a bit
for _ in range(20):
assert cluster.raft_exec('INCR', 'testkey')
# Compact to get rid of logs
node3 = cluster.node(3)
assert node3.client.execute_command('RAFT.DEBUG', 'COMPACT') == b'OK'
# Restart node
node3.restart()
node3.wait_for_node_voting()
# Bump the log a bit
for _ in range(20):
assert cluster.raft_exec('INCR', 'testkey')
# Recompact
cluster.wait_for_unanimity()
assert node3.client.execute_command('RAFT.DEBUG', 'COMPACT') == b'OK'
node3.restart()
node3.wait_for_node_voting()
assert node3.raft_info()['num_nodes'] == 3
def test_removed_node_remains_dead(cluster):
"""
A removed node stays down and does not resurrect in any case.
"""
cluster.create(3)
# Some baseline data
for _ in range(100):
cluster.raft_exec('INCR', 'counter')
# Remove node 3
cluster.node(1).client.execute_command('RAFT.NODE', 'REMOVE', '3')
cluster.node(1).wait_for_num_voting_nodes(2)
# Add more data
for _ in range(100):
cluster.raft_exec('INCR', 'counter')
# Check
node = cluster.node(3)
# Verify node 3 does not accept writes
with raises(RedisError):
node.client.execute_command('RAFT', 'INCR', 'counter')
# Verify node 3 still does not accept writes after a restart
node.terminate()
node.start()
with raises(RedisError):
node.client.execute_command('RAFT', 'INCR', 'counter')
def test_fuzzing_with_restarts_and_rewrites(cluster):
"""
Counter fuzzer with log rewrites.
"""
nodes = 3
cycles = 100
cluster.create(nodes)
# Randomize max log entries
for node in cluster.nodes.values():
node.client.execute_command('RAFT.CONFIG', 'SET',
'raft-log-max-file-size',
str(random.randint(1000, 2000)))
for i in range(cycles):
assert cluster.raft_exec('INCRBY', 'counter', 1) == i + 1
logging.info('---------- Executed INCRBY # %s', i)
if random.randint(1, 7) == 1:
r = random.randint(1, nodes)
logging.info('********** Restarting node %s **********', r)
cluster.node(r).restart()
cluster.node(r).wait_for_election()
logging.info('********** Node %s is UP **********', r)
assert int(cluster.raft_exec('GET', 'counter')) == cycles
def test_reelection_basic_flow(cluster):
"""
Basic reelection flow
"""
cluster.create(3)
assert cluster.leader == 1
assert cluster.raft_exec('SET', 'key', 'value') == b'OK'
cluster.node(1).terminate()
cluster.node(2).wait_for_election()
assert cluster.raft_exec('SET', 'key2', 'value2') == b'OK'
cluster.exec_all('GET', 'key2')
def test_proxying(cluster):
"""
Command proxying from follower to leader works
"""
cluster.create(3)
assert cluster.leader == 1
with raises(ResponseError, match='MOVED'):
assert cluster.node(2).raft_exec('SET', 'key', 'value') == b'OK'
assert cluster.node(2).client.execute_command('RAFT.CONFIG', 'SET',
'follower-proxy',
'yes') == b'OK'
# Basic sanity
assert cluster.node(2).raft_exec('SET', 'key', 'value') == b'OK'
assert cluster.raft_exec('GET', 'key') == b'value'
# Numeric values
assert cluster.node(2).raft_exec('SADD', 'myset', 'a') == 1
assert cluster.node(2).raft_exec('SADD', 'myset', 'b') == 1
# Multibulk
assert set(cluster.node(2).raft_exec('SMEMBERS',
'myset')) == set([b'a', b'b'])
# Nested multibulk
assert set(
cluster.node(2).raft_exec('EVAL', 'return {{\'a\',\'b\',\'c\'}};',
0)[0]) == set([b'a', b'b', b'c'])
# Error
with raises(ResponseError, match='WRONGTYPE'):
cluster.node(2).raft_exec('INCR', 'myset')
def test_stale_reads_on_leader_election(cluster):
"""
"""
cluster.create(3)
# Try 10 times
for _ in range(10):
val_written = cluster.raft_exec("INCR", "counter-1")
leader = cluster.node(cluster.leader)
leader.terminate()
leader.start(verify=False)
val_read = cluster.raft_exec('GET', 'counter-1')
assert val_read is not None
assert val_written == int(val_read)
time.sleep(1)
def test_log_rollback(cluster):
"""
Rollback of log entries that were written in the minority.
"""
cluster.create(3)
assert cluster.leader == 1
assert cluster.raft_exec('INCRBY', 'key', '111') == 111
# Break cluster
cluster.node(2).terminate()
cluster.node(3).terminate()
# Load a command which can't be committed
assert cluster.node(1).current_index() == 6
conn = cluster.node(1).client.connection_pool.get_connection('RAFT')
conn.send_command('RAFT', 'INCRBY', 'key', '222')
assert cluster.node(1).current_index() == 7
cluster.node(1).terminate()
# We want to be sure the last entry is in the log
log = RaftLog(cluster.node(1).raftlog)
log.read()
assert log.entry_count() == 7
# Restart the cluster without node 1, make sure the write was
# not committed.
cluster.node(2).start()
cluster.node(3).start()
cluster.node(2).wait_for_election()
assert cluster.node(2).current_index() == 7 # 6 + 1 no-op entry
# Restart node 1
cluster.node(1).start()
cluster.node(1).wait_for_election()
# Make another write and make sure it overwrites the previous one in
# node 1's log
assert cluster.raft_exec('INCRBY', 'key', '333') == 444
cluster.wait_for_unanimity()
# Make sure log reflects the change
log.reset()
log.read()
assert match(r'.*INCRBY.*333', str(log.entries[-1].data()))
def test_node_history_with_same_address(cluster):
""
""
cluster.create(5)
cluster.raft_exec("INCR", "step-counter")
# Remove nodes
ports = []
for node_id in [2, 3, 4, 5]:
ports.append(cluster.node(node_id).port)
cluster.remove_node(node_id)
cluster.leader_node().wait_for_log_applied()
cluster.node(cluster.leader).wait_for_num_nodes(1)
# Now add and remove several more times
for _ in range(5):
for port in ports:
n = cluster.add_node(port=port)
cluster.leader_node().wait_for_num_nodes(2)
cluster.leader_node().wait_for_log_applied()
cluster.remove_node(n.id)
cluster.leader_node().wait_for_num_nodes(1)
cluster.leader_node().wait_for_log_applied()
# Add enough data in the log to satisfy timing
for _ in range(3000):
cluster.raft_exec("INCR", "step-counter")
# Add another node
new_node = cluster.add_node(port=ports[0])
new_node.wait_for_node_voting()
# Terminate all
cluster.terminate()
# Start new node
cluster.start()
# need some time to start applying logs..
time.sleep(2)
assert cluster.raft_exec("GET", "step-counter") == b'3001'
def test_fuzzing_with_config_changes(cluster):
"""
Basic Raft fuzzer test
"""
nodes = 5
cycles = 100
cluster.create(nodes)
for i in range(cycles):
assert cluster.raft_exec('INCRBY', 'counter', 1) == i + 1
if random.randint(1, 7) == 1:
try:
node_id = cluster.random_node_id()
cluster.remove_node(node_id)
except ResponseError:
continue
cluster.add_node().wait_for_node_voting()
assert int(cluster.raft_exec('GET', 'counter')) == cycles
def test_fuzzing_with_restarts(cluster):
"""
Basic Raft fuzzer test
"""
nodes = 3
cycles = 100
cluster.create(nodes)
for i in range(cycles):
assert cluster.raft_exec('INCRBY', 'counter', 1) == i + 1
logging.info('---------- Executed INCRBY # %s', i)
if i % 7 == 0:
r = random.randint(1, nodes)
logging.info('********** Restarting node %s **********', r)
cluster.node(r).restart()
cluster.node(r).wait_for_election()
logging.info('********** Node %s is UP **********', r)
assert int(cluster.raft_exec('GET', 'counter')) == cycles
def test_stale_log_trim(cluster):
"""
When starting up, if log is older than snapshot it should be trimmed.
"""
cluster.create(3, prepopulate_log=20)
# Stop node 3 and advance the log; then overwrite node3
# with a recent snapshot and start it. This simulates delivery of snapshot
# and a crash sometime before the log is adjusted.
cluster.node(3).terminate()
for _ in range(20):
assert cluster.raft_exec('INCR', 'testkey')
assert cluster.node(1).client.execute_command('RAFT.DEBUG', 'COMPACT') == b'OK'
for _ in range(20):
assert cluster.raft_exec('INCR', 'testkey')
shutil.copyfile(os.path.join(os.curdir, cluster.node(1).dbfilename),
os.path.join(os.curdir, cluster.node(3).dbfilename))
cluster.node(3).start()
cluster.node(3).wait_for_node_voting()
assert cluster.raft_exec('INCR', 'last-key')
def test_log_reset_on_snapshot_load(cluster):
"""
Test correct reset of log when a snapshot is received.
"""
cluster.create(3, prepopulate_log=20)
# Stop node 3, advance the log, then compact.
cluster.node(3).terminate()
for _ in range(20):
assert cluster.raft_exec('INCR', 'testkey')
assert cluster.node(1).client.execute_command('RAFT.DEBUG', 'COMPACT') == b'OK'
# Start node 3 and wait for it to receive a snapshot
cluster.node(3).start()
cluster.node(3).wait_for_node_voting()
# Restart node 3 and make sure it correctly started
cluster.node(3).terminate()
cluster.node(3).start()
cluster.node(3).wait_for_node_voting()
assert cluster.raft_exec('INCR', 'last-key')
cluster.wait_for_unanimity()
def test_new_uncommitted_during_rewrite(cluster):
cluster.create(3)
# Take down majority to create uncommitted entries and check rewrite
cluster.node(1).raft_exec('SET', 'key', '1')
# Initiate compaction and wait to see it's in progress
conn = cluster.node(1).client.connection_pool.get_connection('COMPACT')
conn.send_command('RAFT.DEBUG', 'COMPACT', '2')
cluster.node(1).wait_for_info_param('snapshot_in_progress', 'yes')
assert cluster.node(1).raft_info()['snapshot_in_progress'] == 'yes'
# Send a bunch of writes
cluster.node(1).raft_exec('INCRBY', 'key', '2')
cluster.node(1).raft_exec('INCRBY', 'key', '3')
cluster.node(1).raft_exec('INCRBY', 'key', '4')
# Wait for compaction to end
assert cluster.node(1).raft_info()['snapshot_in_progress'] == 'yes'
cluster.node(1).wait_for_info_param('snapshot_in_progress', 'no')
# Make sure our writes made it to the log
log = RaftLog(cluster.node(1).raftlog)
log.read()
assert log.entry_count(LogEntry.LogType.NORMAL) == 3
# Extra check -- Make sure we can read it back. Note that we need to start
# all nodes because we don't log the commit index.
cluster.node(1).terminate()
cluster.node(2).terminate()
cluster.node(3).terminate()
cluster.node(1).start()
cluster.node(2).start()
cluster.node(3).start()
cluster.node(1).wait_for_info_param('state', 'up')
# Make sure cluster state is as expected
assert cluster.raft_exec('get', 'key') == b'10'
# Make sure node 1 state is as expected
cluster.node(1).wait_for_log_applied()
assert cluster.node(1).client.get('key') == b'10'
def test_read_before_commits(cluster):
"""
"""
cluster.create(3)
assert cluster.raft_exec('GET', 'somekey') is None