def start():
"""Start casssandra nodes"""
product = dse if config['product'] == 'dse' else cstar
cassandra_path = product.get_cassandra_path()
# Place environment file on host:
env = config.get('env', '')
fab.puts('env is: {}'.format(env))
if isinstance(env, list) or isinstance(env, tuple):
env = "\n".join(env)
env += "\n"
fab.puts('env is: {}'.format(env))
if not config['use_jna']:
env = 'JVM_EXTRA_OPTS=-Dcassandra.boot_without_jna=true\n\n' + env
# Turn on GC logging:
fab.run("mkdir -p ~/fab/cassandra/logs")
log_dir = fab.run("readlink -m {log_dir}".format(log_dir=config['log_dir']))
try:
ip_address = cluster_config['hosts'][fab.env.host]['internal_ip']
except:
ip_address = fab.env.host
env = "JVM_OPTS=\"$JVM_OPTS -Djava.rmi.server.hostname={hostname} -Xloggc:{log_dir}/gc.log\"\n\n".format(
hostname=ip_address, log_dir=log_dir) + env
# Enable JMX without authentication
env = "JVM_OPTS=\"$JVM_OPTS -Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.authenticate=false\"\n" + env
# Flamegraph
if flamegraph.is_enabled(config):
env += "JVM_OPTS=\"$JVM_OPTS -XX:+PreserveFramePointer\""
if profiler.yourkit_is_enabled(config):
execute(profiler.yourkit_clean)
env += profiler.yourkit_get_jvm_opts()
fab.puts("running with token allocation type: {}".format(config['token_allocation']))
if config['use_vnodes'] and config['token_allocation'] in ('static-random', 'static-algorithmic'):
env += "JVM_OPTS=\"$JVM_OPTS -Dcassandra.initial_token={}\"\n".format(
get_static_vnode_tokens(fab.env.host,
fab.env.hosts,
partitioner=config['partitioner'],
group=config['token_allocation']))
env_script = "{name}.sh".format(name=uuid.uuid1())
env_file = StringIO(env)
fab.run('mkdir -p ~/fab/scripts')
fab.put(env_file, '~/fab/scripts/{env_script}'.format(env_script=env_script))
fab.puts('env is: {}'.format(env))
if len(env_script) > 0:
env_path = os.path.join(cassandra_path, 'conf/cassandra-env.sh')
fab.run('echo >> ~/fab/scripts/{env_script}'.format(**locals()))
fab.run('cat {env_path} >> ~/fab/scripts/{env_script}'.format(**locals()))
fab.run('cp ~/fab/scripts/{env_script} {env_path}'.format(**locals()))
product.start(config)
def bootstrap_cluster(cfg):
config = copy.copy(pristine_config)
config.update(cfg)
# Flamegraph Setup
if flamegraph.is_enabled():
execute(flamegraph.setup)
git_id = bootstrap(config, destroy=True)
return git_id
def bootstrap_cluster(cfg):
config = copy.copy(pristine_config)
config.update(cfg)
# Flamegraph Setup
if flamegraph.is_enabled():
execute(flamegraph.setup)
git_id = bootstrap(config, destroy=True)
return git_id
def stress_compare(revisions,
title,
log,
operations = [],
subtitle = '',
capture_fincore=False,
initial_destroy=True,
leave_data=False,
keep_page_cache=False
):
"""
Run Stress on multiple C* branches and compare them.
revisions - List of dictionaries that contain cluster configurations
to trial. This is combined with the default config.
title - The title of the comparison
subtitle - A subtitle for more information (displayed smaller underneath)
log - The json file path to record stats to
operations - List of dictionaries indicating the operations. Example:
[# cassandra-stress command, node defaults to cluster defined 'stress_node'
{'type': 'stress',
'command': 'write n=19M -rate threads=50',
'node': 'node1',
'wait_for_compaction': True},
# nodetool command to run in parallel on nodes:
{'type': 'nodetool',
'command': 'decomission',
'nodes': ['node1','node2']},
# cqlsh script, node defaults to cluster defined 'stress_node'
{'type': 'cqlsh',
'script': "use my_ks; INSERT INTO blah (col1, col2) VALUES (val1, val2);",
'node': 'node1'}
]
capture_fincore - Enables capturing of linux-fincore logs of C* data files.
initial_destroy - Destroy all data before the first revision is run.
leave_data - Whether to leave the Cassandra data/commitlog/etc directories intact between revisions.
keep_page_cache - Whether to leave the linux page cache intact between revisions.
"""
validate_revisions_list(revisions)
validate_operations_list(operations)
pristine_config = copy.copy(fab_config)
# initial_destroy setting can be set in the job
# configuration, or manually in the call to this function.
# Either is fine, but they shouldn't conflict. If they do,
# ValueError is raised.
if initial_destroy == True and pristine_config.get('initial_destroy', None) == False:
raise ValueError('setting for initial_destroy conflicts in job config and stress_compare() call')
else:
initial_destroy = bool(distutils.util.strtobool(str(pristine_config.get('initial_destroy', initial_destroy))))
if initial_destroy:
logger.info("Cleaning up from prior runs of stress_compare ...")
teardown(destroy=True, leave_data=False)
# Update our local cassandra git remotes and branches
_, localhost_entry = get_localhost()
with common.fab.settings(hosts=[localhost_entry]):
execute(cstar.update_cassandra_git)
# Flamegraph Setup
if flamegraph.is_enabled():
execute(flamegraph.setup)
clean_stress()
stress_revisions = set([operation['stress_revision'] for operation in operations if 'stress_revision' in operation])
stress_shas = setup_stress(stress_revisions)
with GracefulTerminationHandler() as handler:
for rev_num, revision_config in enumerate(revisions):
config = copy.copy(pristine_config)
config.update(revision_config)
revision = revision_config['revision']
config['log'] = log
config['title'] = title
config['subtitle'] = subtitle
product = dse if config.get('product') == 'dse' else cstar
# leave_data setting can be set in the revision
# configuration, or manually in the call to this function.
# Either is fine, but they shouldn't conflict. If they do,
# ValueError is raised.
if leave_data == True and revision_config.get('leave_data', None) == False:
raise ValueError('setting for leave_data conflicts in job config and stress_compare() call')
else:
leave_data = bool(distutils.util.strtobool(str(revision_config.get('leave_data', leave_data))))
logger.info("Bringing up {revision} cluster...".format(revision=revision))
# Drop the page cache between each revision, especially
# important when leave_data=True :
if not keep_page_cache:
drop_page_cache()
# Only fetch from git on the first run:
git_fetch = True if rev_num == 0 else False
revision_config['git_id'] = git_id = bootstrap(config,
destroy=initial_destroy,
leave_data=leave_data,
git_fetch=git_fetch)
if flamegraph.is_enabled(revision_config):
execute(flamegraph.ensure_stopped_perf_agent)
execute(flamegraph.start_perf_agent, rev_num)
if capture_fincore:
start_fincore_capture(interval=10)
last_stress_operation_id = 'None'
for operation_i, operation in enumerate(operations, 1):
try:
start = datetime.datetime.now()
stats = {
"id": str(uuid.uuid1()),
"type": operation['type'],
"revision": revision,
"git_id": git_id,
"start_date": start.isoformat(),
"label": revision_config.get('label', revision_config['revision']),
"test": '{operation_i}_{operation}'.format(
operation_i=operation_i,
operation=operation['type'])
}
if operation['type'] == 'stress':
last_stress_operation_id = stats['id']
# Default to all the nodes of the cluster if no
# nodes were specified in the command:
if operation.has_key('nodes'):
cmd = "{command} -node {hosts}".format(
command=operation['command'],
hosts=",".join(operation['nodes']))
elif '-node' in operation['command']:
cmd = operation['command']
else:
cmd = "{command} -node {hosts}".format(
command=operation['command'],
hosts=",".join([n for n in fab_config['hosts']]))
stats['command'] = cmd
stats['intervals'] = []
stats['test'] = '{operation_i}_{operation}'.format(
operation_i=operation_i, operation=cmd.strip().split(' ')[0]).replace(" ", "_")
logger.info('Running stress operation : {cmd} ...'.format(cmd=cmd))
# Run stress:
# (stress takes the stats as a parameter, and adds
# more as it runs):
stress_sha = stress_shas[operation.get('stress_revision', 'default')]
stats = stress(cmd, revision, stress_sha, stats=stats)
# Wait for all compactions to finish (unless disabled):
if operation.get('wait_for_compaction', True):
compaction_throughput = revision_config.get("compaction_throughput_mb_per_sec", 16)
wait_for_compaction(compaction_throughput=compaction_throughput)
elif operation['type'] == 'nodetool':
if 'nodes' not in operation:
operation['nodes'] = 'all'
if operation['nodes'] in ['all','ALL']:
nodes = [n for n in fab_config['hosts']]
else:
nodes = operation['nodes']
set_nodetool_path(os.path.join(product.get_bin_path(), 'nodetool'))
logger.info("Running nodetool on {nodes} with command: {command}".format(nodes=operation['nodes'], command=operation['command']))
stats['command'] = operation['command']
output = nodetool_multi(nodes, operation['command'])
stats['output'] = output
logger.info("Nodetool command finished on all nodes")
elif operation['type'] == 'cqlsh':
logger.info("Running cqlsh commands on {node}".format(node=operation['node']))
set_cqlsh_path(os.path.join(product.get_bin_path(), 'cqlsh'))
output = cqlsh(operation['script'], operation['node'])
stats['output'] = output.split("\n")
stats['command'] = operation['script']
logger.info("Cqlsh commands finished")
elif operation['type'] == 'bash':
nodes = operation.get('nodes', [n for n in fab_config['hosts']])
logger.info("Running bash commands on: {nodes}".format(nodes=nodes))
stats['output'] = bash(operation['script'], nodes)
stats['command'] = operation['script']
logger.info("Bash commands finished")
elif operation['type'] == 'spark_cassandra_stress':
node = operation['node']
logger.info("Running spark_cassandra_stress on {node}".format(node=node))
output = spark_cassandra_stress(operation['script'], node)
stats['output'] = output
logger.info("spark_cassandra_stress finished")
elif operation['type'] == 'ctool':
logger.info("Running ctool with parameters: {command}".format(command=operation['command']))
ctool = Ctool(operation['command'], common.config)
output = execute(ctool.run)
stats['output'] = output
logger.info("ctool finished")
elif operation['type'] == 'dsetool':
if 'nodes' not in operation:
operation['nodes'] = 'all'
if operation['nodes'] in ['all','ALL']:
nodes = [n for n in fab_config['hosts']]
else:
nodes = operation['nodes']
dsetool_options = operation['script']
logger.info("Running dsetool {command} on {nodes}".format(nodes=operation['nodes'], command=dsetool_options))
stats['command'] = dsetool_options
output = dsetool_cmd(nodes=nodes, options=dsetool_options)
stats['output'] = output
logger.info("dsetool command finished on all nodes")
elif operation['type'] == 'dse':
logger.info("Running dse command on {node}".format(node=operation['node']))
output = dse_cmd(node=operation['node'], options=operation['script'])
stats['output'] = output.split("\n")
stats['command'] = operation['script']
logger.info("dse commands finished")
end = datetime.datetime.now()
stats['end_date'] = end.isoformat()
stats['op_duration'] = str(end - start)
log_stats(stats, file=log)
finally:
# Copy node logs:
retrieve_logs_and_create_tarball(job_id=stats['id'])
revision_config['last_log'] = stats['id']
if capture_fincore:
stop_fincore_capture()
log_dir = os.path.join(CSTAR_PERF_LOGS_DIR, stats['id'])
retrieve_fincore_logs(log_dir)
# Restart fincore capture if this is not the last
# operation:
if operation_i < len(operations):
start_fincore_capture(interval=10)
if flamegraph.is_enabled(revision_config):
# Generate and Copy node flamegraphs
execute(flamegraph.stop_perf_agent)
execute(flamegraph.generate_flamegraph, rev_num)
flamegraph_dir = os.path.join(os.path.expanduser('~'),'.cstar_perf', 'flamegraph')
flamegraph_test_dir = os.path.join(flamegraph_dir, last_stress_operation_id)
retrieve_flamegraph(flamegraph_test_dir, rev_num+1)
sh.tar('cfvz', "{}.tar.gz".format(stats['id']), last_stress_operation_id, _cwd=flamegraph_dir)
shutil.rmtree(flamegraph_test_dir)
log_add_data(log, {'title':title,
'subtitle': subtitle,
'revisions': revisions})
if revisions[-1].get('leave_data', leave_data):
teardown(destroy=False, leave_data=True)
else:
kill_delay = 300 if profiler.yourkit_is_enabled(revision_config) else 0
teardown(destroy=True, leave_data=False, kill_delay=kill_delay)
if profiler.yourkit_is_enabled(revision_config):
yourkit_config = profiler.yourkit_get_config()
yourkit_dir = os.path.join(os.path.expanduser('~'),'.cstar_perf', 'yourkit')
yourkit_test_dir = os.path.join(yourkit_dir, last_stress_operation_id)
retrieve_yourkit(yourkit_test_dir, rev_num+1)
sh.tar('cfvz', "{}.tar.gz".format(stats['id']),
last_stress_operation_id, _cwd=yourkit_dir)
shutil.rmtree(yourkit_test_dir)
def stress_compare(revisions,
title,
log,
operations = [],
subtitle = '',
capture_fincore=False,
initial_destroy=True,
leave_data=False,
keep_page_cache=False,
git_fetch_before_test=True,
bootstrap_before_test=True,
teardown_after_test=True
):
"""
Run Stress on multiple C* branches and compare them.
revisions - List of dictionaries that contain cluster configurations
to trial. This is combined with the default config.
title - The title of the comparison
subtitle - A subtitle for more information (displayed smaller underneath)
log - The json file path to record stats to
operations - List of dictionaries indicating the operations. Example:
[# cassandra-stress command, node defaults to cluster defined 'stress_node'
{'type': 'stress',
'command': 'write n=19M -rate threads=50',
'node': 'node1',
'wait_for_compaction': True},
# nodetool command to run in parallel on nodes:
{'type': 'nodetool',
'command': 'decomission',
'nodes': ['node1','node2']},
# cqlsh script, node defaults to cluster defined 'stress_node'
{'type': 'cqlsh',
'script': "use my_ks; INSERT INTO blah (col1, col2) VALUES (val1, val2);",
'node': 'node1'}
]
capture_fincore - Enables capturing of linux-fincore logs of C* data files.
initial_destroy - Destroy all data before the first revision is run.
leave_data - Whether to leave the Cassandra data/commitlog/etc directories intact between revisions.
keep_page_cache - Whether to leave the linux page cache intact between revisions.
git_fetch_before_test (bool): If True, will update the cassandra.git with fab_common.git_repos
bootstrap_before_test (bool): If True, will bootstrap DSE / C* before running the operations
teardown_after_test (bool): If True, will shutdown DSE / C* after all of the operations
"""
validate_revisions_list(revisions)
validate_operations_list(operations)
pristine_config = copy.copy(fab_config)
# initial_destroy and git_fetch_before_test can be set in the job configuration,
# or manually in the call to this function.
# Either is fine, but they shouldn't conflict. If they do, a ValueError is raised.
initial_destroy = get_bool_if_method_and_config_values_do_not_conflict('initial_destroy',
initial_destroy,
pristine_config,
method_name='stress_compare')
if initial_destroy:
logger.info("Cleaning up from prior runs of stress_compare ...")
teardown(destroy=True, leave_data=False)
# https://datastax.jira.com/browse/CSTAR-633
git_fetch_before_test = get_bool_if_method_and_config_values_do_not_conflict('git_fetch_before_test',
git_fetch_before_test,
pristine_config,
method_name='stress_compare')
stress_shas = maybe_update_cassandra_git_and_setup_stress(operations, git_fetch=git_fetch_before_test)
# Flamegraph Setup
if flamegraph.is_enabled():
execute(flamegraph.setup)
with GracefulTerminationHandler() as handler:
for rev_num, revision_config in enumerate(revisions):
config = copy.copy(pristine_config)
config.update(revision_config)
revision = revision_config['revision']
config['log'] = log
config['title'] = title
config['subtitle'] = subtitle
product = dse if config.get('product') == 'dse' else cstar
# leave_data, bootstrap_before_test, and teardown_after_test can be set in the job configuration,
# or manually in the call to this function.
# Either is fine, but they shouldn't conflict. If they do, a ValueError is raised.
leave_data = get_bool_if_method_and_config_values_do_not_conflict('leave_data',
leave_data,
revision_config,
method_name='stress_compare')
# https://datastax.jira.com/browse/CSTAR-638
bootstrap_before_test = get_bool_if_method_and_config_values_do_not_conflict('bootstrap_before_test',
bootstrap_before_test,
revision_config,
method_name='stress_compare')
# https://datastax.jira.com/browse/CSTAR-639
teardown_after_test = get_bool_if_method_and_config_values_do_not_conflict('teardown_after_test',
teardown_after_test,
revision_config,
method_name='stress_compare')
logger.info("Bringing up {revision} cluster...".format(revision=revision))
# Drop the page cache between each revision, especially
# important when leave_data=True :
if not keep_page_cache:
drop_page_cache()
# Only fetch from git on the first run and if git_fetch_before_test is True
git_fetch_before_bootstrap = True if rev_num == 0 and git_fetch_before_test else False
if bootstrap_before_test:
revision_config['git_id'] = git_id = bootstrap(config,
destroy=initial_destroy,
leave_data=leave_data,
git_fetch=git_fetch_before_bootstrap)
else:
revision_config['git_id'] = git_id = config['revision']
if flamegraph.is_enabled(revision_config):
execute(flamegraph.ensure_stopped_perf_agent)
execute(flamegraph.start_perf_agent, rev_num)
if capture_fincore:
start_fincore_capture(interval=10)
last_stress_operation_id = 'None'
for operation_i, operation in enumerate(operations, 1):
try:
start = datetime.datetime.now()
stats = {
"id": str(uuid.uuid1()),
"type": operation['type'],
"revision": revision,
"git_id": git_id,
"start_date": start.isoformat(),
"label": revision_config.get('label', revision_config['revision']),
"test": '{operation_i}_{operation}'.format(
operation_i=operation_i,
operation=operation['type'])
}
if operation['type'] == 'stress':
last_stress_operation_id = stats['id']
# Default to all the nodes of the cluster if no
# nodes were specified in the command:
if operation.has_key('nodes'):
cmd = "{command} -node {hosts}".format(
command=operation['command'],
hosts=",".join(operation['nodes']))
elif '-node' in operation['command']:
cmd = operation['command']
else:
cmd = "{command} -node {hosts}".format(
command=operation['command'],
hosts=",".join([n for n in fab_config['hosts']]))
stats['command'] = cmd
stats['intervals'] = []
stats['test'] = '{operation_i}_{operation}'.format(
operation_i=operation_i, operation=cmd.strip().split(' ')[0]).replace(" ", "_")
logger.info('Running stress operation : {cmd} ...'.format(cmd=cmd))
# Run stress:
# (stress takes the stats as a parameter, and adds
# more as it runs):
stress_sha = stress_shas[operation.get('stress_revision', 'default')]
stats = stress(cmd, revision, stress_sha, stats=stats)
# Wait for all compactions to finish (unless disabled):
if operation.get('wait_for_compaction', True):
compaction_throughput = revision_config.get("compaction_throughput_mb_per_sec", 16)
wait_for_compaction(compaction_throughput=compaction_throughput)
elif operation['type'] == 'nodetool':
if 'nodes' not in operation:
operation['nodes'] = 'all'
if operation['nodes'] in ['all','ALL']:
nodes = [n for n in fab_config['hosts']]
else:
nodes = operation['nodes']
set_nodetool_path(os.path.join(product.get_bin_path(), 'nodetool'))
logger.info("Running nodetool on {nodes} with command: {command}".format(nodes=operation['nodes'], command=operation['command']))
stats['command'] = operation['command']
output = nodetool_multi(nodes, operation['command'])
stats['output'] = output
logger.info("Nodetool command finished on all nodes")
elif operation['type'] == 'cqlsh':
logger.info("Running cqlsh commands on {node}".format(node=operation['node']))
set_cqlsh_path(os.path.join(product.get_bin_path(), 'cqlsh'))
output = cqlsh(operation['script'], operation['node'])
stats['output'] = output.split("\n")
stats['command'] = operation['script']
logger.info("Cqlsh commands finished")
elif operation['type'] == 'bash':
nodes = operation.get('nodes', [n for n in fab_config['hosts']])
logger.info("Running bash commands on: {nodes}".format(nodes=nodes))
stats['output'] = bash(operation['script'], nodes)
stats['command'] = operation['script']
logger.info("Bash commands finished")
elif operation['type'] == 'spark_cassandra_stress':
nodes = operation.get('nodes', [n for n in fab_config['hosts']])
stress_node = config.get('stress_node', None)
# Note: once we have https://datastax.jira.com/browse/CSTAR-617, we should fix this to use
# client-tool when DSE_VERSION >= 4.8.0
# https://datastax.jira.com/browse/DSP-6025: dse client-tool
master_regex = re.compile(r"(.|\n)*(?P<master>spark:\/\/\d+.\d+.\d+.\d+:\d+)(.|\n)*")
master_out = dsetool_cmd(nodes[0], options='sparkmaster')[nodes[0]]
master_match = master_regex.match(master_out)
if not master_match:
raise ValueError('Could not find master address from "dsetool sparkmaster" cmd\n'
'Found output: {f}'.format(f=master_out))
master_string = master_match.group('master')
build_spark_cassandra_stress = bool(distutils.util.strtobool(
str(operation.get('build_spark_cassandra_stress', 'True'))))
remove_existing_spark_data = bool(distutils.util.strtobool(
str(operation.get('remove_existing_spark_data', 'True'))))
logger.info("Running spark_cassandra_stress on {stress_node} "
"using spark.cassandra.connection.host={node} and "
"spark-master {master}".format(stress_node=stress_node,
node=nodes[0],
master=master_string))
output = spark_cassandra_stress(operation['script'], nodes, stress_node=stress_node,
master=master_string,
build_spark_cassandra_stress=build_spark_cassandra_stress,
remove_existing_spark_data=remove_existing_spark_data)
stats['output'] = output.get('output', 'No output captured')
stats['spark_cass_stress_time_in_seconds'] = output.get('stats', {}).get('TimeInSeconds', 'No time captured')
stats['spark_cass_stress_ops_per_second'] = output.get('stats', {}).get('OpsPerSecond', 'No ops/s captured')
logger.info("spark_cassandra_stress finished")
elif operation['type'] == 'ctool':
logger.info("Running ctool with parameters: {command}".format(command=operation['command']))
ctool = Ctool(operation['command'], common.config)
output = execute(ctool.run)
stats['output'] = output
logger.info("ctool finished")
elif operation['type'] == 'dsetool':
if 'nodes' not in operation:
operation['nodes'] = 'all'
if operation['nodes'] in ['all','ALL']:
nodes = [n for n in fab_config['hosts']]
else:
nodes = operation['nodes']
dsetool_options = operation['script']
logger.info("Running dsetool {command} on {nodes}".format(nodes=operation['nodes'], command=dsetool_options))
stats['command'] = dsetool_options
output = dsetool_cmd(nodes=nodes, options=dsetool_options)
stats['output'] = output
logger.info("dsetool command finished on all nodes")
elif operation['type'] == 'dse':
logger.info("Running dse command on {node}".format(node=operation['node']))
output = dse_cmd(node=operation['node'], options=operation['script'])
stats['output'] = output.split("\n")
stats['command'] = operation['script']
logger.info("dse commands finished")
end = datetime.datetime.now()
stats['end_date'] = end.isoformat()
stats['op_duration'] = str(end - start)
log_stats(stats, file=log)
finally:
# Copy node logs:
retrieve_logs_and_create_tarball(job_id=stats['id'])
revision_config['last_log'] = stats['id']
if capture_fincore:
stop_fincore_capture()
log_dir = os.path.join(CSTAR_PERF_LOGS_DIR, stats['id'])
retrieve_fincore_logs(log_dir)
# Restart fincore capture if this is not the last
# operation:
if operation_i < len(operations):
start_fincore_capture(interval=10)
if flamegraph.is_enabled(revision_config):
# Generate and Copy node flamegraphs
execute(flamegraph.stop_perf_agent)
execute(flamegraph.generate_flamegraph, rev_num)
flamegraph_dir = os.path.join(os.path.expanduser('~'),'.cstar_perf', 'flamegraph')
flamegraph_test_dir = os.path.join(flamegraph_dir, last_stress_operation_id)
retrieve_flamegraph(flamegraph_test_dir, rev_num+1)
sh.tar('cfvz', "{}.tar.gz".format(stats['id']), last_stress_operation_id, _cwd=flamegraph_dir)
shutil.rmtree(flamegraph_test_dir)
log_add_data(log, {'title':title,
'subtitle': subtitle,
'revisions': revisions})
if teardown_after_test:
if revisions[-1].get('leave_data', leave_data):
teardown(destroy=False, leave_data=True)
else:
kill_delay = 300 if profiler.yourkit_is_enabled(revision_config) else 0
teardown(destroy=True, leave_data=False, kill_delay=kill_delay)
if profiler.yourkit_is_enabled(revision_config):
yourkit_config = profiler.yourkit_get_config()
yourkit_dir = os.path.join(os.path.expanduser('~'),'.cstar_perf', 'yourkit')
yourkit_test_dir = os.path.join(yourkit_dir, last_stress_operation_id)
retrieve_yourkit(yourkit_test_dir, rev_num+1)
sh.tar('cfvz', "{}.tar.gz".format(stats['id']),
last_stress_operation_id, _cwd=yourkit_dir)
shutil.rmtree(yourkit_test_dir)
def stress_compare(revisions,
title,
log,
operations=[],
subtitle='',
capture_fincore=False,
initial_destroy=True,
leave_data=False,
keep_page_cache=False,
git_fetch_before_test=True,
bootstrap_before_test=True,
teardown_after_test=True):
"""
Run Stress on multiple C* branches and compare them.
revisions - List of dictionaries that contain cluster configurations
to trial. This is combined with the default config.
title - The title of the comparison
subtitle - A subtitle for more information (displayed smaller underneath)
log - The json file path to record stats to
operations - List of dictionaries indicating the operations. Example:
[# cassandra-stress command, node defaults to cluster defined 'stress_node'
{'type': 'stress',
'command': 'write n=19M -rate threads=50',
'node': 'node1',
'wait_for_compaction': True},
# nodetool command to run in parallel on nodes:
{'type': 'nodetool',
'command': 'decomission',
'nodes': ['node1','node2']},
# cqlsh script, node defaults to cluster defined 'stress_node'
{'type': 'cqlsh',
'script': "use my_ks; INSERT INTO blah (col1, col2) VALUES (val1, val2);",
'node': 'node1'}
]
capture_fincore - Enables capturing of linux-fincore logs of C* data files.
initial_destroy - Destroy all data before the first revision is run.
leave_data - Whether to leave the Cassandra data/commitlog/etc directories intact between revisions.
keep_page_cache - Whether to leave the linux page cache intact between revisions.
git_fetch_before_test (bool): If True, will update the cassandra.git with fab_common.git_repos
bootstrap_before_test (bool): If True, will bootstrap DSE / C* before running the operations
teardown_after_test (bool): If True, will shutdown DSE / C* after all of the operations
"""
validate_revisions_list(revisions)
validate_operations_list(operations)
pristine_config = copy.copy(fab_config)
# initial_destroy and git_fetch_before_test can be set in the job configuration,
# or manually in the call to this function.
# Either is fine, but they shouldn't conflict. If they do, a ValueError is raised.
initial_destroy = get_bool_if_method_and_config_values_do_not_conflict(
'initial_destroy',
initial_destroy,
pristine_config,
method_name='stress_compare')
if initial_destroy:
logger.info("Cleaning up from prior runs of stress_compare ...")
teardown(destroy=True, leave_data=False)
# https://datastax.jira.com/browse/CSTAR-633
git_fetch_before_test = get_bool_if_method_and_config_values_do_not_conflict(
'git_fetch_before_test',
git_fetch_before_test,
pristine_config,
method_name='stress_compare')
stress_shas = maybe_update_cassandra_git_and_setup_stress(
operations, git_fetch=git_fetch_before_test)
# Flamegraph Setup
if flamegraph.is_enabled():
execute(flamegraph.setup)
with GracefulTerminationHandler() as handler:
for rev_num, revision_config in enumerate(revisions):
config = copy.copy(pristine_config)
config.update(revision_config)
revision = revision_config['revision']
config['log'] = log
config['title'] = title
config['subtitle'] = subtitle
product = dse if config.get('product') == 'dse' else cstar
# leave_data, bootstrap_before_test, and teardown_after_test can be set in the job configuration,
# or manually in the call to this function.
# Either is fine, but they shouldn't conflict. If they do, a ValueError is raised.
leave_data = get_bool_if_method_and_config_values_do_not_conflict(
'leave_data',
leave_data,
revision_config,
method_name='stress_compare')
# https://datastax.jira.com/browse/CSTAR-638
bootstrap_before_test = get_bool_if_method_and_config_values_do_not_conflict(
'bootstrap_before_test',
bootstrap_before_test,
revision_config,
method_name='stress_compare')
# https://datastax.jira.com/browse/CSTAR-639
teardown_after_test = get_bool_if_method_and_config_values_do_not_conflict(
'teardown_after_test',
teardown_after_test,
revision_config,
method_name='stress_compare')
logger.info(
"Bringing up {revision} cluster...".format(revision=revision))
# Drop the page cache between each revision, especially
# important when leave_data=True :
if not keep_page_cache:
drop_page_cache()
# Only fetch from git on the first run and if git_fetch_before_test is True
git_fetch_before_bootstrap = True if rev_num == 0 and git_fetch_before_test else False
if bootstrap_before_test:
revision_config['git_id'] = git_id = bootstrap(
config,
destroy=initial_destroy,
leave_data=leave_data,
git_fetch=git_fetch_before_bootstrap)
else:
revision_config['git_id'] = git_id = config['revision']
if flamegraph.is_enabled(revision_config):
execute(flamegraph.ensure_stopped_perf_agent)
execute(flamegraph.start_perf_agent, rev_num)
if capture_fincore:
start_fincore_capture(interval=10)
last_stress_operation_id = 'None'
for operation_i, operation in enumerate(operations, 1):
try:
start = datetime.datetime.now()
stats = {
"id":
str(uuid.uuid1()),
"type":
operation['type'],
"revision":
revision,
"git_id":
git_id,
"start_date":
start.isoformat(),
"label":
revision_config.get('label',
revision_config['revision']),
"test":
'{operation_i}_{operation}'.format(
operation_i=operation_i,
operation=operation['type'])
}
if operation['type'] == 'stress':
last_stress_operation_id = stats['id']
# Default to all the nodes of the cluster if no
# nodes were specified in the command:
if operation.has_key('nodes'):
cmd = "{command} -node {hosts}".format(
command=operation['command'],
hosts=",".join(operation['nodes']))
elif '-node' in operation['command']:
cmd = operation['command']
else:
cmd = "{command} -node {hosts}".format(
command=operation['command'],
hosts=",".join(
[n for n in fab_config['hosts']]))
stats['command'] = cmd
stats['intervals'] = []
stats['test'] = '{operation_i}_{operation}'.format(
operation_i=operation_i,
operation=cmd.strip().split(' ')[0]).replace(
" ", "_")
logger.info(
'Running stress operation : {cmd} ...'.format(
cmd=cmd))
# Run stress:
# (stress takes the stats as a parameter, and adds
# more as it runs):
stress_sha = stress_shas[operation.get(
'stress_revision', 'default')]
stats = stress(cmd, revision, stress_sha, stats=stats)
# Wait for all compactions to finish (unless disabled):
if operation.get('wait_for_compaction', True):
compaction_throughput = revision_config.get(
"compaction_throughput_mb_per_sec", 16)
wait_for_compaction(
compaction_throughput=compaction_throughput)
elif operation['type'] == 'nodetool':
if 'nodes' not in operation:
operation['nodes'] = 'all'
if operation['nodes'] in ['all', 'ALL']:
nodes = [n for n in fab_config['hosts']]
else:
nodes = operation['nodes']
set_nodetool_path(
os.path.join(product.get_bin_path(), 'nodetool'))
logger.info(
"Running nodetool on {nodes} with command: {command}"
.format(nodes=operation['nodes'],
command=operation['command']))
stats['command'] = operation['command']
output = nodetool_multi(nodes, operation['command'])
stats['output'] = output
logger.info("Nodetool command finished on all nodes")
elif operation['type'] == 'cqlsh':
logger.info("Running cqlsh commands on {node}".format(
node=operation['node']))
set_cqlsh_path(
os.path.join(product.get_bin_path(), 'cqlsh'))
output = cqlsh(operation['script'], operation['node'])
stats['output'] = output.split("\n")
stats['command'] = operation['script']
logger.info("Cqlsh commands finished")
elif operation['type'] == 'bash':
nodes = operation.get('nodes',
[n for n in fab_config['hosts']])
logger.info("Running bash commands on: {nodes}".format(
nodes=nodes))
stats['output'] = bash(operation['script'], nodes)
stats['command'] = operation['script']
logger.info("Bash commands finished")
elif operation['type'] == 'spark_cassandra_stress':
nodes = operation.get('nodes',
[n for n in fab_config['hosts']])
stress_node = config.get('stress_node', None)
# Note: once we have https://datastax.jira.com/browse/CSTAR-617, we should fix this to use
# client-tool when DSE_VERSION >= 4.8.0
# https://datastax.jira.com/browse/DSP-6025: dse client-tool
master_regex = re.compile(
r"(.|\n)*(?P<master>spark:\/\/\d+.\d+.\d+.\d+:\d+)(.|\n)*"
)
master_out = dsetool_cmd(
nodes[0], options='sparkmaster')[nodes[0]]
master_match = master_regex.match(master_out)
if not master_match:
raise ValueError(
'Could not find master address from "dsetool sparkmaster" cmd\n'
'Found output: {f}'.format(f=master_out))
master_string = master_match.group('master')
build_spark_cassandra_stress = bool(
distutils.util.strtobool(
str(
operation.get(
'build_spark_cassandra_stress',
'True'))))
remove_existing_spark_data = bool(
distutils.util.strtobool(
str(
operation.get('remove_existing_spark_data',
'True'))))
logger.info(
"Running spark_cassandra_stress on {stress_node} "
"using spark.cassandra.connection.host={node} and "
"spark-master {master}".format(
stress_node=stress_node,
node=nodes[0],
master=master_string))
output = spark_cassandra_stress(
operation['script'],
nodes,
stress_node=stress_node,
master=master_string,
build_spark_cassandra_stress=
build_spark_cassandra_stress,
remove_existing_spark_data=
remove_existing_spark_data)
stats['output'] = output.get('output',
'No output captured')
stats[
'spark_cass_stress_time_in_seconds'] = output.get(
'stats', {}).get('TimeInSeconds',
'No time captured')
stats['spark_cass_stress_ops_per_second'] = output.get(
'stats', {}).get('OpsPerSecond',
'No ops/s captured')
logger.info("spark_cassandra_stress finished")
elif operation['type'] == 'ctool':
logger.info(
"Running ctool with parameters: {command}".format(
command=operation['command']))
ctool = Ctool(operation['command'], common.config)
output = execute(ctool.run)
stats['output'] = output
logger.info("ctool finished")
elif operation['type'] == 'dsetool':
if 'nodes' not in operation:
operation['nodes'] = 'all'
if operation['nodes'] in ['all', 'ALL']:
nodes = [n for n in fab_config['hosts']]
else:
nodes = operation['nodes']
dsetool_options = operation['script']
logger.info(
"Running dsetool {command} on {nodes}".format(
nodes=operation['nodes'],
command=dsetool_options))
stats['command'] = dsetool_options
output = dsetool_cmd(nodes=nodes,
options=dsetool_options)
stats['output'] = output
logger.info("dsetool command finished on all nodes")
elif operation['type'] == 'dse':
logger.info("Running dse command on {node}".format(
node=operation['node']))
output = dse_cmd(node=operation['node'],
options=operation['script'])
stats['output'] = output.split("\n")
stats['command'] = operation['script']
logger.info("dse commands finished")
end = datetime.datetime.now()
stats['end_date'] = end.isoformat()
stats['op_duration'] = str(end - start)
log_stats(stats, file=log)
finally:
# Copy node logs:
retrieve_logs_and_create_tarball(job_id=stats['id'])
revision_config['last_log'] = stats['id']
if capture_fincore:
stop_fincore_capture()
log_dir = os.path.join(CSTAR_PERF_LOGS_DIR, stats['id'])
retrieve_fincore_logs(log_dir)
# Restart fincore capture if this is not the last
# operation:
if operation_i < len(operations):
start_fincore_capture(interval=10)
if flamegraph.is_enabled(revision_config):
# Generate and Copy node flamegraphs
execute(flamegraph.stop_perf_agent)
execute(flamegraph.generate_flamegraph, rev_num)
flamegraph_dir = os.path.join(os.path.expanduser('~'),
'.cstar_perf', 'flamegraph')
flamegraph_test_dir = os.path.join(flamegraph_dir,
last_stress_operation_id)
retrieve_flamegraph(flamegraph_test_dir, rev_num + 1)
sh.tar('cfvz',
"{}.tar.gz".format(stats['id']),
last_stress_operation_id,
_cwd=flamegraph_dir)
shutil.rmtree(flamegraph_test_dir)
log_add_data(log, {
'title': title,
'subtitle': subtitle,
'revisions': revisions
})
if teardown_after_test:
if revisions[-1].get('leave_data', leave_data):
teardown(destroy=False, leave_data=True)
else:
kill_delay = 300 if profiler.yourkit_is_enabled(
revision_config) else 0
teardown(destroy=True,
leave_data=False,
kill_delay=kill_delay)
if profiler.yourkit_is_enabled(revision_config):
yourkit_config = profiler.yourkit_get_config()
yourkit_dir = os.path.join(os.path.expanduser('~'),
'.cstar_perf', 'yourkit')
yourkit_test_dir = os.path.join(yourkit_dir,
last_stress_operation_id)
retrieve_yourkit(yourkit_test_dir, rev_num + 1)
sh.tar('cfvz',
"{}.tar.gz".format(stats['id']),
last_stress_operation_id,
_cwd=yourkit_dir)
shutil.rmtree(yourkit_test_dir)