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 settting 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 = 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)
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 settting 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 = 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=True, 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")
end = datetime.datetime.now()
stats['end_date'] = end.isoformat()
stats['op_duration'] = str(end - start)
log_stats(stats, file=log)
finally:
# Copy node logs:
logs_dir = os.path.join(os.path.expanduser('~'),'.cstar_perf','logs')
log_dir = os.path.join(logs_dir, stats['id'])
os.makedirs(log_dir)
retrieve_logs(log_dir)
revision_config['last_log'] = stats['id']
# Tar them for archiving:
subprocess.Popen(shlex.split('tar cfvz {id}.tar.gz {id}'.format(id=stats['id'])), cwd=logs_dir).communicate()
shutil.rmtree(log_dir)
if capture_fincore:
stop_fincore_capture()
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
):
"""
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=19000000 -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 settting 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 = 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)
# Clean stress builds
stress_builds = [b for b in os.listdir(CASSANDRA_STRESS_PATH)
if b not in ['default', 'trunk']]
for stress_build in stress_builds:
path = os.path.join(CASSANDRA_STRESS_PATH, stress_build)
logger.info("Removing stress build '{}'".format(path))
shutil.rmtree(path)
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
# leave_data settting 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 = 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=True, leave_data=leave_data, git_fetch=git_fetch)
if capture_fincore:
start_fincore_capture(interval=10)
for operation_i, operation in enumerate(operations, 1):
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'])}
if operation['type'] == 'stress':
# 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(host=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):
stats = stress(cmd, revision, stats, stress_revision=revision_config.get('stress_revision', None))
# 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']
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']))
output = cqlsh(operation['script'], operation['node'])
stats['output'] = output.split("\n")
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 {node}".format(nodes=nodes))
output = bash(operation['script'], nodes)
stats['output'] = output.split("\n")
logger.info("Bash commands finished")
end = datetime.datetime.now()
stats['end_date'] = end.isoformat()
stats['op_duration'] = str(end - start)
log_stats(stats, file=log)
#Copy node logs:
logs_dir = os.path.join(os.path.expanduser('~'),'.cstar_perf','logs')
log_dir = os.path.join(logs_dir, stats['id'])
os.makedirs(log_dir)
retrieve_logs(log_dir)
revision_config['last_log'] = stats['id']
#Tar them for archiving:
subprocess.Popen(shlex.split('tar cfvz {id}.tar.gz {id}'.format(id=stats['id'])), cwd=logs_dir).communicate()
shutil.rmtree(log_dir)
if capture_fincore:
stop_fincore_capture()
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)
log_add_data(log, {'title':title,
'subtitle': subtitle,
'revisions': revisions})
if revisions[-1].get('leave_data', False):
teardown(destroy=False, leave_data=True)
else:
teardown(destroy=True, leave_data=False)
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 settting 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 = 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)
clean_stress()
stress_revisions = set([
operation['stress_revision'] for operation in operations
if 'stress_revision' in operation
])
stress_shas = setup_stress(stress_revisions)
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['product'] == 'dse' else cstar
# leave_data settting 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 = 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=True,
leave_data=leave_data,
git_fetch=git_fetch)
if capture_fincore:
start_fincore_capture(interval=10)
for operation_i, operation in enumerate(operations, 1):
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'])
}
if operation['type'] == 'stress':
# 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(host=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")
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)
logger.info("Bash commands finished")
end = datetime.datetime.now()
stats['end_date'] = end.isoformat()
stats['op_duration'] = str(end - start)
log_stats(stats, file=log)
#Copy node logs:
logs_dir = os.path.join(os.path.expanduser('~'), '.cstar_perf',
'logs')
log_dir = os.path.join(logs_dir, stats['id'])
os.makedirs(log_dir)
retrieve_logs(log_dir)
revision_config['last_log'] = stats['id']
#Tar them for archiving:
subprocess.Popen(shlex.split(
'tar cfvz {id}.tar.gz {id}'.format(id=stats['id'])),
cwd=logs_dir).communicate()
shutil.rmtree(log_dir)
if capture_fincore:
stop_fincore_capture()
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)
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:
teardown(destroy=True, leave_data=False)
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 settting 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 = 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)
clean_stress()
stress_revisions = set([operation["stress_revision"] for operation in operations if "stress_revision" in operation])
stress_shas = setup_stress(stress_revisions)
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["product"] == "dse" else cstar
# leave_data settting 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 = 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=True, leave_data=leave_data, git_fetch=git_fetch)
if capture_fincore:
start_fincore_capture(interval=10)
for operation_i, operation in enumerate(operations, 1):
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"]),
}
if operation["type"] == "stress":
# 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(host=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")
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)
logger.info("Bash commands finished")
end = datetime.datetime.now()
stats["end_date"] = end.isoformat()
stats["op_duration"] = str(end - start)
log_stats(stats, file=log)
# Copy node logs:
logs_dir = os.path.join(os.path.expanduser("~"), ".cstar_perf", "logs")
log_dir = os.path.join(logs_dir, stats["id"])
os.makedirs(log_dir)
retrieve_logs(log_dir)
revision_config["last_log"] = stats["id"]
# Tar them for archiving:
subprocess.Popen(
shlex.split("tar cfvz {id}.tar.gz {id}".format(id=stats["id"])), cwd=logs_dir
).communicate()
shutil.rmtree(log_dir)
if capture_fincore:
stop_fincore_capture()
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)
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:
teardown(destroy=True, leave_data=False)