def RunDd(vms, data_size_in_mb, metadata):
"""Run dd benchmark and parses results.
Args:
vms: The VMs running dd benchmarks.
data_size_in_mb: The size of the data file in MB.
metadata: The metadata to attach to the sample.
Returns:
A list of samples. Each sample is a 4-tuple of (benchmark_name, value, unit,
metadata), as accepted by PerfKitBenchmarkerPublisher.AddSamples.
"""
vm = vms[0]
cmd = ('rm -rf %s/*; sudo sync; sudo sysctl vm.drop_caches=3; '
'time for i in {0..99}; do dd if=%s/data/file-$i.dat '
'of=%s/file-$i.dat bs=262144; done' %
(vm.GetScratchDir(1), vm.GetScratchDir(0), vm.GetScratchDir(1)))
_, res = vm.RemoteCommand(cmd)
logging.info(res)
time_used = vm_util.ParseTimeCommandResult(res)
return [
sample.Sample('dd throughput',
data_size_in_mb / time_used,
UNIT,
metadata=metadata)
]
def RunForIpAddress(ip_address, ip_type):
"""Run SCP benchmark against a destination IP address."""
target_dir = posixpath.join(receiving_vm.GetScratchDir(0), ip_type)
cmd = cmd_template % (ip_address, target_dir)
receiving_vm.RemoteCommand('mkdir %s' % target_dir)
meta = metadata.copy()
meta['ip_type'] = ip_type
_, res = sending_vm.RemoteCommand(cmd)
time_used = vm_util.ParseTimeCommandResult(res)
result = DATA_SIZE_IN_MB / time_used
receiving_vm.RemoteCommand('rm -rf %s' % target_dir)
return sample.Sample('scp throughput', result, UNIT, meta)
def RunCp(vms):
"""Runs cp benchmarks and parses results.
Args:
vms: The VMs running cp benchmarks.
Returns:
A list of sample.Sample objects.
"""
cmd = ('rm -rf %s/*; sudo sync; sudo sysctl vm.drop_caches=3; '
'time cp %s/data/* %s/; ' %
(vms[0].GetScratchDir(1), vms[0].GetScratchDir(0),
vms[0].GetScratchDir(1)))
_, res = vms[0].RemoteCommand(cmd)
logging.info(res)
time_used = vm_util.ParseTimeCommandResult(res)
return [sample.Sample('cp throughput', DATA_SIZE_IN_MB / time_used, UNIT)]
def RunCp(vms, data_size_in_mb, metadata):
"""Runs cp benchmarks and parses results.
Args:
vms: The VMs running cp benchmarks.
data_size_in_mb: The size of the data file in MB.
metadata: The base metadata to attach to the sample.
Returns:
A list of sample.Sample objects.
"""
cmd = ('rm -rf %s/*; sudo sync; sudo sysctl vm.drop_caches=3; '
'time cp %s/data/* %s/; ' %
(vms[0].GetScratchDir(1), vms[0].GetScratchDir(0),
vms[0].GetScratchDir(1)))
_, res = vms[0].RemoteCommand(cmd)
logging.info(res)
time_used = vm_util.ParseTimeCommandResult(res)
return [sample.Sample('cp throughput', data_size_in_mb / time_used, UNIT,
metadata=metadata)]
def _CliBasedTests(output_results, metadata, vm, iteration_count,
clean_up_bucket_cmd, upload_cmd, cleanup_local_temp_cmd,
download_cmd):
""" Performs tests via cli tools
We will upload and download a set of files from/to a local directory via
cli tools and observe the throughput.
Args:
output_results: The collection to put results in.
metadata: The metadata to be included in the result.
vm: The vm to run the tests.
iteration_count: The number of iterations to run for this test.
clean_up_bucket_cmd: The command to run to cleanup the bucket.
upload_cmd: The command to run to upload the objects.
cleanup_local_temp_cmd: The command to run to cleanup the local temp dir.
download_cmd: The command to run to download the content.
Raises:
NotEnoughResultsError: if we failed too many times to upload or download.
"""
if (FLAGS.object_storage_scenario != 'all'
and FLAGS.object_storage_scenario != 'cli'):
# User does not want to run this scenario, do nothing.
return
# CLI tool based tests.
cli_upload_results = []
cli_download_results = []
data_size_in_mbits = 0
if FLAGS.cli_test_size == 'normal':
data_size_in_mbits = DATA_SIZE_IN_MBITS
else:
data_size_in_mbits = LARGE_DATA_SIZE_IN_MBITS
for _ in range(iteration_count):
vm.RemoteCommand(clean_up_bucket_cmd, ignore_failure=True)
upload_successful = False
try:
_, res = vm.RemoteCommand(upload_cmd)
upload_successful = True
except:
logging.info('failed to upload, skip this iteration.')
pass
if upload_successful:
logging.debug(res)
throughput = data_size_in_mbits / vm_util.ParseTimeCommandResult(
res)
# Output some log traces to show we are making progress
logging.info('cli upload throughput %f', throughput)
cli_upload_results.append(throughput)
download_successful = False
vm.RemoteCommand(cleanup_local_temp_cmd, ignore_failure=True)
try:
_, res = vm.RemoteCommand(download_cmd)
download_successful = True
except:
logging.info('failed to download, skip this iteration.')
pass
if download_successful:
logging.debug(res)
throughput = data_size_in_mbits / vm_util.ParseTimeCommandResult(
res)
logging.info('cli download throughput %f', throughput)
cli_download_results.append(throughput)
expected_successes = iteration_count * (1 - CLI_TEST_FAILURE_TOLERANCE)
if (len(cli_download_results) < expected_successes
or len(cli_upload_results) < expected_successes):
raise NotEnoughResultsError(
'Failed to complete the required number of '
'iterations.')
# Report various percentiles.
metrics_prefix = ''
if FLAGS.cli_test_size != 'normal':
metrics_prefix = '%s ' % FLAGS.cli_test_size
_AppendPercentilesToResults(
output_results, cli_upload_results,
'%s%s' % (metrics_prefix, UPLOAD_THROUGHPUT_VIA_CLI), THROUGHPUT_UNIT,
metadata)
_AppendPercentilesToResults(
output_results, cli_download_results,
'%s%s' % (metrics_prefix, DOWNLOAD_THROUGHPUT_VIA_CLI),
THROUGHPUT_UNIT, metadata)
def CLIThroughputBenchmark(output_results, metadata, vm, command_builder,
service, bucket, regional_bucket):
"""A benchmark for CLI tool throughput.
We will upload and download a set of files from/to a local directory
via cli tools and observe the throughput.
Args:
results: the results array to append to.
metadata: a dictionary of metadata to add to samples.
vm: the VM to run the benchmark on.
command_builder: an APIScriptCommandBuilder.
service: an ObjectStorageService.
bucket_name: the primary bucket to benchmark.
regional_bucket_name: the secondary bucket to benchmark.
Raises:
NotEnoughResultsError: if we failed too many times to upload or download.
"""
data_directory = '%s/run/data' % vm.GetScratchDir()
download_directory = '%s/run/temp' % vm.GetScratchDir()
# The real solution to the iteration count issue is dynamically
# choosing the number of iterations based on how long they
# take. This will work for now, though.
if FLAGS.storage == providers.AZURE:
iteration_count = CLI_TEST_ITERATION_COUNT_AZURE
elif FLAGS.cli_test_size == 'normal':
iteration_count = CLI_TEST_ITERATION_COUNT
else:
iteration_count = LARGE_CLI_TEST_ITERATION_COUNT
# The CLI-based tests require some provisioning on the VM first.
vm.RemoteCommand('cd %s/run/; bash cloud-storage-workload.sh %s' %
(vm.GetScratchDir(), FLAGS.cli_test_size))
# CLI tool based tests.
cli_upload_results = []
cli_download_results = []
if FLAGS.cli_test_size == 'normal':
data_size_in_mbits = DATA_SIZE_IN_MBITS
file_names = ['file-%s.dat' % i for i in range(100)]
else:
data_size_in_mbits = LARGE_DATA_SIZE_IN_MBITS
file_names = ['file_large_3gib.dat']
for _ in range(iteration_count):
try:
service.EmptyBucket(bucket)
except Exception:
pass
try:
_, res = service.CLIUploadDirectory(vm, data_directory, file_names,
bucket)
except errors.VirtualMachine.RemoteCommandError:
logging.info('failed to upload, skip this iteration.')
continue
throughput = data_size_in_mbits / vm_util.ParseTimeCommandResult(res)
logging.info('cli upload throughput %f', throughput)
cli_upload_results.append(throughput)
try:
vm.RemoveFile(posixpath.join(download_directory, '*'))
except Exception:
pass
try:
_, res = service.CLIDownloadBucket(vm, bucket, file_names,
download_directory)
except errors.VirtualMachine.RemoteCommandError:
logging.info('failed to download, skip this iteration.')
continue
throughput = data_size_in_mbits / vm_util.ParseTimeCommandResult(res)
logging.info('cli download throughput %f', throughput)
cli_download_results.append(throughput)
expected_successes = iteration_count * (1 - CLI_TEST_FAILURE_TOLERANCE)
if (len(cli_download_results) < expected_successes
or len(cli_upload_results) < expected_successes):
raise NotEnoughResultsError(
'Failed to complete the required number of '
'iterations.')
# Report various percentiles.
metrics_prefix = ''
if FLAGS.cli_test_size != 'normal':
metrics_prefix = '%s ' % FLAGS.cli_test_size
_AppendPercentilesToResults(
output_results, cli_upload_results,
'%s%s' % (metrics_prefix, UPLOAD_THROUGHPUT_VIA_CLI), THROUGHPUT_UNIT,
metadata)
_AppendPercentilesToResults(
output_results, cli_download_results,
'%s%s' % (metrics_prefix, DOWNLOAD_THROUGHPUT_VIA_CLI),
THROUGHPUT_UNIT, metadata)
