Add version 2 benchmark list formatting

This commit is contained in:
Joshua Boniface 2021-10-02 02:47:17 -04:00
parent 1e8841ce69
commit f70f052df1
1 changed files with 45 additions and 25 deletions

View File

@ -81,8 +81,8 @@ def format_ops_tohuman(dataops):
datahuman = '' datahuman = ''
for unit in sorted(ops_unit_matrix, key=ops_unit_matrix.get, reverse=True): for unit in sorted(ops_unit_matrix, key=ops_unit_matrix.get, reverse=True):
new_ops = int(math.ceil(dataops / ops_unit_matrix[unit])) new_ops = int(math.ceil(dataops / ops_unit_matrix[unit]))
# Round up if 5 or more digits # Round up if 6 or more digits
if new_ops > 9999: if new_ops > 99999:
# We can jump down another level # We can jump down another level
continue continue
else: else:
@ -1453,6 +1453,46 @@ def ceph_benchmark_list(config, job):
return retvalue, retdata return retvalue, retdata
def get_benchmark_list_results_legacy(benchmark_data):
benchmark_bandwidth = dict()
benchmark_iops = dict()
for test in ["seq_read", "seq_write", "rand_read_4K", "rand_write_4K"]:
benchmark_bandwidth[test] = format_bytes_tohuman(int(benchmark_data[test]['overall']['bandwidth']) * 1024)
benchmark_iops[test] = format_ops_tohuman(int(benchmark_data[test]['overall']['iops']))
return benchmark_bandwidth, benchmark_iops
def get_benchmark_list_results_json(benchmark_data):
benchmark_bandwidth = dict()
benchmark_iops = dict()
for test in ['seq_read', 'seq_write', 'rand_read_4K', 'rand_write_4K']:
benchmark_test_data = benchmark_data[test]
active_class = None
for io_class in ['read', 'write']:
if benchmark_test_data['jobs'][0][io_class]['io_bytes'] > 0:
active_class = io_class
if active_class is not None:
benchmark_bandwidth[test] = format_bytes_tohuman(int(benchmark_test_data['jobs'][0][active_class]['bw_bytes']))
benchmark_iops[test] = format_ops_tohuman(int(benchmark_test_data['jobs'][0][active_class]['iops']))
return benchmark_bandwidth, benchmark_iops
def get_benchmark_list_results(benchmark_format, benchmark_data):
if benchmark_format == 0:
benchmark_bandwidth, benchmark_iops = get_benchmark_list_results_legacy(benchmark_data)
elif benchmark_format == 1:
benchmark_bandwidth, benchmark_iops = get_benchmark_list_results_json(benchmark_data)
seq_benchmark_bandwidth = "{} / {}".format(benchmark_bandwidth['seq_read'], benchmark_bandwidth['seq_write'])
seq_benchmark_iops = "{} / {}".format(benchmark_iops['seq_read'], benchmark_iops['seq_write'])
rand_benchmark_bandwidth = "{} / {}".format(benchmark_bandwidth['rand_read_4K'], benchmark_bandwidth['rand_write_4K'])
rand_benchmark_iops = "{} / {}".format(benchmark_iops['rand_read_4K'], benchmark_iops['rand_write_4K'])
return seq_benchmark_bandwidth, seq_benchmark_iops, rand_benchmark_bandwidth, rand_benchmark_iops
def format_list_benchmark(config, benchmark_information): def format_list_benchmark(config, benchmark_information):
benchmark_list_output = [] benchmark_list_output = []
@ -1482,18 +1522,7 @@ def format_list_benchmark(config, benchmark_information):
continue continue
benchmark_data = benchmark['benchmark_result'] benchmark_data = benchmark['benchmark_result']
seq_benchmark_bandwidth, seq_benchmark_iops, rand_benchmark_bandwidth, rand_benchmark_iops = get_benchmark_list_results(benchmark_format, benchmark_data)
benchmark_bandwidth = dict()
benchmark_iops = dict()
for test in ["seq_read", "seq_write", "rand_read_4K", "rand_write_4K"]:
benchmark_data = benchmark['benchmark_result']
benchmark_bandwidth[test] = format_bytes_tohuman(int(benchmark_data[test]['overall']['bandwidth']) * 1024)
benchmark_iops[test] = format_ops_tohuman(int(benchmark_data[test]['overall']['iops']))
seq_benchmark_bandwidth = "{} / {}".format(benchmark_bandwidth['seq_read'], benchmark_bandwidth['seq_write'])
seq_benchmark_iops = "{} / {}".format(benchmark_iops['seq_read'], benchmark_iops['seq_write'])
rand_benchmark_bandwidth = "{} / {}".format(benchmark_bandwidth['rand_read_4K'], benchmark_bandwidth['rand_write_4K'])
rand_benchmark_iops = "{} / {}".format(benchmark_iops['rand_read_4K'], benchmark_iops['rand_write_4K'])
_benchmark_seq_bw_length = len(seq_benchmark_bandwidth) + 1 _benchmark_seq_bw_length = len(seq_benchmark_bandwidth) + 1
if _benchmark_seq_bw_length > benchmark_seq_bw_length: if _benchmark_seq_bw_length > benchmark_seq_bw_length:
@ -1558,17 +1587,8 @@ def format_list_benchmark(config, benchmark_information):
rand_benchmark_bandwidth = 'Running' rand_benchmark_bandwidth = 'Running'
rand_benchmark_iops = 'Running' rand_benchmark_iops = 'Running'
else: else:
benchmark_bandwidth = dict()
benchmark_iops = dict()
for test in ["seq_read", "seq_write", "rand_read_4K", "rand_write_4K"]:
benchmark_data = benchmark['benchmark_result'] benchmark_data = benchmark['benchmark_result']
benchmark_bandwidth[test] = format_bytes_tohuman(int(benchmark_data[test]['overall']['bandwidth']) * 1024) seq_benchmark_bandwidth, seq_benchmark_iops, rand_benchmark_bandwidth, rand_benchmark_iops = get_benchmark_list_results(benchmark_format, benchmark_data)
benchmark_iops[test] = format_ops_tohuman(int(benchmark_data[test]['overall']['iops']))
seq_benchmark_bandwidth = "{} / {}".format(benchmark_bandwidth['seq_read'], benchmark_bandwidth['seq_write'])
seq_benchmark_iops = "{} / {}".format(benchmark_iops['seq_read'], benchmark_iops['seq_write'])
rand_benchmark_bandwidth = "{} / {}".format(benchmark_bandwidth['rand_read_4K'], benchmark_bandwidth['rand_write_4K'])
rand_benchmark_iops = "{} / {}".format(benchmark_iops['rand_read_4K'], benchmark_iops['rand_write_4K'])
benchmark_list_output.append('{bold}\ benchmark_list_output.append('{bold}\
{benchmark_job: <{benchmark_job_length}} \ {benchmark_job: <{benchmark_job_length}} \