|
|
@ -101,7 +101,7 @@ for bm_name in sys.argv[1:]: |
|
|
|
if len(benchmarks) >= min(4, multiprocessing.cpu_count()): |
|
|
|
if len(benchmarks) >= min(4, multiprocessing.cpu_count()): |
|
|
|
# run up to half the cpu count: each benchmark can use up to two cores |
|
|
|
# run up to half the cpu count: each benchmark can use up to two cores |
|
|
|
# (one for the microbenchmark, one for the data flush) |
|
|
|
# (one for the microbenchmark, one for the data flush) |
|
|
|
jobset.run(benchmarks, maxjobs=multiprocessing.cpu_count()/2, |
|
|
|
jobset.run(benchmarks, maxjobs=max(1, multiprocessing.cpu_count()/2), |
|
|
|
add_env={'GRPC_TEST_PORT_SERVER': 'localhost:%d' % port_server_port}) |
|
|
|
add_env={'GRPC_TEST_PORT_SERVER': 'localhost:%d' % port_server_port}) |
|
|
|
jobset.run(profile_analysis, maxjobs=multiprocessing.cpu_count()) |
|
|
|
jobset.run(profile_analysis, maxjobs=multiprocessing.cpu_count()) |
|
|
|
jobset.run(cleanup, maxjobs=multiprocessing.cpu_count()) |
|
|
|
jobset.run(cleanup, maxjobs=multiprocessing.cpu_count()) |
|
|
@ -110,7 +110,7 @@ for bm_name in sys.argv[1:]: |
|
|
|
cleanup = [] |
|
|
|
cleanup = [] |
|
|
|
# run the remaining benchmarks that weren't flushed |
|
|
|
# run the remaining benchmarks that weren't flushed |
|
|
|
if len(benchmarks): |
|
|
|
if len(benchmarks): |
|
|
|
jobset.run(benchmarks, maxjobs=multiprocessing.cpu_count()/2, |
|
|
|
jobset.run(benchmarks, maxjobs=max(1, multiprocessing.cpu_count()/2), |
|
|
|
add_env={'GRPC_TEST_PORT_SERVER': 'localhost:%d' % port_server_port}) |
|
|
|
add_env={'GRPC_TEST_PORT_SERVER': 'localhost:%d' % port_server_port}) |
|
|
|
jobset.run(profile_analysis, maxjobs=multiprocessing.cpu_count()) |
|
|
|
jobset.run(profile_analysis, maxjobs=multiprocessing.cpu_count()) |
|
|
|
jobset.run(cleanup, maxjobs=multiprocessing.cpu_count()) |
|
|
|
jobset.run(cleanup, maxjobs=multiprocessing.cpu_count()) |
|
|
|