diff options
Diffstat (limited to 'tools/gcp/stress_test/run_server.py')
| -rwxr-xr-x | tools/gcp/stress_test/run_server.py | 20 |
1 files changed, 16 insertions, 4 deletions
diff --git a/tools/gcp/stress_test/run_server.py b/tools/gcp/stress_test/run_server.py index 0d9a653d18..a666ae2900 100755 --- a/tools/gcp/stress_test/run_server.py +++ b/tools/gcp/stress_test/run_server.py @@ -30,6 +30,7 @@ import datetime import os +import resource import select import subprocess import sys @@ -56,6 +57,10 @@ def run_server(): might want to connect to the pod for examining logs). This is the reason why the script waits forever in case of failures. """ + # Set the 'core file' size to 'unlimited' so that 'core' files are generated + # if the server crashes (Note: This is not relevant for Java and Go servers) + resource.setrlimit(resource.RLIMIT_CORE, + (resource.RLIM_INFINITY, resource.RLIM_INFINITY)) # Read the parameters from environment variables env = dict(os.environ) @@ -69,13 +74,19 @@ def run_server(): dataset_id = env['DATASET_ID'] summary_table_id = env['SUMMARY_TABLE_ID'] qps_table_id = env['QPS_TABLE_ID'] + # The following parameter is to inform us whether the server runs forever + # until forcefully stopped or will it naturally stop after sometime. + # This way, we know that the process should not terminate (even if it does + # with a success exit code) and flag any termination as a failure. + will_run_forever = env.get('WILL_RUN_FOREVER', '1') logfile_name = env.get('LOGFILE_NAME') print('pod_name: %s, project_id: %s, run_id: %s, dataset_id: %s, ' - 'summary_table_id: %s, qps_table_id: %s') % ( - pod_name, project_id, run_id, dataset_id, summary_table_id, - qps_table_id) + 'summary_table_id: %s, qps_table_id: %s') % (pod_name, project_id, + run_id, dataset_id, + summary_table_id, + qps_table_id) bq_helper = BigQueryHelper(run_id, image_type, pod_name, project_id, dataset_id, summary_table_id, qps_table_id) @@ -106,7 +117,8 @@ def run_server(): stderr=subprocess.STDOUT) returncode = stress_p.wait() - if returncode != 0: + + if will_run_forever == '1' or returncode != 0: end_time = datetime.datetime.now().isoformat() event_type = EventType.FAILURE details = 'Returncode: %d; End time: %s' % (returncode, end_time) |