aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--functest/opnfv_tests/openstack/rally/rally.py175
-rw-r--r--functest/tests/unit/openstack/rally/test_rally.py14
2 files changed, 76 insertions, 113 deletions
diff --git a/functest/opnfv_tests/openstack/rally/rally.py b/functest/opnfv_tests/openstack/rally/rally.py
index 103c3a7e2..b3aaab659 100644
--- a/functest/opnfv_tests/openstack/rally/rally.py
+++ b/functest/opnfv_tests/openstack/rally/rally.py
@@ -21,6 +21,7 @@ import time
import uuid
import pkg_resources
+import prettytable
import yaml
from functest.core import testcase
@@ -354,8 +355,9 @@ class RallyBase(testcase.TestCase):
proc = subprocess.Popen(cmd, stdout=subprocess.PIPE,
stderr=subprocess.STDOUT)
- output = self._get_output(proc, test_name)
+ output = self.get_cmd_output(proc)
task_id = self.get_task_id(output)
+
LOGGER.debug('task_id : %s', task_id)
if task_id is None:
@@ -375,93 +377,62 @@ class RallyBase(testcase.TestCase):
self.RESULTS_DIR)
os.makedirs(self.RESULTS_DIR)
- # write html report file
- report_html_name = 'opnfv-{}.html'.format(test_name)
- report_html_dir = os.path.join(self.RESULTS_DIR, report_html_name)
- cmd = (["rally", "task", "report", task_id, "--out", report_html_dir])
-
+ # get and save rally operation JSON result
+ cmd = (["rally", "task", "detailed", task_id])
LOGGER.debug('running command: %s', cmd)
- subprocess.Popen(cmd, stdout=subprocess.PIPE,
- stderr=subprocess.STDOUT)
+ proc = subprocess.Popen(cmd, stdout=subprocess.PIPE,
+ stderr=subprocess.STDOUT)
+ json_detailed = self.get_cmd_output(proc)
+ LOGGER.info('%s', json_detailed)
- # get and save rally operation JSON result
cmd = (["rally", "task", "results", task_id])
LOGGER.debug('running command: %s', cmd)
proc = subprocess.Popen(cmd, stdout=subprocess.PIPE,
stderr=subprocess.STDOUT)
json_results = self.get_cmd_output(proc)
+ self._append_summary(json_results, test_name)
report_json_name = 'opnfv-{}.json'.format(test_name)
report_json_dir = os.path.join(self.RESULTS_DIR, report_json_name)
with open(report_json_dir, 'w') as r_file:
LOGGER.debug('saving json file')
r_file.write(json_results)
+ # write html report file
+ report_html_name = 'opnfv-{}.html'.format(test_name)
+ report_html_dir = os.path.join(self.RESULTS_DIR, report_html_name)
+ cmd = (["rally", "task", "report", task_id, "--out", report_html_dir])
+ LOGGER.debug('running command: %s', cmd)
+ subprocess.Popen(cmd, stdout=subprocess.PIPE,
+ stderr=subprocess.STDOUT)
+
# parse JSON operation result
if self.task_succeed(json_results):
LOGGER.info('Test scenario: "{}" OK.'.format(test_name) + "\n")
else:
LOGGER.info('Test scenario: "{}" Failed.'.format(test_name) + "\n")
- def _get_output(self, proc, test_name):
- result = ""
+ def _append_summary(self, json_raw, test_name):
nb_tests = 0
+ nb_success = 0
overall_duration = 0.0
- success = 0.0
- nb_totals = 0
- for line in proc.stdout:
- if ("Load duration" in line or
- "started" in line or
- "finished" in line or
- " Preparing" in line or
- "+-" in line or
- "|" in line):
- result += line
- elif "test scenario" in line:
- result += "\n" + line
- elif "Full duration" in line:
- result += line + "\n\n"
-
- # parse output for summary report
- if ("| " in line and
- "| action" not in line and
- "| Starting" not in line and
- "| Completed" not in line and
- "| ITER" not in line and
- "| " not in line and
- "| total" not in line):
- nb_tests += 1
- elif "| total" in line:
- percentage = ((line.split('|')[8]).strip(' ')).strip('%')
- try:
- success += float(percentage)
- except ValueError:
- LOGGER.info('Percentage error: %s, %s',
- percentage, line)
- nb_totals += 1
- elif "Full duration" in line:
- duration = line.split(': ')[1]
- try:
- overall_duration += float(duration)
- except ValueError:
- LOGGER.info('Duration error: %s, %s', duration, line)
-
- overall_duration = "{:10.2f}".format(overall_duration)
- if nb_totals == 0:
- success_avg = 0
- else:
- success_avg = "{:0.2f}".format(success / nb_totals)
+ rally_report = json.loads(json_raw)
+ for report in rally_report:
+ if report.get('full_duration'):
+ overall_duration += report.get('full_duration')
+
+ if report.get('result'):
+ for result in report.get('result'):
+ nb_tests += 1
+ if not result.get('error'):
+ nb_success += 1
scenario_summary = {'test_name': test_name,
'overall_duration': overall_duration,
'nb_tests': nb_tests,
- 'success': success_avg}
+ 'nb_success': nb_success}
self.summary.append(scenario_summary)
- LOGGER.debug("\n" + result)
-
- return result
-
def _prepare_env(self):
LOGGER.debug('Validating the test name...')
if self.test_name not in self.TESTS:
@@ -559,75 +530,57 @@ class RallyBase(testcase.TestCase):
self._run_task(self.test_name)
def _generate_report(self):
- report = (
- "\n"
- " "
- "\n"
- " Rally Summary Report\n"
- "\n"
- "+===================+============+===============+===========+"
- "\n"
- "| Module | Duration | nb. Test Run | Success |"
- "\n"
- "+===================+============+===============+===========+"
- "\n")
+ total_duration = 0.0
+ total_nb_tests = 0
+ total_nb_success = 0
payload = []
+ res_table = prettytable.PrettyTable(
+ padding_width=2,
+ field_names=['Module', 'Duration', 'nb. Test Run', 'Success'])
+ res_table.align['Module'] = "l"
+ res_table.align['Duration'] = "r"
+ res_table.align['Success'] = "r"
+
# for each scenario we draw a row for the table
- total_duration = 0.0
- total_nb_tests = 0
- total_success = 0.0
for item in self.summary:
- name = "{0:<17}".format(item['test_name'])
- duration = float(item['overall_duration'])
- total_duration += duration
- duration = time.strftime("%M:%S", time.gmtime(duration))
- duration = "{0:<10}".format(duration)
- nb_tests = "{0:<13}".format(item['nb_tests'])
- total_nb_tests += int(item['nb_tests'])
- success = "{0:<10}".format(str(item['success']) + '%')
- total_success += float(item['success'])
- report += ("" +
- "| " + name + " | " + duration + " | " +
- nb_tests + " | " + success + "|\n" +
- "+-------------------+------------"
- "+---------------+-----------+\n")
- payload.append({'module': name,
+ total_duration += item['overall_duration']
+ total_nb_tests += item['nb_tests']
+ total_nb_success += item['nb_success']
+ try:
+ success_avg = 100 * item['nb_success'] / item['nb_tests']
+ except ZeroDivisionError:
+ success_avg = 0
+ success_str = str("{:0.2f}".format(success_avg)) + '%'
+ duration_str = time.strftime("%M:%S",
+ time.gmtime(item['overall_duration']))
+ res_table.add_row([item['test_name'], duration_str,
+ item['nb_tests'], success_str])
+ payload.append({'module': item['test_name'],
'details': {'duration': item['overall_duration'],
'nb tests': item['nb_tests'],
- 'success': item['success']}})
+ 'success': success_str}})
total_duration_str = time.strftime("%H:%M:%S",
time.gmtime(total_duration))
- total_duration_str2 = "{0:<10}".format(total_duration_str)
- total_nb_tests_str = "{0:<13}".format(total_nb_tests)
-
try:
- self.result = total_success / len(self.summary)
+ self.result = 100 * total_nb_success / total_nb_tests
except ZeroDivisionError:
self.result = 100
-
success_rate = "{:0.2f}".format(self.result)
- success_rate_str = "{0:<10}".format(str(success_rate) + '%')
- report += ("+===================+============"
- "+===============+===========+")
- report += "\n"
- report += ("| TOTAL: | " + total_duration_str2 + " | " +
- total_nb_tests_str + " | " + success_rate_str + "|\n")
- report += ("+===================+============"
- "+===============+===========+")
- report += "\n"
-
- LOGGER.info("\n" + report)
+ success_rate_str = str(success_rate) + '%'
+ res_table.add_row(["", "", "", ""])
+ res_table.add_row(["TOTAL:", total_duration_str, total_nb_tests,
+ success_rate_str])
+
+ LOGGER.info("Rally Summary Report:\n\n%s\n", res_table.get_string())
+ LOGGER.info("Rally '%s' success_rate is %s%%",
+ self.case_name, success_rate)
payload.append({'summary': {'duration': total_duration,
'nb tests': total_nb_tests,
'nb success': success_rate}})
-
self.details = payload
- LOGGER.info("Rally '%s' success_rate is %s%%",
- self.case_name, success_rate)
-
def _clean_up(self):
for creator in reversed(self.creators):
try:
diff --git a/functest/tests/unit/openstack/rally/test_rally.py b/functest/tests/unit/openstack/rally/test_rally.py
index f48585274..ba8a241fb 100644
--- a/functest/tests/unit/openstack/rally/test_rally.py
+++ b/functest/tests/unit/openstack/rally/test_rally.py
@@ -220,7 +220,7 @@ class OSRallyTesting(unittest.TestCase):
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
'_build_task_args', return_value={})
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
- '_get_output')
+ '_append_summary')
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
'get_task_id', return_value=None)
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
@@ -242,7 +242,7 @@ class OSRallyTesting(unittest.TestCase):
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
'_build_task_args', return_value={})
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
- '_get_output')
+ '_append_summary')
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
'get_task_id', return_value='1')
@mock.patch('functest.opnfv_tests.openstack.rally.rally.RallyBase.'
@@ -424,6 +424,16 @@ class OSRallyTesting(unittest.TestCase):
self.assertEqual(self.rally_base.run(), testcase.TestCase.EX_RUN_ERROR)
mock_prep_env.assert_called()
+ def test_append_summary(self):
+ text = '[{"result":[{"error":[]},{"error":["err"]}],' \
+ '"full_duration": 17.312026}]'
+ self.rally_base._append_summary(text, "foo_test")
+ self.assertEqual(self.rally_base.summary[0]['test_name'], "foo_test")
+ self.assertEqual(self.rally_base.summary[0]['overall_duration'],
+ 17.312026)
+ self.assertEqual(self.rally_base.summary[0]['nb_tests'], 2)
+ self.assertEqual(self.rally_base.summary[0]['nb_success'], 1)
+
if __name__ == "__main__":
logging.disable(logging.CRITICAL)