diff options
author | Ross Brattain <ross.b.brattain@intel.com> | 2017-10-11 13:30:14 -0700 |
---|---|---|
committer | Ross Brattain <ross.b.brattain@intel.com> | 2017-10-13 13:59:37 -0700 |
commit | f8fc6a898ea65e8f98a3e9e2ff9483b437a0e33f (patch) | |
tree | 5dc61933f7f47b2b280214a61f00bb7df4fd818d | |
parent | 06f8d7a47036fe92698a9b528282a2e96eb2d32a (diff) |
task: drain background runner queues
We were not draining the queues in the background
runners.
Modify the main runner_join loop to loop over
all the runners and drain them.
The runner join method does extra work for periodic actions
so we can't change its behavior.
Instead create a new poll() method and use that
to check runner status
Change-Id: I9466ba40a6a4c45c82cedff279cbb4817c6b66ad
Signed-off-by: Ross Brattain <ross.b.brattain@intel.com>
-rw-r--r-- | yardstick/benchmark/core/task.py | 20 | ||||
-rwxr-xr-x | yardstick/benchmark/runners/base.py | 4 |
2 files changed, 18 insertions, 6 deletions
diff --git a/yardstick/benchmark/core/task.py b/yardstick/benchmark/core/task.py index 53298d8d3..1512ca718 100644 --- a/yardstick/benchmark/core/task.py +++ b/yardstick/benchmark/core/task.py @@ -39,7 +39,6 @@ output_file_default = "/tmp/yardstick.out" config_file = '/etc/yardstick/yardstick.conf' test_cases_dir_default = "tests/opnfv/test_cases/" LOG = logging.getLogger(__name__) -JOIN_TIMEOUT = 60 class Task(object): # pragma: no cover @@ -260,7 +259,7 @@ class Task(object): # pragma: no cover # Wait for runners to finish for runner in runners: - status = runner_join(runner, self.outputs, result) + status = runner_join(runner, background_runners, self.outputs, result) if status != 0: raise RuntimeError( "{0} runner status {1}".format(runner.__execution_type__, status)) @@ -270,7 +269,7 @@ class Task(object): # pragma: no cover for scenario in scenarios: if not _is_background_scenario(scenario): runner = self.run_one_scenario(scenario, output_file) - status = runner_join(runner, self.outputs, result) + status = runner_join(runner, background_runners, self.outputs, result) if status != 0: LOG.error('Scenario NO.%s: "%s" ERROR!', scenarios.index(scenario) + 1, @@ -285,11 +284,11 @@ class Task(object): # pragma: no cover # Wait for background runners to finish for runner in background_runners: - status = runner.join(self.outputs, result, JOIN_TIMEOUT) + status = runner.join(self.outputs, result) if status is None: # Nuke if it did not stop nicely base_runner.Runner.terminate(runner) - runner.join(self.outputs, result, JOIN_TIMEOUT) + runner.join(self.outputs, result) base_runner.Runner.release(runner) print("Background task ended") @@ -641,13 +640,22 @@ def get_networks_from_nodes(nodes): return networks -def runner_join(runner, outputs, result): +def runner_join(runner, background_runners, outputs, result): """join (wait for) a runner, exit process at runner failure + :param background_runners: + :type background_runners: :param outputs: :type outputs: dict :param result: :type result: list """ + while runner.poll() is None: + outputs.update(runner.get_output()) + result.extend(runner.get_result()) + # drain all the background runner queues + for background in background_runners: + outputs.update(background.get_output()) + result.extend(background.get_result()) status = runner.join(outputs, result) base_runner.Runner.release(runner) return status diff --git a/yardstick/benchmark/runners/base.py b/yardstick/benchmark/runners/base.py index 13718d793..a887fa5b3 100755 --- a/yardstick/benchmark/runners/base.py +++ b/yardstick/benchmark/runners/base.py @@ -210,6 +210,10 @@ class Runner(object): QUEUE_JOIN_INTERVAL = 5 + def poll(self, timeout=QUEUE_JOIN_INTERVAL): + self.process.join(timeout) + return self.process.exitcode + def join(self, outputs, result, interval=QUEUE_JOIN_INTERVAL): while self.process.exitcode is None: # drain the queue while we are running otherwise we won't terminate |