def test_task_not_lost():
driver_task_id = utils.submit_job(app_url=utils.SPARK_EXAMPLES,
app_args="1500", # Long enough to examine the Executor's task info
args=["--conf spark.cores.max=1",
"--class org.apache.spark.examples.SparkPi"])
# Wait until executor is running
sdk_tasks.check_running(SPARK_PI_FW_NAME, 1, timeout_seconds=600)
# Check Executor task ID - should end with 0, the first task.
# If it's > 0, that means the first task was lost.
assert sdk_tasks.get_task_ids(SPARK_PI_FW_NAME, '')[0].endswith('-0')
# Check job output
utils.check_job_output(driver_task_id, "Pi is roughly 3")
def test_task_not_lost():
driver_task_id = utils.submit_job(
app_url=utils.SPARK_EXAMPLES,
app_args="1500", # Long enough to examine the Executor's task info
args=[
"--conf", "spark.cores.max=1", "--class",
"org.apache.spark.examples.SparkPi"
])
# Wait until executor is running
utils.wait_for_executors_running(SPARK_PI_FW_NAME, 1)
# Check Executor task ID - should be 0, the first task.
# If it's > 0, that means the first task was lost.
executor_task = shakedown.get_service_tasks(SPARK_PI_FW_NAME)[0]
assert executor_task['id'] == "0"
# Check job output
utils.check_job_output(driver_task_id, "Pi is roughly 3")
def test_executor_gpus_allocated():
"""
Checks that the specified executor.gpus is allocated for each executor.
"""
num_executors = 2
executor_gpus = 1
driver_task_id = _submit_gpu_app(num_executors=num_executors,
executor_gpus=executor_gpus,
gpus_max=num_executors * executor_gpus)
# Wait until executors are running
spark_utils.wait_for_executors_running(GPU_PI_APP_NAME, num_executors)
# Check Executor gpus - should be 1.
for i in range(0, num_executors):
executor_task = shakedown.get_service_tasks(GPU_PI_APP_NAME)[i]
assert executor_task['resources']['gpus'] == 1.0
# Check job output
spark_utils.check_job_output(driver_task_id,
"Pi calculated with GPUs: 3.14")