def main():
killer = GracefulKiller()
evalai = EvalAI_Interface(
AUTH_TOKEN=AUTH_TOKEN,
EVALAI_API_SERVER=EVALAI_API_SERVER,
QUEUE_NAME=QUEUE_NAME,
)
logger.info("Deploying Worker for {}".format(
evalai.get_challenge_by_queue_name()["title"]))
challenge = evalai.get_challenge_by_queue_name()
cluster_details = evalai.get_aws_eks_cluster_details(challenge.get("id"))
cluster_name = cluster_details.get("name")
cluster_endpoint = cluster_details.get("cluster_endpoint")
api_instance = get_api_client(cluster_name, cluster_endpoint, challenge,
evalai)
install_gpu_drivers(api_instance)
while True:
message = evalai.get_message_from_sqs_queue()
message_body = message.get("body")
if message_body:
submission_pk = message_body.get("submission_pk")
challenge_pk = message_body.get("challenge_pk")
phase_pk = message_body.get("phase_pk")
submission = evalai.get_submission_by_pk(submission_pk)
if submission:
api_instance = get_api_object(cluster_name, cluster_endpoint,
challenge, evalai)
core_v1_api_instance = get_core_v1_api_object(
cluster_name, cluster_endpoint, challenge, evalai)
if (submission.get("status") == "finished"
or submission.get("status") == "failed"
or submission.get("status") == "cancelled"):
# Fetch the last job name from the list as it is the latest running job
job_name = submission.get("job_name")[-1]
delete_job(api_instance, job_name)
message_receipt_handle = message.get("receipt_handle")
evalai.delete_message_from_sqs_queue(
message_receipt_handle)
elif submission.get("status") == "running":
job_name = submission.get("job_name")[-1]
update_failed_jobs_and_send_logs(
api_instance,
core_v1_api_instance,
evalai,
job_name,
submission_pk,
challenge_pk,
phase_pk,
)
else:
logger.info(
"Processing message body: {0}".format(message_body))
challenge_phase = evalai.get_challenge_phase_by_pk(
challenge_pk, phase_pk)
process_submission_callback(api_instance, message_body,
challenge_phase, evalai)
if killer.kill_now:
break
def main():
killer = GracefulKiller()
evalai = EvalAI_Interface(
AUTH_TOKEN=AUTH_TOKEN,
EVALAI_API_SERVER=EVALAI_API_SERVER,
QUEUE_NAME=QUEUE_NAME,
)
logger.info("Deploying Worker for {}".format(
evalai.get_challenge_by_queue_name()["title"]))
while True:
logger.info(
"Fetching new messages from the queue {}".format(QUEUE_NAME))
message = evalai.get_message_from_sqs_queue()
message_body = message.get("body")
if message_body:
submission_pk = message_body.get("submission_pk")
submission = evalai.get_submission_by_pk(submission_pk)
if submission:
if (submission.get("status") == "finished"
or submission.get("status") == "failed"):
# Fetch the last job name from the list as it is the latest running job
job_name = submission.get("job_name")[-1]
delete_job(batch_v1, job_name)
message_receipt_handle = message.get("receipt_handle")
evalai.delete_message_from_sqs_queue(
message_receipt_handle)
elif submission.get("status") == "running":
continue
else:
message_receipt_handle = message.get("receipt_handle")
logger.info(
"Processing message body: {0}".format(message_body))
process_submission_callback(message_body, evalai)
if killer.kill_now:
break
def main():
killer = GracefulKiller()
evalai = EvalAI_Interface(
AUTH_TOKEN=AUTH_TOKEN,
EVALAI_API_SERVER=EVALAI_API_SERVER,
QUEUE_NAME=QUEUE_NAME,
)
logger.info(
"Deploying Worker for {}".format(
evalai.get_challenge_by_queue_name()["title"]
)
)
challenge = evalai.get_challenge_by_queue_name()
cluster_details = evalai.get_aws_eks_cluster_details(challenge.get("id"))
cluster_name = cluster_details.get("name")
cluster_endpoint = cluster_details.get("cluster_endpoint")
api_instance_client = get_api_client(
cluster_name, cluster_endpoint, challenge, evalai
)
install_gpu_drivers(api_instance_client)
api_instance = get_api_object(
cluster_name, cluster_endpoint, challenge, evalai
)
core_v1_api_instance = get_core_v1_api_object(
cluster_name, cluster_endpoint, challenge, evalai
)
if challenge.get("is_static_dataset_code_upload"):
# Create and Mount Script Volume
script_config_map = create_script_config_map(script_config_map_name)
create_configmap(core_v1_api_instance, script_config_map)
submission_meta = {}
submission_meta["submission_time_limit"] = challenge.get(
"submission_time_limit"
)
while True:
time.sleep(2)
message = evalai.get_message_from_sqs_queue()
message_body = message.get("body")
if message_body:
if challenge.get(
"is_static_dataset_code_upload"
) and not message_body.get(
"is_static_dataset_code_upload_submission"
):
time.sleep(35)
continue
api_instance = get_api_object(
cluster_name, cluster_endpoint, challenge, evalai
)
core_v1_api_instance = get_core_v1_api_object(
cluster_name, cluster_endpoint, challenge, evalai
)
message_body["submission_meta"] = submission_meta
submission_pk = message_body.get("submission_pk")
challenge_pk = message_body.get("challenge_pk")
phase_pk = message_body.get("phase_pk")
submission = evalai.get_submission_by_pk(submission_pk)
if submission:
if (
submission.get("status") == "finished"
or submission.get("status") == "failed"
or submission.get("status") == "cancelled"
):
try:
# Fetch the last job name from the list as it is the latest running job
job_name = submission.get("job_name")[-1]
delete_job(api_instance, job_name)
message_receipt_handle = message.get("receipt_handle")
evalai.delete_message_from_sqs_queue(
message_receipt_handle
)
except Exception as e:
logger.exception(
"Failed to delete submission job: {}".format(e)
)
# Delete message from sqs queue to avoid re-triggering job delete
evalai.delete_message_from_sqs_queue(
message_receipt_handle
)
elif submission.get("status") == "running":
job_name = submission.get("job_name")[-1]
update_failed_jobs_and_send_logs(
api_instance,
core_v1_api_instance,
evalai,
job_name,
submission_pk,
challenge_pk,
phase_pk,
message,
)
else:
logger.info(
"Processing message body: {0}".format(message_body)
)
challenge_phase = evalai.get_challenge_phase_by_pk(
challenge_pk, phase_pk
)
process_submission_callback(
api_instance, message_body, challenge_phase, evalai
)
if killer.kill_now:
break