mirror of
https://github.com/newnius/YAO-agent.git
synced 2025-06-07 13:51:56 +00:00
update version based heart beats
This commit is contained in:
parent
b94e13d3e3
commit
6229798726
16
agent.py
16
agent.py
@ -39,7 +39,7 @@ def launch_tasks(stats):
|
|||||||
entries_to_remove = []
|
entries_to_remove = []
|
||||||
lock.acquire()
|
lock.acquire()
|
||||||
for task_id, task in pending_tasks.items():
|
for task_id, task in pending_tasks.items():
|
||||||
if int(utils[task['gpus'][0]]) < 85:
|
if int(utils[task['gpus'][0]]) < 70:
|
||||||
entries_to_remove.append(task_id)
|
entries_to_remove.append(task_id)
|
||||||
script = " ".join([
|
script = " ".join([
|
||||||
"docker exec",
|
"docker exec",
|
||||||
@ -307,12 +307,14 @@ def report_msg(stats):
|
|||||||
flag = True
|
flag = True
|
||||||
if abs(last_version['mem_available'] - post_fields['mem_available']) > 2.0:
|
if abs(last_version['mem_available'] - post_fields['mem_available']) > 2.0:
|
||||||
flag = True
|
flag = True
|
||||||
for stat in stats:
|
for i in range(len(stats)):
|
||||||
if abs(last_version['status']['memory_total'] - stat['memory_total']) > 0.0:
|
if abs(last_version['status'][i]['memory_total'] - post_fields['status'][i]['memory_total']) > 0.0:
|
||||||
flag = True
|
flag = True
|
||||||
if abs(last_version['status']['memory_free'] - stat['memory_free']) > 512.0:
|
if abs(last_version['status'][i]['memory_free'] - post_fields['status'][i]['memory_free']) > 512.0:
|
||||||
flag = True
|
flag = True
|
||||||
if abs(last_version['status']['utilization_gpu'] - stat['utilization_gpu']) > 15.0:
|
if abs(last_version['status'][i]['utilization_gpu'] - post_fields['status'][i]['utilization_gpu']) > 15.0:
|
||||||
|
flag = True
|
||||||
|
else:
|
||||||
flag = True
|
flag = True
|
||||||
|
|
||||||
if flag:
|
if flag:
|
||||||
@ -322,6 +324,10 @@ def report_msg(stats):
|
|||||||
post_fields['flag'] = ver
|
post_fields['flag'] = ver
|
||||||
data = json.dumps(post_fields)
|
data = json.dumps(post_fields)
|
||||||
|
|
||||||
|
if flag:
|
||||||
|
print(ver)
|
||||||
|
print(post_fields)
|
||||||
|
|
||||||
producer = KafkaProducer(bootstrap_servers=KafkaBrokers)
|
producer = KafkaProducer(bootstrap_servers=KafkaBrokers)
|
||||||
future = producer.send('yao', value=data.encode(), partition=0)
|
future = producer.send('yao', value=data.encode(), partition=0)
|
||||||
result = future.get(timeout=10)
|
result = future.get(timeout=10)
|
||||||
|
Loading…
Reference in New Issue
Block a user