Hi All, We Have Clearml-Server Running On A Kube Pod, And Then A Gpu Server Running The Clearml-Agent Which We Use To Queue Jobs. For Some Reason, Our Kube Pod Restarted (We'Re Looking Into Why), But In The Process Of This Happening All Jobs On The Worke

Answered

Hi all, we have clearml-server running on a kube pod, and then a GPU server running the clearml-agent which we use to queue jobs.

For some reason, our kube pod restarted (we're looking into why), but in the process of this happening all jobs on the worker were aborted ( Process terminated by user ). Obviously the first thing we need to fix is the kube issue, but is there any way of making the worker resilient to a network outage/dropped connection so that any currently running jobs are not automatically terminated? Maybe a timeout setting somewhere?

Edit: on closer inspection, the pod did not restart this morning, so something else is causing the processes to terminate

  				
Posted 
	one day ago

					More
				  		
  Report
		
					DepravedBee82
				
					0
					 × 1

Votes Newest

Answers 2

It seems like the worker lost network connectivity, and then aborted the jobs 😞

2024-11-21T06:56:01.958962+00:00 mrl-plswh100 systemd-networkd-wait-online[2279529]: Timeout occurred while waiting for network connectivity.
2024-11-21T06:56:01.976055+00:00 mrl-plswh100 apt-helper[2279520]: E: Sub-process /lib/systemd/systemd-networkd-wait-online returned an error code (1)
2024-11-21T06:57:15.810747+00:00 mrl-plswh100 clearml-agent[2304481]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:15.810789+00:00 mrl-plswh100 clearml-agent[2304481]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:15.810825+00:00 mrl-plswh100 clearml-agent[2304481]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:15.810876+00:00 mrl-plswh100 clearml-agent[2304481]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:15.874338+00:00 mrl-plswh100 clearml-agent[2304484]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:15.874350+00:00 mrl-plswh100 clearml-agent[2304484]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:15.874363+00:00 mrl-plswh100 clearml-agent[2304484]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:15.874375+00:00 mrl-plswh100 clearml-agent[2304484]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:15.903967+00:00 mrl-plswh100 clearml-agent[2304489]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:15.903979+00:00 mrl-plswh100 clearml-agent[2304489]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:15.903991+00:00 mrl-plswh100 clearml-agent[2304489]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:15.904002+00:00 mrl-plswh100 clearml-agent[2304489]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:16.101657+00:00 mrl-plswh100 clearml-agent[2304490]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:16.101692+00:00 mrl-plswh100 clearml-agent[2304490]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:16.101727+00:00 mrl-plswh100 clearml-agent[2304490]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:16.101761+00:00 mrl-plswh100 clearml-agent[2304490]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:23.867991+00:00 mrl-plswh100 clearml-agent[2305219]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:23.868026+00:00 mrl-plswh100 clearml-agent[2305219]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:23.868061+00:00 mrl-plswh100 clearml-agent[2305219]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:23.868097+00:00 mrl-plswh100 clearml-agent[2305219]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:24.117654+00:00 mrl-plswh100 clearml-agent[2305224]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:24.117688+00:00 mrl-plswh100 clearml-agent[2305224]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:24.117722+00:00 mrl-plswh100 clearml-agent[2305224]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:24.117756+00:00 mrl-plswh100 clearml-agent[2305224]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:24.313929+00:00 mrl-plswh100 clearml-agent[2305230]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:24.313964+00:00 mrl-plswh100 clearml-agent[2305230]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:24.313999+00:00 mrl-plswh100 clearml-agent[2305230]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:24.314043+00:00 mrl-plswh100 clearml-agent[2305230]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:25.319252+00:00 mrl-plswh100 clearml-agent[2305229]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:25.319286+00:00 mrl-plswh100 clearml-agent[2305229]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:25.319322+00:00 mrl-plswh100 clearml-agent[2305229]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:25.319364+00:00 mrl-plswh100 clearml-agent[2305229]: sdk.network.iteration.retry_backoff_factor_sec = 10
2024-11-21T06:57:26.322691+00:00 mrl-plswh100 clearml-agent[2305231]: sdk.network.metrics.file_upload_threads = 4
2024-11-21T06:57:26.322725+00:00 mrl-plswh100 clearml-agent[2305231]: sdk.network.metrics.file_upload_starvation_warning_sec = 120
2024-11-21T06:57:26.322759+00:00 mrl-plswh100 clearml-agent[2305231]: sdk.network.iteration.max_retries_on_server_error = 5
2024-11-21T06:57:26.322793+00:00 mrl-plswh100 clearml-agent[2305231]: sdk.network.iteration.retry_backoff_factor_sec = 10

  				
Posted 
	one day ago

					More
				  		
  Report
		
					DepravedBee82
				
					0
					 × 1

@<1724960464275771392:profile|DepravedBee82> the agent (and SDK) will wait for quite some time and retry even if a server is not available. The printout you've attached is not something the agent or SDK print out - is this something your code prints? In general, I can easily test that (and just did 🙂 ) by running an agent with a task and simply disconnecting the network cable - the agent will keep trying for a very long time before giving up (backoff times keep increasing, and the max retries for network connectivity is 254 by default).
By the way, the sdk.network.iteration.max_retries_on_server_error is not actually used by the clearml python package, only by the ClearML enterprise python package

  				
Posted 
	14 hours ago

					More
				  		
  Report
		
					SuccessfulKoala55
				
					0
					 × 1

Write your answer

12 Views

2 Answers

one day ago

12 hours ago