Sometimes the port param works, sometimes it doesn't. I did not find any differences in the logs so far.

we want to use a session/a port per gpu on the machine but are behind a firewall. currently ports 10020-10025 are allowed, but even with the port param set it sometimes jumps out of this range. I cannot always reproduce it, but I suspect it somehow depends on other sessions running at the time

oh, bummer 😕 but thanks a lot for clarifying

Hi @<1523701070390366208:profile|CostlyOstrich36> , sure (a little anonymised). I took the output from the "console" tab because somehow wsl overwrites the lines in my terminal

clearml-session --project session_test2 --remote-ssh-port 10023 --queue my_queue --docker my_image:310.0
clearml-session - CLI for launching JupyterLab / VSCode on a remote machine
Verifying credentials
Use previous queue (resource) 'my_queue' [Y]/n? y

Interactive session config:
{
    "base_task_id": null,
  ...

Other frameworks I've used either offer their SDKs in multiple programming languages or hide their logic behind an REST API. From what I've seen in the ClearML python SDK, the logic really is in the SDK.

Just trying to understand how it works in ClearML 🙃

Hi @<1523701070390366208:profile|CostlyOstrich36> ,

thanks for your help. Sorry, but I'm not really sure what you're suggesting. To clarify: I want to upload the files from the java application. The upload to "the application servers" is already working, the data should then be forwarded to ClearML. For this forwarding I cannot use the python SDK.

This is planned to be an automated process so I cannot use the CLI either.

If it's really not possible, I will probably write a flask service...