hello , i've got exactly same issue recently. first I would advice you to check if you can connect to ssh within the docker container (if it runs in docker) or from your host.

try
ssh -T git@gitlab.com
does it connect?

2. the config as you show will try to use https connection in the first place, you have to remove user/pass for it to use ssh as far as I understand

Funny, because when I faced the issue I thought it would be easier to use ssh and not user/pass (because gitgub uses 2FA and I was not sure (still not) how ClearML works with it).

What if you'd try manually change REPOSITORY prop in cloned task from ssh style address to https style?

like so:

@<1523701070390366208:profile|CostlyOstrich36> sure! Thanks for taking a look!)

@<1523701070390366208:profile|CostlyOstrich36> any update here, by chance?

CLEARML-AGENT version 1.8.1

When i started agent with clearml-agent daemon --queue train it was 3.10.12. but I've installed the clearml itself and started the agent with bash were venv was not activated.

no, if you open the task logs you can see that it's git@github.com :[domain]/[repo]

unless I'm missing something

that issue was resolved by upgrading virtualenv to the latest version

Hey, @<1523701087100473344:profile|SuccessfulKoala55> it works!) Thanks:)

Two more questions, if you will.

• For a regular worker I can specify python executable using agent.python_binary setting. How can I do in for a service worker? Should I tweak it inside the container as well?
• (kinda stupid question) Do you know if specifying agent.git_user & agent.git_pass using https works for github as it requires 2FA now?

And one more thing that relates to the q#1 - is it safe to use more resent base image for agent-services service in docker compose? I see the specified latest tag point to image which was updated like 3 years ago..

And here's when I've tried to test the connection within the agent's container

Yes, looks like the solution, I've been working on it for the past 5min
Will let you know:)

Hi! Self hosted.

Error I've got:

Thanks a lot @<1523701070390366208:profile|CostlyOstrich36> ! Will take a look in a few hours.

@<1523701087100473344:profile|SuccessfulKoala55> could you assist, please?

nvm, just saw it has to have a different signature for aws/non-aws addresses

probably you have to have ffmpeg libsm6 libxext6
packages installed in the docker container where you use opencv.

@<1523701087100473344:profile|SuccessfulKoala55> sure. Thanks for taking a look.

Hi @<1523701087100473344:profile|SuccessfulKoala55> , have you spot smth interesting here?
Thanks!

Hello @<1523701070390366208:profile|CostlyOstrich36>
Any thoughts on this maybe?

Hello @<1523701070390366208:profile|CostlyOstrich36>
Yes, I'm self hosting. Yes, there's an error (see the pic). It says ERR_BLOCKED_BY_ORB. But the address it sent request to is wrong. As I said it's double-encoded.

No errors in the containers logs at that moment.

In my cleaml.conf I do not specify default_output_uri nor did I provide it when I initialize a task. Might it be that this is the reason? Because I've overwritten the config file a few times, and don't remember how it looked in the past.

Hello @<1523701070390366208:profile|CostlyOstrich36>
Nothing you can help with here?
Thanks.

Hi, @<1523701070390366208:profile|CostlyOstrich36>
No, I did not.

The address looks like this:
https://<my_file_server>/my_project/task_id/image.jpg

When I visit https://<my_file_server> I got OK in response. The image.jpg itself is also available to view locally at /opt/clearml/my_project/task_id/image.jpg and inside the file server container at /mnt/fileserver/ my_project/task_id/image.jpg