yes, if I manually put any task in this queue, it starts without problems

yes I see this tast in queue, but they just ‘pending’

Okay thank you so much
But I think I solve problem with credentials by using clearml_agent v1.8.1rc2
But now I get an issue with local python modules 🫠
Even when I set

agent.skip_pip_venv_install = 1
agent.skip_python_env_install = /usr/bin/python

In worker logs I see:

Environment setup completed successfully
Starting Task Execution:

@<1689446563463565312:profile|SmallTurkey79> did you solved this issue with fatal: could not read Username ?

@<1709740168430227456:profile|HomelyBluewhale47> @<1523701070390366208:profile|CostlyOstrich36> any updates here?
Also face same problem, envs that I set in clearml.conf can’t be used when I run agent

In my script:

And in my script I using

python-dotenv

from dotenv import load_dotenv
load_dotenv()
login = os.getenv("login")
password = os.getenv("pass")

I use them to connect to DB

But as I said before it seems that agent got them:

And in agent log I can see:

agent.environment.login = ****
agent.environment.pass = ****

So I need to use EXPORT before spinning up clearml-agent daemon?

@<1523701070390366208:profile|CostlyOstrich36> any updates?
Is it possible to avoid using docker mode?

what if I can’t use the docker mode because the agent is already running inside docker

@<1523701070390366208:profile|CostlyOstrich36> any updates here?

Can you provide an information, how to do that?

no, without docker

@<1698868530394435584:profile|QuizzicalFlamingo74> did u find solution?)

@<1523701070390366208:profile|CostlyOstrich36>
And if I use docker mode, an agent will read env variables from clearml.conf file?

Thank you this works 🙌

But by the way, why agent didn’t read env variables from clearml.conf file ?

Sorry, I didn’t really understand about top level
What do you mean?

Get the same result
But then I will have an empty credentials inside S3BucketConfigurations , here

Upd: if I delete keys ‘key’ and ‘secret_key’ from top section I got

2024-05-27 16:29:51,597 - clearml.storage - ERROR - Failed creating storage object 

 Reason: Missing key and secret for S3 storage access (

)

@<1523701070390366208:profile|CostlyOstrich36> looks the same to me, I use it this example to setup my confign
Or is there any diff?

@<1523701070390366208:profile|CostlyOstrich36> mb any other ideas, how to fix? 🫠

Now I use same config, as in original question:

s3 {
                key: "key"
                secret: "sec_key"
    
                credentials: [
                    {
                        host: "host:port"
                        bucket: "bucket_name"
                        key: "key"
                        secret: "sec_key"
                        multipart: false
                        secure: false
                    }
                ]
            } 

And pass output...

I think the sdk is aware that this is not amazon, because when I specify the ‘region’ field then it ignores my host and uses the amazon host

I tried None :port/bucket
But result still the same 🥲

I tried:

1. output_url = "

"
2. output_url = "

"
3. output_url = "s3://"

And one more thing(seems like agent didn’t pull all necessary code)
I’m trying to run task export.py with remote agent. But in this script there are some imports from my other .py scripts:

import torch - this module should be installed with pip
from from generate_triton_config import generate_configs - this import from generate_triton_config.py 

And it also raise an error:

Traceback (most recent call last):
  File "/root/.clearml/venvs-builds/3.10/code/export.py", line 8,...

Thank you!
That works

CLEARML_AGENT_SKIP_PIP_VENV_INSTALL & CLEARML_AGENT_SKIP_PYTHON_ENV_INSTALL

But still have a problem with, agent didn’t pull all necessary code
Do you have any ideas why that happens?
In “Installed packages” section, I see:

# Local modules found - skipping:
# generate_triton_config == ../generate_triton_config.py

Mb this relate somehow?

It even failed to create correct enviroment
And when it tried to execute import torch
It raise large error:

ImportError: cannot import name '_get_sequence_nr' from 'torch._C._autograd' (unknown location) 

@<1523701070390366208:profile|CostlyOstrich36> yes, I use same clearml.conf file

sdk{
    environment{
                test_var1: "var1_value"
                test_var2: "var2_value"

from clearml import Task
from clearml.automation import TaskScheduler


# Create the scheduler and make it poll quickly for demo purpose
scheduler = TaskScheduler(
    sync_frequency_minutes=1,
    force_create_task_project='name',
    force_create_task_name='Scheduler'
)

# Get the task that we want to rerun
# task_to_schedule = Task.get_task(project_name='name', task_name='')
task_to_schedule = Task.get_task(task_id='8ee7b88505a...')

# Add the scheduler based on task above and overri...