Hi! Is there any example/documentation to the new ClearML Reports feature?

edit: never mind. Apparently it is just an embedded Markdown editor... :man-facepalming:

CostlyOstrich36 it worked! thank you so much :grinning_face_with_star_eyes:

CostlyOstrich36 ok the issue is not the clearml-agent version, it is the conda environment I'm trying to run with. I usually run my agents from one conda environment and when I pass CLEARML_AGENT_SKIP_PIP_VENV_INSTALL=1, it just uses the conda interpreter and all is good. When I try to do this with a different conda env, this does not work and tries forcibly to create a new venv. any idea why?

Edit: if it is any help, the conda env that works is installed in the conda install dir, and the n...

@<1533257407419912192:profile|DilapidatedDucks58> I'm had the same issue and solved it by removing the server's CORS config (from any config file under /opt/clearml/config).

see this thread for details: None )

@<1523701070390366208:profile|CostlyOstrich36> @<1523701087100473344:profile|SuccessfulKoala55> any idea how to fix this?

UPDATE: The issue in clearml.conf
In the API settings, the server address used an alias which was not defined in the docker. Once that was replaced with the explicit IP address, everything worked as expected

well maybe, but that kinda ruins the whole point of auto logging. The scalars are already logged, it is just a matter of plotting so I assumed it could be done in the webapp

AnxiousSeal95 looking forward to it!

AgitatedDove14 I have suggested fix, should I open an issue on GitHub, or can I directly open a PR?
compression=compression or ZIP_DEFLATED if compression else ZIP_STORED

@<1523701087100473344:profile|SuccessfulKoala55> they're running in the background. Is there an easy way to tail the logs or something?

I have a preprocessing task that is CPU parallelized, but I wanted to create a ClearML task for each instance since it reports some parameters that I want to later aggregate. Running this quickly requires 10-30 agents listening on the same queue

Thanks SuccessfulKoala55 - I will take a look. It is worth adding this information somewhere in the documentation though 🤓

SuccessfulKoala55 - Ok, new question that might focus this down:

I want to be able to query the fileserver using HTTP URLs that have "Accept-Ranges" in the header. I exposed this in the apiserver.conf but this does not affect the file server, only the web-server. Any chance to add a fileserver.conf to enable this functionallity?

these are the mounts I add:
-v /home/some_username/workspace/:/root/workspace -v /software:/software -v /images:/images -v /data:/data -v /processedData:/processedData -v /disk1:/disk1 -v /disk2:/disk2 -v /disk3:/disk3 -v /disk4:/disk4 -v /disk5:/disk5 -v /disk6:/disk6 -v /disk8:/disk8

None of these seem problematic to me. The only issue I can think of is that /home is an external mount on the host machine (outside of docker). Should I mount it somewhere?

@<1523701205467926528:profile|AgitatedDove14> actually no. I was unaware that was needed. I will try that and let you know

CostlyOstrich36 any news? I currently resort to reporting these scalars manually in order to get the desired result (which makes auto logging redundant)

it is /home This is how we share user dirs across multiple servers. I think this might be a permissions issue though. will try changing the permissions on .gitconfig and let you know if that helps

OutrageousSheep60 passing None means using default compression. You need to pass compression=0

yes.

I checked the base docker image I use and noticed that a $HOME/.gitconfig file already exists. would that be an issue given that the .gitconfig of the current user is mounted to that path once the clearml task is run?

@<1523701087100473344:profile|SuccessfulKoala55> any update on how to solve this?

Ideally, I would've wanted to use the GUI to sent shutdown/restart signals to the worker itself similar to what I currently do with tasks. From I understand, this is not possible so I would settle for maybe just adding an easier way to find what command to run to kill a certain worker. Even a simple button in the workers/queues GUI to copy a worker's ID and Name would make things easier.

@<1523701205467926528:profile|AgitatedDove14> exactly what I was looking for. Thanks!

sure

It takes around 3-5mins to upload 100-500k plain text files. I just assumed that the added size includes the entire dataset including metadata, SHA2 and other stuff required for dataset functionallity

that's an interesting use case for services mode. I'll try it out. thanks!

I put it there since I tried using clearml fileserver URLs as input to foxglove studio (playback of remote .bag files recorded in ROS). Foxglove required these CORS settings, but we since then pivoted to a different solution so it is no longer needed.

UPDATE:
Tried to kill the services mode agent and start a new one instead and I get a similar error:

CLEARML_WORKER_NAME=pls_work clearml-agent daemon -d --services-mode --queue foo_bar

clearml_agent: ERROR: create.<locals>.Validator.__init__() got an unexpected keyword argument 'types'

CostlyOstrich36 from what I can tell, the user_id argument is new and is indeed missing from the enqueue call

makes sense, just wanted to make sure I'm not missing anything obvious. I'll probably just avoid the pipeline grouping view for now. Thanks anyway

AgitatedDove14 thanks for the tip. I tried it but it still compresses the data to zip format. I suspect it is since ZIP_STORED is a constant that equals 0:

This is problematic due to line 689 in dataset.py ( https://github.com/allegroai/clearml/blob/d17903d4e9f404593ffc1bdb7b4e710baae54662/clearml/datasets/dataset.py#L689 ):
compression=compression or ZIP_DEFLATED

Since ZIP_DEFLATED=8, passing compression=0 still causes data to be compressed

AgitatedDove14 Is there a way to change colors of embedded plot just like in the UI? Some default colors make it hard to view in dark mode. Also, the color of embedded plots is the same regardless of the color of the original plot