Hi,
I'm running on Dell ECS storage appliance, which offers S3 compatibility.
yes http://ECS.ai is the DNS name of the server.
ClearML-models is the bucket.
Let me try with ip:port.

Thanks SuccessfulKoala55 , how might I do this clean up? Does this increase with more use of ClearML? And to add, we save all artifacts onto a remote S3 server.

ok thanks. this would mean that increasing the disk space for my ClearML is the only option as we are not at liberty to delete.

Yes it is! But ClearML didn't support multi node training out of the box in a way that it streamline the process. So we are trying to figure out a way to do it.

Hi, by agent logs i suppose you meant the logs from the ClearML server console panel?

Hi erez, i think i would want to reference the code that transformed the data. Take for example, i received 10k images, i performed some transformation and save it as a next version before i split it up for my ML training. Some time later, i receive a new set of 10k images and wants to apply the same transformation and then append it to the previous 10k as another version. Clearml-data does well for the data-versioning part, but in terms of data provenance, its not clear how i can associate t...

I used nvcr pytorch image and instruct clearml to inherit global dependencies. No need to install torch and work well.

Hi.

We tried as advised above and it still didn't work.
Host: http://ecs.ai:443
output_uri = S3://ecs.ai:443/bucketname

This time round the client gave this error.
Botocore.exceptions.connectiinclosederror: connection was closed before we received a valid response from endpoint URL: ' http://ecs.ai/bucketname/.clearml.test '.

It's quite apparent that whatever clearml passed to boto3 ends up as a http call instead of https, which is wrong.

Thanks AgitatedDove14 , will take a look.

Yes! I definitely think this is important, and hopefully we will see something there 

 (or at least in the docs)

Hi AgitatedDove14 , any updates in the docs to demonstrate this yet?

Hi AgitatedDove14 , thanks.
In this case i am running k8s glue (machine glue), which will then spawn off pods in kubernetes worker (machine worker). So when you say direct access, are you refering to the Glue machine or K8S Worker machine?

Hi, how may i task.init() within these sub processes without write access to the 3rd party scripts and python executables?

After some churning, this is the answer. Change it in the clearml-agent init generated clearml.conf.

` default_docker: {
    # default docker image to use when running in docker mode
    image: "nvidia/cuda:10.1-runtime-ubuntu18.04"

    # optional arguments to pass to docker image
    # arguments: ["--ipc=host", ]
    arguments: ["--env GIT_SSL_NO_VERIFY=true",]
  } `

[root@2c7498711bef elasticsearch]# curl `
{
"cluster_name" : "clearml",
"status" : "red",
"timed_out" : false,
"number_of_nodes" : 1,
"number_of_data_nodes" : 1,
"active_primary_shards" : 4,
"active_shards" : 4,
"relocating_shards" : 0,
"initializing_shards" : 0,
"unassigned_shards" : 8,
"delayed_unassigned_shards" : 0,
"number_of_pending_tasks" : 0,
"number_of_in_flight_fetch" : 0,
"task_max_waiting_in_queue_millis" : 0,
"active_shards_percent_as_number" ...

Ok thanks, we'll try it out on next availability.

[root@2c7498711bef elasticsearch]# curl `
{
"index" : "events-training_stats_scalar-d1bd92a3b039400cbafc60a7a5b1e52b",
"shard" : 0,
"primary" : false,
"current_state" : "unassigned",
"unassigned_info" : {
"reason" : "CLUSTER_RECOVERED",
"at" : "2021-05-22T11:33:38.932Z",
"last_allocation_status" : "no_attempt"
},
"can_allocate" : "no",
"allocate_explanation" : "cannot allocate because allocation is not permitted to any of the nodes",
"node_allocation_decisi...

ah thanks. Hopefully the old ones get flushed out by Google soon.

Hi, Self-hosted using docker-compose.

The first stage is a rank0 pytorch script. The downstream stages are rankN scripts, they are waiting for the IP address of the first stage. But the first stage doesn’t return, it simply waits for the rankN scripts to connect to it. But in this case, the rankN scripts doesn’t start. So its probably necessary to have just a single stage.

If i were to start a single rank0, and subsequent rankN tasks, it would be rather messy on ClearML Dashboard. Best to have either a single clearml application...

Alright thanks, i will work on that.

i see. Can i take it that when the client uses
task.execute_remotely(queue_name="1gpu", exit_process=True)then none of the content in its clearml.conf will be used, except for the API part. And Clearml simply uses whatever is on the Agent side.
api { # Notice: 'host' is the api server (default port 8008), not the web server. api_server: web_server: files_server: # Credentials are generated using the webapp, `
# Override with os environment: ...

My assumption is that the agent will have pulled that off the client's clearml.conf.

Do you have more info on vault?
Actually it only make sense if the entire department or organisation are saving their models in a common repo. In our case this is not possible due to client security (e.g. training data from clients can potentially be 'reverse engineered' from trained models in future). So each department and even projects will need their own repo.

It would make sense on a very large resource cluster. Unfortunately we only have less than 50 GPUs to share across. A multi-tenant SAAS would cut the resources into even more smaller clusters and not help with efficiency. Or would you have a suggestion?

which clearml.conf is it refering to? I'm executing on my client, which is then remotely executed by the agent. Both of them has ~/clearml.conf.

yup. in this case it wasn't root. Removing that USER and -u in pip solves the problem. However, in our production images, we are required to remove root access.
` FROM nvidia/cuda:10.1-cudnn7-devel

ENV DEBIAN_FRONTEND noninteractive
RUN apt-get update && apt-get install -y
python3-opencv ca-certificates python3-dev git wget sudo ninja-build
RUN ln -sv /usr/bin/python3 /usr/bin/python

create a non-root user

ARG USER_ID=1000
RUN useradd -m --no-log-init --system --uid ${USER_ID} a...

AgitatedDove14 , would you elaborate on this resolution process?

Hi AgitatedDove14 , i changed everything to cuda 10.1 and tried again with the same rrror. the section as follows. I made sure torch==1.6.0+cu101 and torchvision==0.8.2+cu101 are in the pypi repo. But the same error still came up.
` # Python 3.6.9 (default, Oct 8 2020, 12:12:24) [GCC 8.4.0]
boto3 == 1.14.56
clearml == 0.17.4
numpy == 1.19.1
torch == 1.6.0
torchvision == 0.7.0

Detailed import analysis

**************************

IMPORT PACKAGE boto3

clearml.storage: 0

IMPORT PACKAG...

I can't seem to find the fix to this. Ended up using an image that comes with torch installed.

Hi AgitatedDove14 , what version i should change it to? I'm currently on v0.17.2rc3.