by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Benjamin Räthlein
    @raethlein

    Hey @sim-san , if you use the Docker-local version you can either create an nfs-volume before (see https://unix.stackexchange.com/a/320717/239011) and add the volume mount to the JupyterHub config (e.g. c.DockerSpawner.volumes={'my-nfs-share': '/workspace/nfs'} or it should also be possible to add something like the following configuration to the JupyterHub config (I did not test it, so it's only meant to be a hint):

    c.DockerSpawner.extra_host_kwargs={
      mounts=[{
        "target": "<container_path>",
        "source": "nfsvolume",
        "type": "volume",
        "driver_config": {"name": "local", options: {"type": "nfs", "device": ":<nfs export path>", "o": "addr=<nfs-server-addr>"}}
      }]
    }

    In the first solution, to avoid having the same nfs folder mounted in all directories, you could create a separate nfs volume (my-nfs-share-{username}) for each workspace and add the directive c.DockerSpawner.volumes={'my-nfs-share-{username}': '/workspace/nfs'}. However, this requires manual effort (as also the host directory has to be created). An automated way of handling nfs shares does not seem to be supported right now by DockerSpawner and also not by our MLHubDockerSpawner.

    In a Kubernetes-setup this might be a little bit easier as you can work with dynamic PVC provisioning etc.

    neer201
    @neer201
    Hey guys! I tried to launch ml-workspace image on vast.ai. My instance has launched and ssh runs normally. But I dont have launched jupyter app with other software. Whet I launch jupyter manually, I dont have access to other applications like VScode with the message in console: 404 GET /tools/vscode (127.0.0.1) 15.69ms referer=http://localhost:8080/tree?
    Lukas Masuch
    @LukasMasuch
    @neer201 do you have more information on how you launched the workspace, e.g. the docker run command? When you run the workspace, jupyter should start automatically and will be accesible on the main port.
    Simon Sander
    @sim-san
    I have copied folders und files in the mounted workspace folder. I can see the file in the notebook overview, but can't access them from the jupyterlab. The folders are not shown. Do you know why this happens ?
    Lukas Masuch
    @LukasMasuch
    @sim-san we haven't seen this issue yet. A few questions: 1) Have you mounted a volume via NFS or is it a basic host path mount? 2) Can you open the file via Jupyter or via terminal (cat)? 3) You could check the permissions of the files and folders. Just type in ls -la within the Jupyter terminal in the /workspace folder. If you like you can also share the screenshot here.
    Simon Sander
    @sim-san
    @LukasMasuch
    1) I copy a folder with .ipynb inside the host path mount.
    2) I can open the files via Jupyter and with terminal in jupyter. But inside the Jupyter Lab and Jupyter Lab terminal I do not see the files.
    3) After I copied the files I chaned the ownership to root:root
    The screenshot from the terminal in the normal Jupyter:
    grafik.png
    The screenshot from the terminal in the Jupyter Lab:
    grafik.png
    Simon Sander
    @sim-san
    But I noticed something else strange. When I opened the named server 'notebooks', I got an error message: 'Server not running. Your server is not running. Would you like to start it? '
    grafik.png
    By clicking on "Launch Server" the general server "My Server" was started. I noticed that the link to the Jupyter Lab in the named server 'notebooks' points to the general server. In the general server I can create a folder in Jupyter and see it in Jupyter Lab from the named server 'notebooks'
    Simon Sander
    @sim-san
    And also a screenshot of the hub logs:
    grafik.png
    neer201
    @neer201
    @LukasMasuch Hey, I didn't run any commands. Sorry, I am a newbie in this area. My config looks like thi
    s. There is an option to run some command when the instance starts but I don't know which.
    image.png
    Lukas Masuch
    @LukasMasuch
    @neer201 I just looked into how to start the workspace with vast.ai, and its possible if you select the mltooling/ml-workspace:latest image, select the SSH option, follow the vast.ai guide to setup the SSH connection, connect to the server with the ssh command (the ssh command should have the port tunneling via -L on port 8080). Once you are connected via SSH, you have to start all the workspace tools via the following command WORKSPACE_BASE_URL="/" WORKSPACE_PORT=8080 VNC_PW="vncpassword" VNC_RESOLUTION="1600x900" VNC_COL_DEPTH="24" python /resources/docker-entrypoint.py
    Within a few seconds/minutes, you should be able to access the workspace from you browser via localhost:8080
    @sim-san Strange issue, seems like that the proxy is redirecting the request to the user workspace instead of the named server when accessing JupyterLab. Can you confirm that what you see within Jupyterlab is the files of your user workspace? Unfortunatly, I wasn't able to reproduce this issue with the newest version of ml-hub on docker. Are you using Kubernetes or Docker?
    Simon Sander
    @sim-san
    I use Docker
    Can you confirm that what you see within Jupyterlab is the files of your user workspace? => Yes
    Lukas Masuch
    @LukasMasuch
    @sim-san Just got an idea. Is your named server called notebooks?
    Lukas Masuch
    @LukasMasuch
    The proxy uses the path /user/<user-name>/notebooks for actually accessing notebook objects within the user workspace. Naming the named server notebooks is therefore problematic since it is conflicting with this path. In the next version we will add blacklist some names for named servers such as notebooks. As a workaround for now, just use any other name than tree, notebooks, tools and it should (hopefully) work fine.
    heayoungsuh
    @heayoungsuh
    @LukasMasuch Sorry I'm late.. The problem went away after ran the dockerfile with build.py. I lost my log files as I was testing out some other things.
    Simon Sander
    @sim-san
    @LukasMasuch thanks for your help. Renaming the workspace solved the problem easily.
    Besides with I have found another bug. I have opened an issue about it. ml-tooling/ml-workspace#32
    Simon Sander
    @sim-san
    And a further Issue I found: ml-tooling/ml-workspace#33
    Lukas Masuch
    @LukasMasuch
    @sim-san Thanks for submitting the issues :) I will try to fix them next week.
    Simon Sander
    @sim-san
    Where can I see logs from the juypter lab if it has not started ?
    I get this error in the web console: Laden fehlgeschlagen für das <script> mit der Quelle "http://192.168.1.150:8899/static/lab/vendors~main.4d6f61962d80393a514d.js".
    Simon Sander
    @sim-san
    Also in an other Browser with an other falvour I get the same error vendors~main.4d6f61962d80393a514d.js:1 Failed to load resource: net::ERR_INVALID_HTTP_RESPONSE
    Lukas Masuch
    @LukasMasuch
    @sim-san you can see all tool logs within this folder: /var/log/supervisor/ inside the workspace. JupyterLab logs are within the jupyter.log logfile. There is also a hidden supervisor UI on port 8059 (open via url path /tools/8059/), but the UI is not fully functional. If you click on the tool name, you will see the last logs of the tool. Calling this url path should give you all jupyter logs: WORKSPACE_BASE_URL/tools/8059/tail.html?processname=jupyter&limit=100000
    neer201
    @neer201
    @LukasMasuch thank you so much! =) It works.
    MangoCubes
    @MangoCubes
    Hello. I am currently trying out the workspace on my machine, and everything seems to be working great, except when I try to access it through reverse proxy. I opened the server on port 8080, and I have Let's Encrypt SSL on it to secure it. However, when I try to use tools like Terminal, the console says that "Firefox can’t establish a connection to the server at wss://domain.com/terminals/websocket/main". Some other tools display similar error message, with the aforementioned error message and error "WebSocket close with status code 1006" being included in all of them. I have a feeling that this isn't quite related to the software itself, and pardon me if it isn't, but I was unsuccessful in finding enough relevant information to fix this. What can I do to fix this, if this is related to the workspace itself?
    Victor Rodriguez-Fernandez
    @vrodriguezf
    Hi! Where is the Jupyter token logged when authenticating with the <generated> option?
    MangoCubes
    @MangoCubes
    Hello and thank you for the reply. I greatly appreciate that you are here to help. Unfortunately, I am not that familiar with this program and my knowledge about it is rather superficial. Can you explain to me what you mean by that? All I know is token authentication method, and logging in with it generates no log on the console.
    This is what it looks like when I log in: https://pastebin.com/rLnRp6Ai
    Victor Rodriguez-Fernandez
    @vrodriguezf
    I refer to this sentence in the README: "You can also use <generated> to let Jupyter generate a random token that is printed out on the container logs."
    MangoCubes
    @MangoCubes
    I just tried --env AUTHENTICATE_VIA_JUPYTER="<generated>", but I don't see any logs regarding the random generated tokens. What have I done wrong?
    MangoCubes
    @MangoCubes
    As in, there is no log about it as if I set the token manually. docker logs <containerid> doesn't reveal anything new either.
    Victor Rodriguez-Fernandez
    @vrodriguezf
    exactly. I understand that it is not logged if you set it manually, but it should be somewhere for the <generated> case
    MangoCubes
    @MangoCubes
    Where can I find them? I understand that the container log should be showing it, but it currently doesn't (https://pastebin.com/rDQtX5qW), and that you can do jupyter notebook list but I would need token to access the console of that container.
    Victor Rodriguez-Fernandez
    @vrodriguezf
    I have no idea of where they are logged. I will open an issue on Github
    MangoCubes
    @MangoCubes
    Hello. I found out that with Portainer, I can access the console without token, and was able to retrieve the token via jupyter notebook list. However, I am not sure as to how I can use this to solve the original issue I had where I could not use some of the tools at all, such as VNC, and instead gives the error message "WebSocket close with status code 1006" in the browser console.
    I have also tried clearing everything and installing NGINX instead, but this returns same result as what I had when I had Apache installed.
    Victor Rodriguez-Fernandez
    @vrodriguezf
    You are right. Just by executing docker exec with the command jupyter notebook list the problem is solved. Thanks!
    MangoCubes
    @MangoCubes
    Oh.
    N. Dorukhan Sergin
    @dorukhansergin
    Hi,
    I've been following the readme about remote development with vs-code. There's an extension recommended there and here's the link for that Remote - SSH. This extension doesn't show up in my VS Code tool?? What can I do about this?
    image.png
    krannnn
    @krannnn
    Hello, I'm having this issue when migrating to docker 19.03
    this is the commande I'm running
    docker run -d -p 5678:8080 --gpus all --name "ml-workspace" -v "${PWD}:/workspace" mltooling/ml-workspace-gpu:latest
    I keep getting this error that is related to IT renaming nvidia-container-runtime-hook to nvidia-container-toolkit .
    docker: Error response from daemon: exec: "nvidia-container-runtime-hook": executable file not found in $PATH.
    Any thoughts ?
    Christophe Lecointe
    @clec_gitlab
    Hello, I'm trying to define a working directory using config c.KubeSpawner.working_dir, but it seems to be hardcoded to /workspace somewhere. Any idea how I can change it ?
    Boris Kosmynin
    @kosmynin_gitlab
    Hi, is there a way to get a vpn client running inside the spawned container?