building a custom docker image

mogul
mogul Registered Posts: 2 ✭✭✭

I want to run dataiku in docker, and the host has a few GPU's which I want to use. So, it seems I have to build my own image for docker.

This is my recipe: (to keep it simple, and make the build faster, I have omitted CUDA and R for now)

mkdir dataiku-build
cd dataiku-build

wget https://cdn.downloads.dataiku.com/public/dss/8.0.2/dataiku-dss-8.0.2.tar.gz
tar xzf dataiku-dss-8.0.2.tar.gz
sudo dataiku-dss-8.0.2/scripts/install/install-deps.sh
dataiku-dss-8.0.2/installer.sh -d DATA_DIR -p 11000

DATA_DIR/bin/dssadmin build-base-image --type container-exec --without-cuda --without-r

Let it cook for at few minutes, and then lets check:

$ docker image ls
REPOSITORY                               TAG                 IMAGE ID            CREATED             SIZE
dku-exec-base-sb59tjz4ll5axakfj8fhfv6g   dss-8.0.2           91329c9f35b1        11 seconds ago      1.39GB
centos                                   7                   8652b9f0cb4c        7 days ago          204MB
docker_web                               latest              fdfe96bedceb        11 months ago       6.27GB
jupyter/datascience-notebook             17aba6048f44        1bb52a350a5a        23 months ago       6.27GB

ah yes, good, there on top, 11 seconds old.

Now, lets fire it up to see whats happens:

$ docker run -p 10000:10000 dku-exec-base-sb59tjz4ll5axakfj8fhfv6g:dss-8.0.2
[2020-11-21 15:06:25,794] [1/MainThread] [INFO] [root] Fetching job definition
Traceback (most recent call last):
  File "/usr/lib64/python2.7/runpy.py", line 162, in _run_module_as_main
Installing debugging signal handler
    "__main__", fname, loader, pkg_name)
  File "/usr/lib64/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/opt/dataiku/python/dataiku/container/runner.py", line 483, in <module>
    execution_id = sys.argv[1]
IndexError: list index out of range

That obviously did not work.

Anyone else having success building and running own dataiku docker images?

Answers

  • Sergey
    Sergey Dataiker, Dataiku DSS Core Designer, Dataiku DSS & SQL, Dataiku DSS Core Concepts Posts: 365 Dataiker

    Hi @mogul

    Not sure I understand what you are trying to achieve with those steps.

    If you are looking to utilize GPUs in your ML tasks, you can just start DSS (as you have already installed it), create code environment and choose GPU during model training:

    https://doc.dataiku.com/dss/latest/machine-learning/deep-learning/runtime-gpu.html

    If you are interested in containerized execution with CUDA support, you will need to set up and configure the container engine (either Kubernetes or Docker) in DSS:

    https://doc.dataiku.com/dss/latest/containers/concepts.html

  • mogul
    mogul Registered Posts: 2 ✭✭✭
    edited July 17

    The latter. I want to run containerised (docker) with CUDA support.

    To do so I followed the guide over here: https://doc.dataiku.com/dss/latest/containers/custom-base-images.html

    Which results in a docker image that does not work.

    Inspecting with nvidia-smi on the docker host reveals two GPU's

    $ nvidia-smi
    Mon Nov 23 11:28:25 2020
    +-----------------------------------------------------------------------------+
    | NVIDIA-SMI 455.28       Driver Version: 455.28       CUDA Version: 11.1     |
    |-------------------------------+----------------------+----------------------+
    | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
    | Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
    |                               |                      |               MIG M. |
    |===============================+======================+======================|
    |   0  TITAN RTX           On   | 00000000:01:00.0 Off |                  N/A |
    | 41%   36C    P8    10W / 280W |     16MiB / 24217MiB |      0%      Default |
    |                               |                      |                  N/A |
    +-------------------------------+----------------------+----------------------+
    |   1  TITAN RTX           Off  | 00000000:21:00.0 Off |                  N/A |
    | 41%   34C    P8    13W / 280W |      5MiB / 24220MiB |      0%      Default |
    |                               |                      |                  N/A |
    +-------------------------------+----------------------+----------------------+
    
    +-----------------------------------------------------------------------------+
    | Processes:                                                                  |
    |  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
    |        ID   ID                                                   Usage      |
    |=============================================================================|
    |    0   N/A  N/A      1371      G   /usr/lib/xorg/Xorg                  9MiB |
    |    0   N/A  N/A      1595      G   /usr/bin/gnome-shell                4MiB |
    |    1   N/A  N/A      1371      G   /usr/lib/xorg/Xorg                  4MiB |
    |    1   N/A  N/A      1595      G   /usr/bin/gnome-shell                0MiB |
    +-----------------------------------------------------------------------------+

    doing the same inside the container running out dataiku instance:

    $ docker exec -it dataiku_dataiku_1 /bin/bash
    [dataiku@5ca65e340f6a ~]$ nvidia-smi
    bash: nvidia-smi: command not found

    Which was why I thought I had to build a new docker image.

    Ah yes, the non CUDA image we are currently running is the one I found on dockerhub: https://hub.docker.com/r/dataiku/dss/

Setup Info
    Tags
      Help me…