MXNet is a deep learning platform that accelerates the transition from research prototyping to production deployment. It is built for full integration into Python that enables you to use it with its libraries and main packages.

$ helm repo add bitnami https://charts.bitnami.com/bitnami
$ helm install my-release bitnami/mxnet

This chart bootstraps a MXNet deployment on a Kubernetes cluster using the Helm package manager.

Bitnami charts can be used with Kubeapps for deployment and management of Helm Charts in clusters. This Helm chart has been tested on top of Bitnami Kubernetes Production Runtime (BKPR). Deploy BKPR to get automated TLS certificates, logging and monitoring for your applications.

• Kubernetes 1.12
• Helm 2.12 or Helm 3.0-beta3
• PV provisioner support in the underlying infrastructure
• ReadWriteMany volumes for deployment scaling

To install the chart with the release name my-release:

$ helm repo add bitnami https://charts.bitnami.com/bitnami
$ helm install my-release bitnami/mxnet

These commands deploy MXNet on the Kubernetes cluster in the default configuration. The Parameters section lists the parameters that can be configured.

Tip: List all releases using helm list

To uninstall/delete the my-release deployment:

$ helm delete my-release

The command removes all the Kubernetes components associated with the chart and deletes the release.

The following table lists the configurable parameters of the MinIO chart and their default values.

Specify each parameter using the --set key=value[,key=value] argument to helm install. For example,

$ helm install my-release \
  --set mode=distributed \
  --set serverCount=2 \
  --set workerCount=3 \
    bitnami/mxnet

The above command creates 6 pods for MXNet: one scheduler, two servers, and three workers.

Alternatively, a YAML file that specifies the values for the parameters can be provided while installing the chart. For example,

$ helm install my-release -f values.yaml bitnami/mxnet

Tip: You can use the default values.yaml

It is strongly recommended to use immutable tags in a production environment. This ensures your deployment does not change automatically if the same tag is updated with a different image.

Bitnami will release a new chart updating its containers if a new version of the main container, significant changes, or critical vulnerabilities exist.

This chart includes a values-production.yaml file where you can find some parameters oriented to production configuration in comparison to the regular values.yaml. You can use this file instead of the default one.

• Run MXNet in distributed mode:

- mode: standalone
  mode: distributed

• Number of server nodes that will execute your code:

- serverCount: 1
  serverCount: 2

• Number of worker nodes that will execute your code:

- workerCount: 1
  workerCount: 4

The MXNet chart supports three different ways to load your files. In order of priority, they are:

1. Existing config map
2. Files under the files directory
3. Cloning a git repository

This means that if you specify a config map with your files, it won't look for the files/ directory nor the git repository.

In order to use use an existing config map you can set the configMap=my-config-map parameter.

To load your files from the files/ directory you don't have to set any option. Just copy your files inside and don't specify a ConfigMap.

Finally, if you want to clone a git repository you can use the following parameters:

cloneFilesFromGit.enabled=true
cloneFilesFromGit.repository=https://github.com/my-user/my-repo
cloneFilesFromGit.revision=master

In case you want to add a file that includes sensitive information, pass a secret object using the existingSecret parameter. All the files in the secret will be mounted in the /secrets folder.

We will use the gluon example from the MXNet official repository. Launch it with the following values:

mode=distributed
cloneFilesFromGit.enabled=true
cloneFilesFromGit.repository=https://github.com/apache/incubator-mxnet.git
cloneFilesFromGit.revision=master
entrypoint.file=image_classification.py
entrypoint.args="--dataset cifar10 --model vgg11 --epochs 1 --kvstore dist_sync"
entrypoint.workDir=/app/example/gluon/

Check the logs of the worker node:

INFO:root:Starting new image-classification task:, Namespace(batch_norm=False, batch_size=32, builtin_profiler=0, data_dir='', dataset='cifar10', dtype='float32', epochs=1, gpus='', kvstore='dist_sync', log_interval=50, lr=0.1, lr_factor=0.1, lr_steps='30,60,90', mode=None, model='vgg11', momentum=0.9, num_workers=4, prefix='', profile=False, resume='', save_frequency=10, seed=123, start_epoch=0, use_pretrained=False, use_thumbnail=False, wd=0.0001)
INFO:root:downloaded http://data.mxnet.io/mxnet/data/cifar10.zip into data/cifar10.zip successfully
[10:05:40] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: data/cifar/train.rec, use 1 threads for decoding..
[10:05:45] src/io/iter_image_recordio_2.cc:172: ImageRecordIOParser2: data/cifar/test.rec, use 1 threads for decoding..

If you want to increase the verbosity, set the environment variable PS_VERBOSE=1 or PS_VERBOSE=2 using the commonEnvVars value.

mode=distributed
cloneFilesFromGit.enabled=true
cloneFilesFromGit.repository=https://github.com/apache/incubator-mxnet.git
cloneFilesFromGit.revision=master
entrypoint.file=image_classification.py
entrypoint.args="--dataset cifar10 --model vgg11 --epochs 1 --kvstore dist_sync"
entrypoint.workDir=/app/example/gluon/
commonExtraEnvVars[0].name=PS_VERBOSE
commonExtraEnvVars[0].value=1

You will now see log entries in the scheduler and server nodes.

[14:22:44] src/van.cc:290: Bind to role=scheduler, id=1, ip=10.32.0.11, port=9092, is_recovery=0
[14:22:53] src/van.cc:56: assign rank=9 to node role=worker, ip=10.32.0.17, port=55423, is_recovery=0
[14:22:53] src/van.cc:56: assign rank=11 to node role=worker, ip=10.32.0.16, port=60779, is_recovery=0
[14:22:53] src/van.cc:56: assign rank=13 to node role=worker, ip=10.32.0.15, port=39817, is_recovery=0
[14:22:53] src/van.cc:56: assign rank=15 to node role=worker, ip=10.32.0.14, port=48119, is_recovery=0
[14:22:53] src/van.cc:56: assign rank=8 to node role=server, ip=10.32.0.13, port=56713, is_recovery=0
[14:22:53] src/van.cc:56: assign rank=10 to node role=server, ip=10.32.0.12, port=57099, is_recovery=0
[14:22:53] src/van.cc:83: the scheduler is connected to 4 workers and 2 servers
[14:22:53] src/van.cc:183: Barrier count for 7 : 1
[14:22:53] src/van.cc:183: Barrier count for 7 : 2
[14:22:53] src/van.cc:183: Barrier count for 7 : 3
[14:22:53] src/van.cc:183: Barrier count for 7 : 4
...

If you have a need for additional containers to run within the same pod as MXNet (e.g. an additional metrics or logging exporter), you can do so via the sidecars config parameter. Simply define your container according to the Kubernetes container spec.

sidecars:
- name: your-image-name
  image: your-image
  imagePullPolicy: Always
  ports:
  - name: portname
   containerPort: 1234

Similarly, you can add extra init containers using the initContainers parameter.

initContainers:
- name: your-image-name
  image: your-image
  imagePullPolicy: Always
  ports:
  - name: portname
   containerPort: 1234

The Bitnami MXNet image can persist data. If enabled, the persisted path is /bitnami/mxnet by default.

The chart mounts a Persistent Volume at this location. The volume is created using dynamic volume provisioning.

As the image run as non-root by default, it is necessary to adjust the ownership of the persistent volume so that the container can write data into it.

By default, the chart is configured to use Kubernetes Security Context to automatically change the ownership of the volume. However, this feature does not work in all Kubernetes distributions. As an alternative, this chart supports using an initContainer to change the ownership of the volume before mounting it in the final destination.

You can enable this initContainer by setting volumePermissions.enabled to true.