Helm Chart Configuration Reference#

Installation Guide

Install Determined on Kubernetes

Chart.yaml Settings#

  • appVersion: Configures which version of Determined to install. Users can specify a release version (e.g., 0.13.0) or specify any commit hash from the upstream Determined repo (e.g., b13461ed06f2fad339e179af8028d4575db71a81). Users are encouraged to use a released version.


If using a non-release branch of the Determined repository, appVersion is going to be set to X.Y.Z.dev0. This is not an official release version and deploying this version result in a ImagePullBackOff error. Users should remove .dev0 to get the latest released version, or they can specify a specific commit hash instead.

values.yaml Settings#

  • masterPort: The port at which the Determined master listens for connections on. (Required)

  • useNodePortForMaster: When set to false (default), a LoadBalancer service is deployed to make the Determined master reachable from outside the cluster. When set to true, the master will instead be exposed behind a NodePort service. When using a NodePort service users will typically have to configure an Ingress to make the Determined master reachable from outside the cluster. NodePort service is recommended when configuring TLS termination in a load-balancer.

  • tlsSecret: enables TLS encryption for all communication with the Determined master (TLS termination is performed in the Determined master). This includes communication between the Determined master and the task containers it launches, but does not include communication between the task containers (distributed training). The specified Secret of type tls must already exist in the same namespace in which Determined is being installed.

  • db: Specifies the configuration of the database.

    • name: The database name to use. (Required)

    • user: The database user to use when logging in the database. (Required)

    • password: The password to use when logging in the database. (Required)

    • port: The database port to use. (Required)

    • hostAddress: Optional configuration to indicate the address of a user provisioned database If configured, the Determined helm chart will not deploy a database.

    • storageSize: Only used when hostAddress is left blank. Configures the size of the PersistentVolumeClaim for the Determined deployed database.

    • cpuRequest: The CPU requirements for the Determined database.

    • memRequest: The memory requirements for the Determined database.

    • cpuLimit: Optional configuration. The CPU limits for the Determined database.

    • memLimit: Optional configuration. The memory limits for the Determined database.

    • useNodePortForDB: Optional configuration that configures whether ClusterIP or NodePort service type is used for the Determined database. By default ClusterIP is used.

    • storageClassName: Optional configuration to indicate StorageClass that should be used by the PersistentVolumeClaim for the Determined deployed database. This can be left blank if a default storage class is specified in the cluster. If dynamic provisioning of PersistentVolumes is disabled, users must manually create a PersistentVolume that will match the PersistentVolumeClaim.

  • checkpointStorage: Specifies where model checkpoints will be stored. This can be overridden on a per-experiment basis in the Experiment Configuration Reference. A checkpoint contains the architecture and weights of the model being trained. Determined currently supports several kinds of checkpoint storage, gcs, s3, azure, shared_fs, and directory, identified by the type subfield.

    • type: gcs: Checkpoints are stored on Google Cloud Storage (GCS). Authentication is done using GCP’s “Application Default Credentials” approach. When using Determined inside Google Kubernetes Engine (GCE), the simplest approach is to ensure that the nodes used by Determined are running in a service account that has the “Storage Object Admin” role on the GCS bucket being used for checkpoints. As an alternative (or when running outside of GKE), you can add the appropriate service account credentials to your container (e.g., via a bind-mount), and then set the GOOGLE_APPLICATION_CREDENTIALS environment variable to the container path where the credentials are located. See Environment Variables for more information on how to set environment variables in trial environments.

      • bucket: The GCS bucket name to use.

      • prefix: The optional path prefix to use. Must not contain ... Note: Prefix is normalized, e.g., /pre/.//fix -> /pre/fix

    • type: s3: Checkpoints are stored in Amazon S3.

      • bucket: The S3 bucket name to use.

      • accessKey: The AWS access key to use.

      • secretKey: The AWS secret key to use.

      • prefix: The optional path prefix to use. Must not contain ... Note: Prefix is normalized, e.g., /pre/.//fix -> /pre/fix

      • endpointUrl: The optional endpoint to use for S3 clones, e.g.,

    • type: azure: Checkpoints are stored in Microsoft’s Azure Blob Storage. Authentication is performed by providing either a connection string, or an account URL and an optional credential.

      • container: The Azure Blob Storage container name to use.

      • connection_string: The connection string for the service account to use.

      • account_url: The account URL for the service account to use.

      • credential: The optional credential to use in conjunction with the account URL.

      Please only specify either connection_string or the account_url and credential pair.

    • type: shared_fs: Checkpoints are written to a hostPath Volume. Users are strongly discouraged from using shared_fs for storage beyond initial testing as most Kubernetes cluster nodes do not have a shared file system.

      • hostPath: The file system path on each node to use. This directory will be mounted to /determined_shared_fs inside the trial pod.

      • mountToServer: Allows users to download checkpoints.

        • false: Default.

        • true: Mounts the shared directory on the server to allow checkpoint.download() to work.

    • type: directory: Checkpoints are written to a local directory in the task container. Users are responsible for mounting a persistent storage at this path, e.g., a shared PVC using pod_spec configuration.

      • containerPath: The file system path inside the task pods to use. The checkpoints and TensorBoard data will be stored under this path.

    • When an experiment finishes, the system will optionally delete some checkpoints to reclaim space. The saveExperimentBest, saveTrialBest and saveTrialLatest parameters specify which checkpoints to save. See Checkpoint Garbage Collection for more details.

  • maxSlotsPerPod: Specifies number of GPUs there are per machine. Determined uses this information when scheduling multi-GPU tasks. Each multi-GPU (distributed training) task will be scheduled as a set of slotsPerTask / maxSlotsPerPod separate pods, with each pod assigned up to maxSlotsPerPod GPUs. Distributed tasks with sizes that are not divisible by maxSlotsPerPod are never scheduled. If you have a cluster of different size nodes, set the maxSlotsPerPod to greatest common divisor of all the sizes. For example, if you have some nodes with 4 GPUs and other nodes with 8 GPUs, set maxSlotsPerPod to 4 so that all distributed experiments will launch with 4 GPUs per pod (with two pods on 8-GPU nodes). (Required)

  • masterCpuRequest: The CPU requirements for the Determined master.

  • masterMemRequest: The memory requirements for the Determined master.

  • masterCpuLimit: Optional configuration. The CPU limits for the Determined master.

  • masterMemLimit: Optional configuration. The memory limits for the Determined master.

  • taskContainerDefaults: Specifies Docker defaults for all task containers. A task represents a single schedulable unit, such as a trial, command, or TensorBoard.

    • networkMode: The Docker network to use for the Determined task containers. If this is set to “host”, Docker host-mode networking will be used instead. Defaults to “bridge”.

    • dtrainNetworkInterface: The network interface to use during distributed training. If not set, Determined automatically determines the network interface. When training a model with multiple machines, the host network interface used by each machine must have the same interface name across machines. This is usually determined automatically, but there may be issues if there is an interface name common to all machines but it is not routable between machines. Determined already filters out common interfaces like lo and docker0, but agent machines may have others. If interface detection is not finding the appropriate interface, the dtrainNetworkInterface option can be used to set it explicitly (e.g., eth11).

    • forcePullImage: Defines the default policy for forcibly pulling images from the Docker registry and bypassing the Docker cache. If a pull policy is specified in the experiment config this default is overriden. Please note that as of November 1st, 2020 unauthenticated users will be capped at 100 pulls from Docker per 6 hours. Defaults to false.

    • cpuPodSpec: Sets the default pod spec for all non-GPU tasks. See Customize a Pod for details.

    • gpuPodSpec: Sets the default pod spec for all GPU tasks. See Customize a Pod for details.

    • checkpointGcPodSpec: Sets the custom pod spec for Checkpoint GC defaults. See Customize a Pod for details.

    • cpuImage: Sets the default Docker image for all non-GPU tasks. If a Docker image is specified in the experiment config this default is overriden. Defaults to: determinedai/pytorch-ngc:0.34.0.

    • startupHook: An optional inline script that will be executed as part of task set up.

    • gpuImage: Sets the default Docker image for all GPU tasks. If a Docker image is specified in the experiment config this default is overriden. Defaults to: determinedai/pytorch-ngc:0.34.0.

    • logPolicies: Sets log policies for trials. For details, visit log_policies.

          - pattern: ".*uncorrectable ECC error encountered.*"
              type: exclude_node
  • enterpriseEdition: Specifies whether to use Determined enterprise edition.

  • imagePullSecretName: Specifies the image pull secret for pulling the Determined master image. Required when using the enterprise edition.

  • telemetry: Specifies whether we collect anonymous information about the usage of Determined.

    • enabled: Whether collection is enabled. Defaults to true.

  • observability: Specifies whether Determined enables Prometheus monitoring routes. See Prometheus for details.

    • enable_prometheus: Whether Prometheus endpoints are present. Defaults to true.

  • tensorboardTimeout: Specifies the duration in seconds before idle TensorBoard instances are automatically terminated. A TensorBoard instance is considered to be idle if it does not receive any HTTP traffic. The default timeout is 300 (5 minutes).

  • initialUserPassword: Specifies a string containing the default password for the admin and determined user accounts. (Required)

  • defaultPassword: (Deprecated) Specifies a string containing the default password for the admin and determined user accounts. Use initialUserPassword instead.

  • logging: Configures where trial logs are stored. This section takes the same shape as the logging configuration in the cluster configuration, except that names are changed to camel case to match Helm conventions (e.g., skip_verify would be skipVerify here).

    • logging.security.tls.certificate: Contains the contents of an expected TLS certificate for the Elasticsearch cluster, rather than a path as it does in the cluster configuration. This can be conveniently set at the command line using helm install --set-file logging.security.tls.certificate=<path>.

  • defaultScheduler: Configures the default scheduler that Determined will use. Currently supports the coscheduler option, which enables the lightweight coscheduling plugin, and the preemption option, which enables a priority-based preemption scheduler. Unless specified as coscheduler, Determined will use the default Kubernetes scheduler.

  • resourcePools: This section contains the names of the resource pools and their linked namespaces. Maps to the resource_pools section from the master configuration.

  • additional_resource_managers: This section includes additional resource managers for launching jobs across multiple Kubernetes clusters. Maps to additional_resource_managers in the master configuration. An example configuration is provided in the values.yaml file.

    • resource_manager: Describes the configuration settings for the resource manager. Maps to resource_manager in the master configuration.

      • kubeconfig_secret_name: Specifies the name of the secret containing the kubeconfig for the resource manager. This kubeconfig is used to connect to the Kubernetes cluster and launch tasks. Note that some kubeconfigs may require additional adjustments or modifications. For example some kubeconfigs reference file paths, which may need to be bind-mounted into the container or have their data paths encoded into the kubeconfig. Other kubeconfigs, like those for GKE, may require installing plugins into the Determined master container and binding certain credential files. (Required)

      • kubeconfig_secret_value: The name of the secret that contains the resource manager’s kubeconfig. (Required)

    • resource_pools: The resource pool configuration. See resource_pools for available configuration options.

  • retentionPolicy: Specifies configuration settings for the retention of trial logs.

    • logRetentionDays: Specifies the number of days to retain logs for by default. Values should be between -1 and 32767. The default value is -1, retaining logs indefinitely. If set to 0, logs will be deleted during the next cleanup.

    • schedule: Specifies the schedule for cleaning up logs. Can be provided as a cron expression or a duration string.


To learn more about distributed training with Determined, visit the conceptual overview or the intro to implementing distributed training.