Configuring Starburst Enterprise in Kubernetes#

The starburst-enterprise Helm chart configures the Starburst Enterprise platform (SEP) coordinator and worker nodes and the internal cluster communication with the values.yaml file detailed in the following sections.

We strongly suggest that you follow best practices when customizing your cluster, creating small, targeted files to override any defaults, and adding any configuration properties. For guidance on creating override files, see our recommended customization file set.

Note

Many section in this page have links to examples of the relevant YAML in our YAML examples page. While you are configuring your cluster, it is is helpful to refer to SEP’s directory structure to ensure you are configuring any file locations correctly.

Throughout this page there are links to more descriptive documentation for a given configuration area. In many cases, these refer to the legacy configuration file names by name.

Configuring SEP with Helm charts#

SEP uses a number of configuration files internally that determine how it behaves:

  • etc/catalog/<catalog name>.properties

  • etc/config.properties

  • etc/jvm.properties

  • etc/log.properties

  • etc/node.properties

  • etc/access-control.properties

With our Kubernetes deployment, these files are built using Helm charts, and are nested in the coordinator.etcFiles YAML node.

Catalog properties files are also built using Helm charts. They are defined under the top-level catalogs: YAML node. The catalogs and properties you create depend on what data sources you connect to with SEP. You can configure as many as you need.

Every Helm-based SEP deployment includes a values.yaml file. The file contains the default values, any of which can be overridden.

Along with basic instance configuration for the various cloud platforms, values.yaml also includes the key-value pairs necessary to build the required *.properties files that configure SEP.

A recommended set of customization files is included later in this topic, including recommendations for creating specific override files, with examples.

Default YAML files#

Default values are provided for a minimum configuration only, not including security or any catalog connector properties, as these vary by customer needs. The configuration in the Helm chart also contains deployment information such as registry credentials and instance size.

Each new release of SEP includes new chart version, and the default values may change. For this reason, we highly recommend that you follow best practices and leave the values.yaml file in the chart untouched, overriding only very specific values as needed in one or more separate YAML files.

Note

Do not change the values.yaml, or copy it in its entirety and make changes to it and specify that as the override file, as new releases may change default values or render your values non-performant. Use separate files with changed values only, as described above.

Using version control#

We very strongly recommend that you manage your customizations in a version control system such as a git repository. Each cluster and deployment has to be managed in separate files.

Creating and using YAML files#

The following is a snippet of default values from the SEP values.yaml file embedded in the Helm chart:

coordinator:
  resources:
    memory: "60Gi" requests:
      cpu: 16

The default 60GB of required memory is potentially larger than any of the available pods in your cluster. As a result, the default prevents your deployment success, since no suitable pod is available.

To create a customization that overrides the default size for a test cluster, copy and paste only that section into a new file named sep-test-setup.yaml, and make any changes. You must also include the relevant structure above that section. The memory settings for workers have the same default values and need to be overridden as well:

coordinator:
  resources:
    memory: "10Gi" requests:
      cpu: 2
worker:
  resources:
    memory: "10Gi" requests:
      cpu: 2

Store the new file in a path accessible from the helm upgrade --install command.

When you are ready to install, specify the new file using the --values argument as in the following example. Replace 4XX.0.0 with the Helm chart version of the desired SEP release as documented on the versions page:

helm upgrade my-sep-test-cluster starburstdata/starburst-enterprise \
  --install \
  --version 4XX.0.0 \
  --values ./registry-access.yaml \
  --values ./sep-test-setup.yaml

You can chain as many override files as you need. If a value appears in multiple files, the value in the rightmost, last specified file takes precedence. Typically it is useful to limit the number of files as well as the size of the individual files. For example, it can be useful to create a separate file that contains all catalog definitions.

To view the built-in configuration of the Helm chart for a specific version of SEP, run the following command:

helm template starburstdata/starburst-enterprise --version 4XX.0.0

Use this command with different version values to compare the configuration of different SEP releases as part of your upgrade process.

To generate the specific configuration files for your deployment, use the template command with your additional values files:

helm template starburstdata/starburst-enterprise \
  --version 4XX.0.0 \
  --values ./registry-access.yaml \
  --values ./sep-test-setup.yaml

Note

The generated files are for review purposes only. To apply changes to your deployment, use the helm upgrade command with your files specified using the --values argument.

Applying configuration changes#

To update SEP with any configuration changes, run the helm upgrade command with the updated YAML files and the --install switch, as in this example:

$ helm upgrade my-sep-prod-cluster starburstdata/starburst-enterprise \
    --install \
    --version 454.0.0 \
    --values ./registry-access.yaml
    --values ./sep-prod-setup.yaml

You can use the same command as if you were updating to a new release. Helm will compare all --values files and the version, and safely ignore any that are unchanged.

Top level nodes#

The top level nodes are as follows, in order of appearance in the default values.yaml file embedded in the Helm chart. Click on a section header to access the relevant documentation:

Top level values.yaml nodes#

Node name

Description

Docker images section

image:

Contains the details for the SEP Docker image.

initImage:

Contains the details for the SEP bootstrap Docker image.

Docker registry access section

registryCredentials:

Defines authentication details for Docker registry access. Cannot be used if using imagePullSecrets:.

imagePullSecrets:

Alternative authentication for selected Docker registry using secrets. Cannot be used if using registryCredentials:.

Internal communications section

sharedSecret:

Sets the shared secret value for internal communications.

environment:

The environment name for the SEP cluster.

internal:

Specifies port numbers used for internal cluster communications.

Ports and networking

expose: node

Defines the mechanisms and their options that expose the SEP cluster to an outside network.

Coordinator and workers

The coordinator and worker nodes have many properties in common. They are detailed in the Coordinator section, and referenced in the Worker section, rather than repeating them. Likewise, examples are provided for the coordinator: top level node, and can be applied to the worker top level node by simply replacing coordinator with worker.

coordinator: node

Extensive section that configures the SEP coordinator pod.

worker: node

Extensive section that configures the SEP worker pods.

Startup options section

initFile: ""

Specifies a shell script to run after coordinator and work pod launch, and before SEP startup on the pods.

extraArguments: []

List of extra arguments to be passed to the initFile script.

extraSecret:

Lists secret to be used for initFile:.

Advanced security options

externalSecrets: node

Defines mounted external secrets from a Kubernetes secrets manager.

userDatabase: node

Configures the user database to use for file-based user authentication. This is disabled by default for versions 356-e and later.

securityContext: {} node

Sets the Kubernetes security context that defines privilege and access control settings for all SEP pods.

Pod lifecycle management section

readinessProbe:

Specifies the Kubernetes readiness probe that monitor containers for readiness to accept traffic.

livenessProbe:

Specifies the Kubernetes liveness probe that monitor containers for the need to restart.

SEP cache, disk and memory management

query: node

Specifies query processing and management configuration properties.

spilling: node

Configures spilling of query processing data to disk.

cache: node for hive configuration

Configures Hive storage caching.

Catalogs to configure connected data sources

catalogs: node

Configures the catalog properties files.

Miscellaneous

additionalVolumes: node

Defines mounted volumes for various uses in the cluster.

prometheus: node

Configures cluster metrics gathering with Prometheus.

commonLabels: {}

Defines common labels to identify objects in a KRM to use with the kustomize utility and other tools.

Docker images#

The default configuration automatically contains the details for the relevant Docker image on the Starburst Harbor instance. Typically you should not configure any overrides. An exception is if you host the Docker image on a different Docker registry.

Top level docker image nodes#

Node name

Description

image:

The image section controls the Docker image to use including the version. Typically, the Helm chart version reflects the SEP version. Specifically, the SEP version 356-e.x is reflected in the chart version of 356.x.0, or patches for the chart are released as 356.x.1, 356.x.2, and similar.

initImage:

The initImage section defines the container image and version to use for bootstrapping configuration files for SEP from the Helm chart and values files. This functionality is internal to the chart, and you should not change or override this configuration.

Defaults and examples:

Docker registry access#

These nodes enable access to the Docker registry. They should be added to the registry-access.yaml file to access the registry as shown in the best practices guide.

Top level docker image and registry credentials nodes#

Node name

Description

registryCredentials:

The registryCredentials section defines authentication details for Docker registry access. Typically, you need to use your username and password for the Starburst Harbor instance. NOTE: Cannot be be used if using imagePullSecrets:.

imagePullSecrets:

Alternative authentication for selected Docker registry using secrets. NOTE: Cannot be used if using registryCredentials:.

Warning

You can only define authentication with registryCredentials or imagePullSecrets. Defining both is invalid, and causes an error.

Defaults and examples:

Internal communication#

SEP provides three top level nodes for internal cluster communication between the coordinator and the workers. Because internal communication configurations and credentials are unique, these are not configured by default.

Ensure to configure sharedSecret and environment values in all your clusters to use secured communication with internal authentication of the nodes.

Top level internal communications nodes#

Node name

Description

environment

The environment name for the SEP cluster, used to set the node.environment node configuration property. Examples are values such as production, staging, batch_cluster, site_analytics, or any other short, meaningful string.

sharedSecret

Sets the shared secret value for secure communication between coordinator and workers in the cluster to a long, random string that you provide. If not set, the node.environment value from the etcFiles.node.properties configuration on the coordinator is used.

internalTls:

When set to true, enables TLS for internal communication and autogenerates trusted certificates. Requires both environment and sharedSecret top level nodes to be configured, and uses the port defined in internal.ports.https.port:. The environment value is used as the CN. Set to false by default.

Defaults and examples:

Exposing the cluster#

You must expose the cluster to allow users to connect to the SEP coordinator with tools such as the CLI, applications using JDBC/ODBC drivers and any other client application. This service-type configuration is defined by the expose top level node. You can choose from four different mechanisms by setting the type value to the common configurations in k8s.

Depending on your choice, you only have to configure the identically-named sections.

Types for expose: top level node#

Type

Description

clusterIp

Default value. Only exposes the coordinator internally within the k8s cluster using an IP address internal to the cluster. Use this in the early stages of configuration.

nodePort

Configures the internal port number of the coordinator for requests from outside the cluster on the nodePort port number. External service requests are made to <nodeIP>:<nodePort>. The clusterIP service is automatically created to supply all internal IP addresses.

loadBalancer

Used with platforms that provide a load balancer. This option automatically creates nodePort and clusterIP services , to which the external load balancer routes requests.

ingress

This option provides a production-level, securable configuration. It allows a load balancer to route to multiple apps in the cluster, and may provide load balancing, SSL termination, and name-based virtual hosting. For example, the SEP coordinator and Ranger server can be in the same cluster, and can be exposed via ingress: configuration.

Defaults and examples:

Coordinator#

The top level coordinator: node contains property nodes that configure the pod of the cluster that runs the SEP coordinator. The default values are suitable to get started with reasonable defaults on a production-sized k8s cluster.

Note

These property nodes also appear under the top level worker: node unless otherwise noted.

coordinator: nodes#

Node name

Description

etcFiles

Specifies the content of the jvm.config file and the *.properties files associated with the coordinator. These files are described in greater detail in a separate table below. NOTE: Not all files described in the detail table are applicable to workers. They are noted as such.

resources

The CPU and memory resources to use for the coordinator pod. These settings can be adjusted to match your workload needs, and available node sizes in the cluster.

nodeMemoryHeadroom

The size of the container memory headroom. The value needs to be less than resource allocation limit for memory defined in resources.

heapSizePercentage

Percentage of container memory reduced with headroom assigned to Java heap. Must be less than 100.

heapHeadroomPercentage

Headroom of Java heap memory not tracked by SEP during query execution. Must be less than 100.

additionalProperties

Any properties described in the reference documentation that are not specific to any other YAML node are set here. Example usages are to set query time-out and memory values and to enable Starburst Insights. Configuration properties can be found throughout the reference documentation, including the Properties reference page.

Warning

Avoid including secrets, such as credentials, in the additionalProperties section of the Helm chart. These values may be exposed in the init container logs. Instead, use the config.properties section to securely include such information.

envFrom: []

Allows for the propagation of environment variables from different sources complying with K8S schema can be used to deliver values to SEP configuration properties file by creating a Kubernetes secret holding variable values.

nodeSelector: {}, affinity: {} and tolerations: []

Configuration to determine the node and pod to use.

deploymentAnnotations: {}

Configuration to annotate the coordinator deployment.

podAnnotations: {}

Configuration to annotate the coordinator pod.

priorityClassName

Priority class for coordinator pod for setting k8s pod priority and preemption.

sidecars: []

Attach additional sidecar containers to the coordinator pod.

initContainers: []

Add extra init containers to the coordinator pod.

Defaults and examples:

etcFiles on the coordinator#

Note

These property nodes also appear under the top level worker: node unless otherwise noted.

etcFiles on the coordinator#

Node name

Description

coordinator.etcFiles.jvm.config

Defines the content of the JVM config for the coordinator.

coordinator.etcFiles.properties

Defines configuration files located in the etc folder. Each nested section defines the filename. Defaults are provided for the main configuration files etc/jvm.config, etc/config.properties, etc/node.properties and etc/log.properties. Additional properties files can be added by adding a nested section and the desired content of the configuration file.

coordinator.etcFiles.properties.config.properties

Defines the content of the default configuration file for the coordinator. You can also use additionalProperties for adding other configuration values.

coordinator.properties.node.properties

Defines the contents of the node.properties file for the coordinator, including paths to installation and log directories. The unique identifier and environment are pulled in from the environment, as documented in the internal communication section.

coordinator.etcFiles.properties.log.properties

Defines the contents of the log configuration files for the coordinator.

coordinator.etcFiles.properties.password-authentication.properties

Configures password authentication for SEP on the coordinator. NOTE: Does not apply to worker.etcFiles.properties.

coordinator.etcFiles.properties.access-control.properties

Enables and configures access control for SEP on the coordinator. NOTE: Does not apply to worker.etcFiles.properties.

coordinator.etcFiles.properties.exchange-manager.properties

Configures an optional exchange manager for Fault-tolerant execution.

coordinator.etcFiles.other

Other files that need to be placed in the etc directory (example).

Node assignment#

All charts allow you to configure criteria to define which node and pod in the cluster is suitable to use for running the relevant container. In SEP, this may be useful if you have a cluster spanning availability zones, or if you are using Ranger or an HMS in your cluster with smaller node sizes. The following configurations are available, and by default are not defined:

nodeSelector: {}
affinity: {}
tolerations: []

Example configurations are available in the k8s documentation. Specific usage and values are highly dependent on your k8s cluster configuration.

Further resources:

Coordinator defaults and examples#

Workers#

The worker section configures the pods of the cluster that run the SEP workers.

The top level worker: node contains property nodes that configure the pod of the cluster that runs the SEP workers. The default values are suitable to get started with reasonable defaults on a production-sized k8s cluster.

Many of the properties for this node also appear in the coordinator: top level node. They are fully documented in the Coordinator section:

  • worker.etcFiles.*

  • worker.resources

  • worker.nodeMemoryHeadroom

  • worker.heapSizePercentage

  • worker.heapHeadroomPercentage

  • worker.additionalProperties

  • worker.envFrom

  • worker.nodeSelector

  • worker.affinity

  • worker.tolerations

  • worker.deploymentAnnotations

  • worker.podAnnotations

  • worker.priorityClassName

  • worker.sidecars

  • worker.initContainers

The following properties apply only to worker:

worker:-only nodes#

Node name

Description

etcFiles

Specifies

worker.replicas

The number of worker pods for a static cluster.

worker.autoscaling

Configuration for the minimum and maximum number of workers. Ensure the additional requirements for scaling are fulfilled on your k8s cluster. Set enabled to true to activate autoscaling. The targetCPUUtilizationPercentage sets the threshold value that triggers scaling up by adding more workers until maxReplicas is reached.

Scaling down proceeds until minReplicas is reached and is controlled by deploymentTerminationGracePeriodSeconds and starburstWorkerShutdownGracePeriodSeconds.

WARNING: The autoscaling feature does not yet work with OpenShift clusters (as of the latest release 4.6), due to a known limitation with OpenShift clusters. HPA does not work with pods having init containers in OpenShift.

Read more information in our scaling section.

worker.kedaScaler

An alternative method of scaling the number of workers is implemented using the KEDA scaler for Kubernetes workloads in the SEP Helm chart. Scaler configuration is described in a dedicated section.

worker.deploymentTerminationGracePeriodSeconds

Specifies the termination grace period for workers. Workers are not terminated until queries running on the pod are finished and the grace period passes.

worker.starburstWorkerShutdownGracePeriodSeconds

Sets shutdown.grace-period to configure the grace period for worker process shutdown.

The following configuration properties for workers are identical to the coordinator properties, documented in preceding section.

Startup shell script for coordinator and workers nodes#

initFile: is a top level node that allows you to create a startup shell script (example) to customize how SEP is started on the coordinator and workers, and pass additional arguments to it. These are undefined by default.

Note

To execute commands in your initFile script that are not available in the main SEP Docker image, Starburst recommends using init containers for initialization and setup tasks. Init containers can use any Docker image. For information on configuring init containers within your Kubernetes deployment, see Init containers.

Startup script nodes#

Node name

Description

initFile

A shell script to run before SEP is launched. The content of the file has to be an inline string in the YAML file. The script is started as /bin/bash <<init_file>>. When called, it is passed the single parameter value coordinator or worker depending on the type of pod. The script needs to invoke the launcher script /usr/lib/starburst/bin/run-starburst for a successful start of SEP. NOTE: In the initFile, for chart versions released before middle of Oct 2020 use the launcher script exec /usr/lib/starburst/bin/launcher run. For newer releases use exec /usr/lib/starburst/bin/run-starburst to enable graceful shutdown of workers .

extraArguments

List of extra arguments to be passed to the initFile script.

User security considerations#

SEP has extensive security options that allow you specify how to authenticate users. Because user security configurations and credentials are unique, these are not configured by default.

Note

We strongly suggest that you watch our Securing Starburst Enterprise training video before customizing your SEP security configuration.

Security context#

You can configure a security context to define privilege and access control settings for the SEP pods.

securityContext:
  runAsNotRoot: true
  runAsUser: 1000
  runAsGroup: 0

Note

These settings are typically not required, and are highly dependent upon your Kubernetes environment. For example, OpenShift requires audit_log to be set in securityContext in order to run sudo, while other platforms use arbitrary user IDs. Refer to the Kubernetes documentation and to the documentation for your particular platform to ensure your securityContext is configured correctly.

Service account#

You can configure a service account for the SEP pods using:

serviceAccountName:

External secret reference#

There are several locations where properties require pointing to files delivered outside of SEP, such as CA certificates. In such cases, you can use a special notation that allows you to point to a k8s secret.

For example, you can configure password authentication using LDAP (example). This requires k8s to create the etc/password-authenticator.properties configuration file, which in turn points to the ca.crt certificate file.

Defining external secrets#

You can automatically mount external secrets, for example from the AWS Secrets Manager, using the secretRef or secretEnv notation.

externalSecrets nodes#

Node name

Description

externalSecrets.type

Type of the external secret provider. Currently, only eso is supported.

Note

The external secret provider needs to be configured with proper access to the specified backend, which is managed through the secretStoreRef property.

externalSecrets.secretPrefix

Prefix of all secrets that need to be mapped to external secret.

externalSecrets.eso.secretStoreRef

Reference to a SecretStore or ClusterSecretStore that you must create to configure the connection to your backend.

The Helm chart scans for all secretRef or secretEnv references in the referenced YAML files which start with the configured secretPrefix string. For each secret found, it generates an``ExternalSecret`` K8s manifest (example).

Note

The selected external secrets provider needs to be deployed and configured separately. The secret names in the external storage must match names of K8s secrets you reference. When using secretRef, the external storage secret must contain only a single value. For each external secret a single K8s secret is created, including one key with external secret value.

File-based authentication#

The unix command htpasswd can generate a user database, which can be used to configure file-based user authentication. It creates the file under /usr/lib/starburst/etc/auth/{{.Values.userDatabase.name}}. This allows you to statically deliver user credentials to the file etc/password-authenticator.properties (example).

Alternatively, you can use a secret to add an externally created user database file to SEP. Set the file.password-file property to the user database file, and ensure to disable the built-in user database (example).

Performance considerations#

SEP has performance management configuration options that help you to optimize aspects of SEP’s performance.

Note

We strongly suggest that you watch our cluster tuning training video before customizing your SEP security configuration.

Query memory usage control#

The top level query: node lets you to set the maxConcurrentQueries configuration property. The maxConcurrentQueries configuration property divides the query memory space by the value set. By setting this value, you establish a limit on the maximum memory usage for each query. If a query exceeds the allocated memory, it fails with an out-of-memory error.

This configuration property defaults to 3, which means that a single query can use up to one-third of the available memory space. To allocate more memory for concurrent queries, decrease the value of this property.

All other query processing properties must be set using the additionalProperties: node on the coordinator.

Spilling#

The top level query: node (example) allows you to configure SEP’s spilling properties.

Spilling uses internal node storage, which is mounted within the container.

Warning

Spilling is disabled by default, and we strongly suggest to leave it disabled. Enabling spill should be used as a method of last resort to allow for rare, memory-intensive queries to succeed on a smaller cluster at the expense of query performance and overall cluster performance.

Hive connector storage caching#

The cache: top level node (example) allows you to configure Hive connector storage caching. It is disabled by default.

Hive connector caching property nodes#

Node name

Description

cache.enabled

Enable or disable caching for all catalogs using the Hive connector. If you want to only enable it for a specific catalog, you have to configure it with the catalog configuration and additionalVolumes.

cache.diskUsagePercentage

Set the value for the hive.cache.disk-usage-percentage property.

cache.ttl

Set the value for the hive.cache.ttl property.

cache.volume

Configure the volume in which to store the cached files.

Catalogs#

The catalogs: top level node allows you to configure a catalog for each of your data sources. The catalogs defined in this node are used to create catalog properties files, which contain key-value pairs for catalog properties. Information for specific properties supported in each catalog can be found with the documentation for the connectors. At the very minimum, a catalog definition must consist of the name of the name of the catalog, and the connector.name property.

For best practices, use the YAML multi-line syntax shown in the examples to configure the content in multiple lines indented below the catalog name.

Defaults and examples:

Additional volumes#

Additional k8s volumes can be necessary for persisting files, for object storage caching, and for a number of other use cases. These can be defined in the additionalVolumes section (example). None are defined by default.

additionalVolume property nodes#

Node name

Description

path

Specifies the path to the mounted volume. If you specify path only, a directory named in path is created. When mounting ConfigMap or Secret, files are created in this directory for each key.

volume

A directory, with or without data, which is accessible to the containers in a pod.

hostpath

A volume with a hostPath defined mounts a file or directory from the host node’s filesystem into your pod.

hostpath.path

REQUIRED when using hostPath-mounted volumes.

subPath

When specified, a specific key named subPath from ConfigMap or Secret is mounted as a file with name provided by path. This is an alternative to using hostpath; you cannot define both for a volume.

Adding files#

Various use cases around security and event listeners need additional config files as properties or XML files. You can add any file to a pod using config maps.

Types of files you may need to add:

  • LDAP authentication file

  • Hive site xml file

Alternatively you can also use additionalVolumes to mount the files and copy the files to appropriate location using path and subPath parameter (example).

Prometheus#

This top level node configures the cluster to create Prometheus metrics. It is not to be confused with the connector of the same name. It is enabled by default.

We strongly suggest reviewing our example Prometheus rules.

SEP built-in directories#

The location of specific directories on the SEP container is important, if you configure additional files or otherwise want to tweak the container.

SEP is configured to use the following directories in the container:

  • /usr/lib/starburst: Top level folder for the SEP binaries.

  • /usr/lib/starburst/plugin: All plugins for SEP including connectors

  • /usr/lib/starburst/etc: etc folder for SEP configuration files such as config.properties and others.

  • /usr/lib/starburst/bin: Location of the run-starburst script, which is invoked with the container start.

  • /data/starburst/var/run: Contains launcher.pid file used by the launcher script.

  • /data/starburst/var/log: Contains secondary, rotated log files. Main log is redirected to stdout as recommended on containers.

Init containers#

Init containers are a Kubernetes feature designed for initialization and setup tasks within a pod. Starburst recommends using init containers as they provide alignment with Kubernetes best practices.

Using an init container within a pod allows you to prepare configuration and plugin files with a different Docker image than your main SEP containers. The init container’s Docker image may contain additional tools and utilities that are not available in the SEP image.

Images#

The SEP Docker image does not include command line tools such as curl, vi, nano, sed, awk, and grep. However, there are several lightweight images you can use for your init container that provide access to some of these tools. Here are some images to consider:

Using init containers with native cloud authentication capabilities may simplify your tasks. There are official images provided by AWS, GCP, or Azure that come with the relevant CLI tools pre-installed:

  • amazon/aws-cli

    • For more information, see AWS CLI.

  • google/cloud-sdk

  • mcr.microsoft.com/azure-cli

Shared volumes#

To make the files and data stored in your init container accessible to the main containers in your pod, you must mount shared volumes in both the init container and main containers. This ensures that your main containers can access your init container’s resources.

The following example demonstrates mounting a shared volume in all of your pod’s containers by defining an emptyDir volume in the additionalVolumes section of the Helm chart:

coordinator:
  initContainers:
  - name: prep-node-properties
    image: busybox:1.28
    command: ["sh"]
    args:
    - -c
    - |
      grep -v node.environment /etc/starburst/node.properties > /mnt/node.properties
      echo node.environment=$SEPENV >> /mnt/node.properties
additionalVolumes:
- path: /mnt
  volume:
    emptyDir: {}
initFile: |
  #!/bin/bash
  mv /mnt/node.properties2 /etc/starburst/node.properties
  exec /usr/lib/starburst/bin/run-starburst

Note

The volumeMounts entries in the initContainers and sidecars sections are automatically added for all additionalVolumes.

For more information on additional volumes, see additional volumes.

Moving files#

You can use init containers to move files within your pod. This is useful for transferring certificates or data from your init container to your pod’s main containers. The following example demonstrates how to use an init container to move a file within your pod:

coordinator:
  initContainers:
  - name: coordinator-pre-init
    image: registry.access.redhat.com/ubi9/ubi:latest
    command: ["sh"]
    args:
    - -c
    - |
      curl https://jdbc.postgresql.org/download/postgresql-42.2.16.jar --output postgresql.jar
      ls -l /downloads

worker:
  initContainers:
  - name: worker-pre-init
    image: registry.access.redhat.com/ubi9/ubi:latest
    command: ["sh"]
    args:
    - -c
    - |
      curl https://jdbc.postgresql.org/download/postgresql-42.2.16.jar --output postgresql.jar
      ls -l /downloads

additionalVolumes:
  - name: downloads
    path: /downloads
    volume:
      emptyDir: {}

Sidecars#

Extra sidecars and init containers can be specified to run together with coordinator or worker pods. You can use these containers for a number of use cases, such as to prepare an additional driver for the SEP runtime. The following example demonstrates declaring both sidecars and init containers:

coordinator:
  sidecars:
    - name: sidecar-1
      image: alpine:3
      command: [ "ping" ]
      args: [ "127.0.0.1" ]
  initContainers:
    - name: init-1
      image: alpine:3
      command: [ "bash", "-c", "printenv > /mnt/InContainer/printenv.txt" ]

worker:
  sidecars:
    - name: sidecar-3
      image: alpine:3
      command: [ "ping" ]
      args: [ "127.0.0.1" ]

additionalVolumes:
  - path: /mnt/InContainer
    volume:
      emptyDir: {}

KEDA scaler#

KEDA is a Kubernetes-based Event Driven Autoscaler. SEP can be configured with an external KEDA scaler to adjust the number of workers automatically based on JVM performance metrics available via JMX. Once it is enabled with the .Values.worker.kedaScaler.enabled property, the coordinator pod runs an additional container called keda-trino-scaler. This container works as a GRPC service, and communicates with KEDA to scale workers.

Note

The .Values.worker.autoscaling and .Values.worker.kedaScaler.enabled properties enable mutually-exclusive features that cannot be enabled together as KEDA also uses HPA to scale Kubernetes Deployments.

Prerequisites#

  • KEDA with Helm - versions 2.6.x and higher are required.

  • Prometheus JMX Exporter - The SEP Helm chart ensures that the exporter is enabled, and configured with all necessary rules.

Configuration#

The following nodes and node sections configure the KEDA scaler:

KEDA configuration nodes and sections for SEP#

Node or section name

Description

.Values.worker.kedaScaler

Set this node to true to enable the KEDA scaler.

scaledObjectSpec node structure

Corresponds to the parameters such as pollingInterval, minReplicaCount, maxReplicaCount, and others found in the KEDA ScaledObject spec.

scaledObjectSpec.triggers[0].metadata section

Defines the scaling method implemented by this scaler. Only the query_queue scaling method is available for SEP. It is enabled by default. The value for query_queue is the number of outstanding queries present on the cluster regardless of processing status.

numberOfQueriesPerWorker

Defines the maximum number of concurrent queries handled by a single worker. For example, if set to 10, and there are currently 34 queries submitted but not completed, KEDA scales out the workers to 4 replicas if maxReplicaCount >= 4.

scaleInToIdleReplicaCountIfNoQueuedQueriesLeft

Reduces the number of workers to idleReplicaCount when the query queue is empty. KEDA requires the value for idleReplicaCount to be less than the value of minReplicaCount. Can be set to 0.

Note

If you are using JMX metrics to monitor your deployment, you must explicitly list those metrics in the JMX metrics whitelist located in the prometheus.whitelistObjectNames section of your values.yaml file. This allows KEDA to access the required JMX metrics for scaling.

Troubleshooting#

When the KEDA scaler does not behave as expected, review the following logs:

  • keda-trino-scaler container logs in the coordinator pod

  • keda-operator logs in the keda namespace

Helm troubleshooting tips#

Here are some things to keep in mind as you implement SEP with Kubernetes and Helm in your organization.

  • Helm is space-sensitive and tabs are not valid. Our default values.yaml file uses 2-space indents. Ensure that your code editor preserves the correct indentation as it copies and pastes text.

  • Special characters such as square brackets [ ], curly braces { }, and non-alphanumeric characters such as $, &, and ! must be escaped or enclosed in quotes to prevent them being misinterpreted as part of YAML syntax. Escaping with a backslash indicates that a character should be treated as a literal character. Enclosing strings in single or double quotes ensures that the entire string is treated as plain text. For example:

    password: 'ExamplePa$$word!'
    
  • If your YAML files fail to parse, The Helm documentation provides several tools to debug this issue.

  • Sometimes problems can arise from improperly formatted YAML files. SEP uses Helm to build its *.properties files from YAML both with YAML key-value pairs and with multi-line strings. We recommend that you review the YAML techniques used in Helm to ensure that you are comfortable with using the Helm multi-line strings feature, and to understand the importance of consistent indentation.

  • Units can also be a source of confusion. We strongly suggest that you pay close attention to any units regarding memory and storage. In general, units provided for multiline strings that feed into SEP configuration files are in traditional metric bytes, such as megabytes (MB) and gigabytes (GB) as used by SEP. However, machine sizing and other values are in binary multiples such as mebibytes (Mi) and gibibytes (Gi), since these are used directly by Helm and Kubernetes.