spotfire-rservice

A Helm chart for Spotfire® Service for R

Homepage: https://github.com/spotfiresoftware/spotfire-cloud-deployment-kit ⧉

Source Code

https://github.com/spotfiresoftware/spotfire-cloud-deployment-kit ⧉

Requirements

Kubernetes: >=1.24.0-0

Repository	Name	Version
file://../spotfire-common	spotfire-common	5.0.0

Overview

This chart deploys the Spotfire® Service for R ⧉ service on a Kubernetes ⧉ cluster using the Helm ⧉ package manager.

The R service pod includes:

A Fluent Bit ⧉ sidecar container for log forwarding.
Service annotations for Prometheus ⧉ scrapers. The Prometheus server discovers the service endpoint using these specifications and scrapes metrics from the exporter.
Predefined configuration for horizontal pod autoscaling with KEDA ⧉ and Prometheus.

This chart is tested to work with Elasticsearch ⧉, Prometheus ⧉ and KEDA ⧉.

Prerequisites

A deployed Spotfire Server release using the Spotfire Server ⧉ chart.

Usage

Replace all placeholders (shown in angle brackets like <NAMESPACE>) with your actual values before running the commands.

Step 1: Create namespace

kubectl create namespace "<NAMESPACE>"

Step 2: Prepare Spotfire Server service addresses

The R service needs the Spotfire Server backend service name, and optionally the log-forwarder service name if you want application logs sent to the Spotfire log-forwarder.

If the Spotfire Server chart was installed in the same namespace with the release name <SPOTFIRE_SERVER_RELEASE>, typical values are:

nodemanagerConfig.serverBackendAddress="<SPOTFIRE_SERVER_RELEASE>-spotfire-server"
logging.logForwarderAddress="<SPOTFIRE_SERVER_RELEASE>-log-forwarder" when the Spotfire Server chart log-forwarder is enabled

Step 3: Deploy the Spotfire Service for R chart

Create a values.yaml file (for example, spotfire-rservice-values.yaml). See the Values table for details of the values.

# spotfire-rservice-values.yaml
acceptEUA: true

global:
  spotfire:
    image:
      registry: "<REGISTRY>"
      pullPolicy: Always

nodemanagerConfig:
  serverBackendAddress: "<SPOTFIRE_SERVER_SERVICE>"

logging:
  logForwarderAddress: "<LOG_FORWARDER_SERVICE>"

Install the chart:

helm install "<SPOTFIRE_RSERVICE_RELEASE>" . \
  --namespace="<NAMESPACE>" \
  --create-namespace \
  --values spotfire-rservice-values.yaml

Note:

Setting acceptEUA: true means you agree that your use of the Spotfire software is governed by the terms of the Cloud Software Group, Inc. End User Agreement ⧉.
Replace <REGISTRY> with your private registry address where the Spotfire container images are stored.
Replace <SPOTFIRE_SERVER_SERVICE> with the Spotfire Server service name that the R service should connect to.
Replace <LOG_FORWARDER_SERVICE> with the log-forwarder service name if you want to forward logs to the Spotfire log-forwarder. If you do not want to use a log-forwarder service, omit logging.logForwarderAddress.

Check pod status:

kubectl get pods --namespace "<NAMESPACE>" \
  -l "app.kubernetes.io/name=spotfire-rservice,app.kubernetes.io/instance=<SPOTFIRE_RSERVICE_RELEASE>"

Configuration

Custom configuration properties

To set Custom configuration properties ⧉, add the property names as keys under the configuration section in your values file.

Example:

configuration:
  # The maximum number of R engine sessions that are allowed to run concurrently in the R service.
  engine.session.max: "5"

  # The number of R engines preallocated and available for new sessions in the R service queue.
  engine.queue.size: "10"

Scaling

To change the number of replicas managed by Helm, update replicaCount in your values file and run helm upgrade.

Example:

replicaCount: 3

helm upgrade "<SPOTFIRE_RSERVICE_RELEASE>" . \
  --namespace="<NAMESPACE>" \
  --values spotfire-rservice-values.yaml

Autoscaling with KEDA

To use KEDA ⧉ for autoscaling, first install KEDA in the Kubernetes cluster. You must also install a Prometheus instance that scrapes metrics from the Spotfire pods.

Example: A values.yaml snippet configuration for enabling autoscaling with KEDA:

resources:
  limits:
    cpu: 5
kedaAutoscaling:
  enabled: true
  spotfireConfig:
    prometheusServerAddress: http://prometheus-server.monitor.svc.cluster.local
  threshold: 3
  minReplicas: 1
  maxReplicas: 3

The spotfire-rservice has the following autoscaling defaults:

metric: spotfire_service_queue_engines_inUse (serviceQueueEnginesInUse R service counter).
query: the sum of spotfire_service_queue_engines_inUse of the R service instances for the release name.

The counter serviceQueueEnginesInUse provides the total number of engines currently executing. By default, the R service has number of cores - 1 available slots, which means that kedaAutoscaling.threshold should be synchronized with resources.limits.cpu. Typically, you want to scale out before all the available capacity is taken. Therefore, the kedaAutoscaling.threshold should be lower than resources.limits.cpu. Note that clients requesting a slot typically wait until a slot is available.

For more information, see Monitoring Spotfire Service for R using JMX ⧉.

Note: You can tune nodemanagerConfig.preStopDrainingTimeoutSeconds and other timeouts (for example, engine.execution.timeout and engine.session.maxtime) so that long-running jobs are not aborted prematurely when an instance is stopped to scale in. See Engine Timeout ⧉ for more details.

For more advanced scenarios, see kedaAutoscaling.advanced ⧉ and kedaAutoscaling.fallback ⧉.

Additionally, you can define your own custom scaling triggers ⧉. Helm template functionality is available:

kedaAutoscaling:
  triggers: {} # list of triggers to activate scaling of the target resource

Note: For more details on the autoscaling defaults, see templates/keda-autoscaling.yaml in the chart.

Update the Pod Deletion Cost annotation automatically

The controller.kubernetes.io/pod-deletion-cost ⧉ pod annotation influences in which order Kubernetes selects the pod to delete, for example, during scale-in.

Note that pod annotations should not be updated 'too' often (minutes rather than seconds), depending on the size of the cluster. The reason for this is that the Kubernetes API server is highly optimized for reads. sleepIntervalSeconds controls how often the updater should run, thresholdPercent and minAbsDelta controls how large the change must be for the annotation to be updated, that is, it specifies the size of a meaningful change.

podDeletionCost:
  enabled: true

The spotfire-rservice has the following defaults:

cost formula: spotfire_service_queue_engines_inUse.
sleepIntervalSeconds: 120.
thresholdPercent: 10
minAbsDelta: 1

Performance and Storage

Improved performance and concurrency for temporary folder

To optimize data reuse, the R service uses its temporary folder (default: /tmp) to store temporary files and intermediate data. In scenarios where large data sets or concurrent computations are involved, the default temporary folder might become a bottleneck, impacting performance and throughput. It is recommended to use a more performant and larger Kubernetes volume.

Example: A values.yaml snippet for optimizing the R service temp disk performance.

extraVolumeMounts:

  - mountPath: /tmp
    name: rservice-temp-dir-volume
extraVolumes:
  - name: rservice-temp-dir-volume
    ephemeral: # See Kubernetes documentation for ephemeral volumes: https://kubernetes.io/docs/concepts/storage/ephemeral-volumes/
      volumeClaimTemplate:
        metadata:
          labels:
            type: rservice-ephemeral-volume
        spec:
          accessModes: ["ReadWriteOnce"]
          storageClassName: <STORAGE_CLASS_NAME> # Replace with your storage class name.
          resources:
            requests:
              storage: 10Gi # Specify the desired storage size.

Note: For Azure AKS clusters, see also: Use Azure Container Storage with local NVMe ⧉.

Uninstalling

To uninstall the <SPOTFIRE_RSERVICE_RELEASE> release:

helm --namespace "<NAMESPACE>" uninstall "<SPOTFIRE_RSERVICE_RELEASE>"

Upgrading

See helm upgrade ⧉ for command documentation.

Upgrading Spotfire Server and Spotfire services

When you upgrade to a newer Spotfire Server version and newer Spotfire services versions, upgrade the Spotfire Server first, and then upgrade the Spotfire services.

Upgrading helm chart version

Please review the release notes ⧉ for any changed, moved, or renamed parameters before upgrading the release.

Values

Key	Type	Default	Description
global.spotfire.acceptEUA	bool	`nil`	Accept the Cloud Software Group, Inc. End User Agreement ⧉ by setting the value to `true`.
global.spotfire.image.pullPolicy	string	`"IfNotPresent"`	The global container image pull policy.
global.spotfire.image.pullSecrets	list	`[]`	The global container image pull secrets.
global.spotfire.image.registry	string	`nil`	The global container image registry. Used for spotfire/ container images, unless it is overridden.
acceptEUA	bool	`nil`	Accept the Cloud Software Group, Inc. End User Agreement ⧉ by setting the value to `true`.
affinity	object	`{}`
configuration	object	`{}`	Add Custom configuration properties ⧉. Keys under configuration should be named the same as the configuration property, for example 'engine.execution.timeout'.
extraContainers	list	`[]`	Additional sidecar containers to add to the service pod.
extraEnvVars	list	`[]`	Additional environment variables.
extraEnvVarsCM	string	`""`	The name of the ConfigMap containing additional environment variables.
extraEnvVarsSecret	string	`""`	The name of the Secret containing extra additional environment variables.
extraInitContainers	list	`[]`	Additional init containers to add to the service pod.
extraVolumeMounts	list	`[]`	Extra volumeMounts for the service container. More info: `kubectl explain deployment.spec.template.spec.containers.volumeMounts`.
extraVolumes	list	`[]`	Extra volumes for the service container. More info: `kubectl explain deployment.spec.template.spec.volumes`.
fluentBitSidecar.image.pullPolicy	string	`"IfNotPresent"`	The image pull policy for the fluent-bit logging sidecar image.
fluentBitSidecar.image.repository	string	`"fluent/fluent-bit"`	The image repository for fluent-bit logging sidecar.
fluentBitSidecar.image.tag	string	`"4.2.3"`	The image tag to use for fluent-bit logging sidecar.
fluentBitSidecar.resources	object	`{}`	The resources setting for fluent-bit sidecar container.
fluentBitSidecar.securityContext	object	`{}`	The securityContext setting for fluent-bit sidecar container. Overrides any securityContext setting on the Pod level.
fullnameOverride	string	`""`
image.pullPolicy	string	`nil`	The spotfire-server image pull policy. Overrides global.spotfire.image.pullPolicy.
image.pullSecrets	list	`[]`	Image pull secrets.
image.registry	string	`nil`	The image registry for spotfire-server. Overrides the global.spotfire.image.registry value.
image.repository	string	`"spotfire/spotfire-rservice"`	The spotfire-server image repository.
image.tag	string	`"15.0.0-v7.0.0"`	The container image tag to use.
kedaAutoscaling	object	`{"advanced":{},"cooldownPeriod":300,"enabled":false,"fallback":{},"maxReplicas":4,"minReplicas":1,"pollingInterval":30,"spotfireConfig":{"prometheusServerAddress":"http://prometheus-server.monitor.svc.cluster.local"},"threshold":null,"triggers":[]}`	KEDA autoscaling configuration. See https://keda.sh/docs/latest/concepts/scaling-deployments/ for more details.
kedaAutoscaling.cooldownPeriod	int	`300`	The period to wait after the last trigger reported active before scaling the resource back to 0.
kedaAutoscaling.maxReplicas	int	`4`	This setting is passed to the HPA definition that KEDA creates for a given resource and holds the maximum number of replicas of the target resource.
kedaAutoscaling.minReplicas	int	`1`	The minimum number of replicas KEDA scales the resource down to.
kedaAutoscaling.pollingInterval	int	`30`	The interval to check each trigger on.
kedaAutoscaling.spotfireConfig	object	`{"prometheusServerAddress":"http://prometheus-server.monitor.svc.cluster.local"}`	Spotfire specific settings.
kedaAutoscaling.spotfireConfig.prometheusServerAddress	string	`"http://prometheus-server.monitor.svc.cluster.local"`	REQUIRED. The URL for the Prometheus server from where metrics are fetched.
livenessProbe.enabled	bool	`true`
livenessProbe.failureThreshold	int	`10`
livenessProbe.httpGet.path	string	`"/spotfire/liveness"`
livenessProbe.httpGet.port	string	`"registration"`
livenessProbe.initialDelaySeconds	int	`60`
livenessProbe.periodSeconds	int	`3`
logging.logForwarderAddress	string	`""`	The spotfire-server log-forwarder name. Template.
logging.logLevel	string	`"debug"`	Set to `debug`, `trace`, `minimal`, or leave empty for info. This applies for both node manager and the service.
nameOverride	string	`""`
nodeSelector	object	`{}`
nodemanagerConfig.preStopDrainingTimeoutSeconds	int	`610`	The draining timeout after which the service is forcefully shut down.
nodemanagerConfig.serverBackendAddress	string	`""`	The spotfire-server service name. This value is evaluated as a helm template.
podAnnotations."prometheus.io/path"	string	`"/spotfire/metrics"`
podAnnotations."prometheus.io/port"	string	`"9080"`
podAnnotations."prometheus.io/scrape"	string	`"true"`
podDeletionCost	object	Default values for Pod Deletion Cost, see values.yaml.	Pod Deletion Cost update configuration. See https://kubernetes.io/docs/concepts/workloads/controllers/replicaset/#pod-deletion-cost for more details.
podDeletionCost.costFormula	string	`"spotfire_service_queue_engines_inUse"`	An awk formula using the Prometheus metric names to calculate deletion cost. Missing or not found metrics default to 0.
podDeletionCost.enabled	bool	`false`	Enable updating of pod deletion cost annotation.
podDeletionCost.image.pullPolicy	string	`"IfNotPresent"`	Image pull policy for the podDeletionCost.
podDeletionCost.image.pullSecrets	list	`[]`	Image pull secrets for the podDeletionCost.
podDeletionCost.image.registry	string	`nil`	Image registry for the podDeletionCost.
podDeletionCost.image.repository	string	`"spotfire/spotfire-config"`	Image repository for the podDeletionCost.
podDeletionCost.image.tag	string	`"15.0.0-v7.0.0"`	Image tag for the podDeletionCost.
podDeletionCost.minAbsDelta	string	`"1"`	Minimum numeric change to trigger a patch.
podDeletionCost.replicaCount	int	`1`	Number of replicas.
podDeletionCost.resources	object	`{}`	Specifies the standard Kubernetes resource requests and/or limits
podDeletionCost.sleepIntervalSeconds	string	`"120"`	How long to wait between checks (seconds).
podDeletionCost.thresholdPercent	string	`"10"`	Minimum % change to trigger a patch.
podSecurityContext	object	`{}`	The Pod securityContext setting applies to all of the containers inside the Pod.
readinessProbe.enabled	bool	`false`
readinessProbe.failureThreshold	int	`10`
readinessProbe.httpGet.path	string	`"/spotfire/readiness"`
readinessProbe.httpGet.port	string	`"registration"`
readinessProbe.initialDelaySeconds	int	`60`
readinessProbe.periodSeconds	int	`3`
replicaCount	int	`1`
resources	object	`{}`
securityContext	object	`{}`	The securityContext setting for the service container. Overrides any securityContext setting on the Pod level.
service.port	int	`9501`
service.type	string	`"ClusterIP"`
serviceAccount.annotations	object	`{}`
serviceAccount.create	bool	`false`
serviceAccount.name	string	`""`
startupProbe.enabled	bool	`true`
startupProbe.failureThreshold	int	`20`
startupProbe.httpGet.path	string	`"/spotfire/started"`
startupProbe.httpGet.port	string	`"registration"`
startupProbe.initialDelaySeconds	int	`60`
startupProbe.periodSeconds	int	`3`
tolerations	list	`[]`
topologySpreadConstraints	list	`[]`
volumes.certificates.existingClaim	string	`""`	Defines an already-existing persistent volume claim.
volumes.certificates.subPath	string	`""`	The subPath of the volume to be used for the volume mount
volumes.packages.existingClaim	string	`""`	If 'persistentVolumeClaim.create' is 'false' (the default), then use this value to define an already existing persistent volume claim.
volumes.packages.persistentVolumeClaim.create	bool	`false`	If 'true', then a 'PersistentVolumeClaim' is created.
volumes.packages.persistentVolumeClaim.resources	object	`{"requests":{"storage":"1Gi"}}`	Specifies the standard Kubernetes resource requests and/or limits for the customExt volume claims.
volumes.packages.persistentVolumeClaim.storageClassName	string	`""`	Specifies the name of the 'StorageClass' to use for the customExt volume-claim.
volumes.packages.persistentVolumeClaim.volumeName	string	`nil`	Specifies the name of the persistent volume to use for the customExt volume-claim.
volumes.packages.subPath	string	`""`	The subPath of the volume to use for the volume mount.
volumes.troubleshooting.existingClaim	string	`""`	If 'persistentVolumeClaim.create' is 'false' (the default), then use this value to define an already existing persistent volume claim.
volumes.troubleshooting.persistentVolumeClaim.create	bool	`false`	If 'true', then a 'PersistentVolumeClaim' is created.
volumes.troubleshooting.persistentVolumeClaim.resources	object	`{"requests":{"storage":"2Gi"}}`	Specifies the standard Kubernetes resource requests and/or limits for the volumes.troubleshooting claims.
volumes.troubleshooting.persistentVolumeClaim.storageClassName	string	`""`	Specifies the name of the 'StorageClass' to use for the volumes.troubleshooting-claim.
volumes.troubleshooting.persistentVolumeClaim.volumeName	string	`nil`	Specifies the name of the persistent volume to use for the volumes.troubleshooting-claim.