Version: Next

Scenario Deployment configuration

Deployment of a scenario onto the Engine is managed by the Designer's extension called Deployment Manager. To enable a given Deployment Manager its jar package has to be placed in the Designer's classpath. Nussknacker is distributed with three default Deployment Managers (flinkStreaming, lite-k8s, lite-embedded). Their jars are located in the managers directory.

Deployment specific configuration is provided in the deploymentConfig section of the configuration file - check configuration areas to understand the structure of the configuration file.

Below is a snippet of scenario deployment configuration.

deploymentConfig {     
  type: "flinkStreaming"
  engineSetupName: "My Flink Cluster"
  
  # Deployment Manager's specific parameters
  restUrl: "http://localhost:8081"
}

Parameters:

type parameter determines the type of the Deployment Manager. Possible options are: flinkStreaming, lite-k8s, lite-embedded
engineSetupName parameter is optional. It specifies how the engine will be displayed in the GUI. If not specified, default name will be used instead (e.g. Flink for flinkStreaming Deployment Manager).

Kubernetes native Lite engine configuration

Please check high level Lite engine description before proceeding to configuration details.

Please note, that K8s Deployment Manager has to be run with properly configured K8s access. If you install the Designer in K8s cluster (e.g. via Helm chart) this comes out of the box. If you want to run the Designer outside the cluster, you have to configure .kube/config properly.

Except the servicePort configuration option, all remaining configuration options apply to both streaming and request-response processing modes.

The table below contains configuration options for the Lite engine. If you install Designer with Helm, you can use Helm values override mechanism to supply your own values for these options. As the the result of the Helm template rendering "classic" Nussknacker configuration file will be generated.

If you install Designer outside the K8s cluster then the required changes should be applied under the deploymentConfig key as any other Nussknacker non K8s configuration.

Parameter	Type	Default value	Description
mode	string		Processing mode: either streaming or request-response
dockerImageName	string	touk/nussknacker-lite-runtime-app	Runtime image (please note that it's not touk/nussknacker - which is designer image)
dockerImageTag	string	current nussknacker version
scalingConfig (Streaming processing mode)	{tasksPerReplica: int}	{ tasksPerReplica: 4 }	see below
scalingConfig (Request - Response processing mode)	{fixedReplicasCount: int}	{ fixedReplicasCount: 2 }	see below
configExecutionOverrides	config	{}	see below
k8sDeploymentConfig	config	{}	see below
nussknackerInstanceName	string	{?NUSSKNACKER_INSTANCE_NAME}	see below
logbackConfigPath	string	{}	see below
commonConfigMapForLogback	string	{}	see below
ingress	config	{enabled: false}	(Request-Response only) see below
servicePort	int	80	(Request-Response only) Port of service exposed
scenarioStateCaching.enabled	boolean	true	Enables scenario state caching in scenario list view
scenarioStateCaching.cacheTTL	duration	10 seconds	TimeToLeave for scenario state cache entries
scenarioStateIdleTimeout	duration	3 seconds	Idle timeout for fetching scenario state from K8s

Customizing K8s deployment resource definition

By default, each scenario is deployed as the following K8s deployment:

apiVersion: apps/v1
kind: Deployment
metadata:
  annotations:
    nussknacker.io/scenarioVersion: |-
      {
        "versionId" : 2,
        "processName" : "DetectLargeTransactions",
        "processId" : 7,
        "user" : "jdoe@sample.pl",
        "modelVersion" : 2
      }
  labels:
    nussknacker.io/nussknackerInstanceName: "helm-release-name"
    nussknacker.io/scenarioId: "7"
    nussknacker.io/scenarioName: detectlargetransactions-080df2c5a7
    nussknacker.io/scenarioVersion: "2"
spec:
  minReadySeconds: 10
  selector:
    matchLabels:
      nussknacker.io/scenarioId: "7"
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        nussknacker.io/scenarioId: "7"
        nussknacker.io/scenarioName: detectlargetransactions-080df2c5a7
        nussknacker.io/scenarioVersion: "2"
      name: scenario-7-detectlargetransactions
    spec:
      containers:
        - env:
            - name: SCENARIO_FILE
              value: /data/scenario.json
            - name: CONFIG_FILE
              value: /opt/nussknacker/conf/application.conf,/runtime-config/runtimeConfig.conf
            - name: DEPLOYMENT_CONFIG_FILE
              value: /data/deploymentConfig.conf
            - name: LOGBACK_FILE
              value: /data/logback.xml
            - name: POD_NAME
              valueFrom:
                fieldRef:
                  apiVersion: v1
                  fieldPath: metadata.name
          image: touk/nussknacker-lite-runtime-app:1.3.0 # filled with dockerImageName/dockerImageTag 
          livenessProbe:
            httpGet:
              path: /alive
              port: 8080
              scheme: HTTP
          name: runtime
          readinessProbe:
            failureThreshold: 60
            httpGet:
              path: /ready
              port: 8080
              scheme: HTTP
            periodSeconds: 1
          volumeMounts:
            - mountPath: /data
              name: configmap
      volumes:
        - configMap:
            defaultMode: 420
            name: scenario-7-detectlargetransactions-ad0834f298
          name: configmap

You can customize it adding e.g. own volumes, deployment strategy etc. with k8sDeploymentConfig settings, e.g. add additional custom label environment variable to the container, add custom sidecar container:

spec {
  metadata: {
    labels: {
      myCustomLabel: addMeToDeployment
    }
  }
  containers: [
    {
      #`runtime` is default container executing scenario 
      name: runtime
      env: [
        CUSTOM_VAR: CUSTOM_VALUE
      ]
    },
    {
      name: sidecar-log-collector
      image: sidecar-log-collector:latest
      command: ["command-to-upload", "/remote/path/of/flink-logs/"]
    }
  ]
}

This config will be merged into the final K8s deployment resource definition. Please note that you cannot override names or labels configured by Nussknacker.

Overriding configuration passed to runtime.

In most cases, the Model configuration values passed to the Lite Engine runtime are the ones from the modelConfig section of main configuration file. However, there are two exception to this rule:

there is application.conf file in the runtime image, which is used as additional source of certain defaults.
you can override the configuration coming from the main configuration file. The paragraph below describes how to use this mechanism.

In some circumstances you want to have different configuration values used by the Designer, and different used by the runtime. E.g. different accounts/credentials should be used in Designer (for schema discovery, tests from file) and in Runtime ( for the production use). For those cases you can use configExecutionOverrides setting:

deploymentConfig {
  configExecutionOverrides {
    special_password: "sfd2323afdf" # this will be used in the Runtime
  }
}
modelConfig {
  special_password: "aaqwmpor909232" # this will be used in the Designer
}

Configuring replicas count

Replicas count is configured under scalingConfig configuration key.

In the Request-Response processing mode you can affect the count of scenario pods (replicas) by setting fixedReplicasCount configuration key; its default value is 2:

{ fixedReplicasCount: x }.

In the Streaming processing mode the scenario parallelism is set in the scenario properties; it determines the minimal number of tasks used to process events. The count of replicas, scenario parallelism and number of tasks per replica are connected with a simple formula:

scenarioParallelism = replicasCount * tasksPerReplica

If you do not change any settings, the number of replicas in the K8s deployment will be set to the ceiling ( scenarioParallelism / tasksPerReplica); the default value of 4 will be used for tasksPerReplica. Alternatively, you can affect the number of replicas in the following ways:

modify the default value of tasksPerReplica by setting { tasksPerReplica: y }; the number of replicas will be computed as before as ceiling (scenarioParallelism / tasksPerReplica)
set fixedReplicasCount directly. The number of tasksPerReplica will be set to the ceiling (scenarioParallelism / fixedReplicaCounts. You cannot use this setting together with tasksPerReplica setting.

Due to rounding, the number of tasks may be different from scenario parallelism (e.g. for fixedReplicasCount = 3, scenario parallelism = 5, there will be 2 tasks per replica, total tasks = 6)

Nussknacker instance name

Value of nussknackerInstanceName will be passed to scenario runtime pods as a nussknacker.io/nussknackerInstanceName Kubernetes label. In a standard scenario, its value is taken from Nussknacker's pod app.kubernetes.io/instance label which, when installed using helm should be set to helm release name.

It can be used to identify scenario deployments and its resources bound to a specific Nussknacker helm release.

Configuring runtime logging

With logbackConfigPath you can provide path to your own logback config file, which will be used by runtime containers. This configuration is optional, if skipped default logging configuration will be used. Please mind, that apart whether you will provide your own logging configuration or use default, you can still modify it in runtime (for each scenario deployment separately*) as described here

*By default, every scenario runtime has its own separate configMap with logback configuration. By setting commonConfigMapForLogback you can enforce usage of single configMap (with such name as configured) with logback.xml for all your runtime containers. Take into account, that DeploymentManager relinquishes control over lifecycle of this ConfigMap (with one exception - it will create it, if not exist).

Configuring runtime ingress

In Request-Response processing mode additional ingress resource can be created when enabled flag is turned on. For now only nginx based K8s ingress controller is supported. It can be configured with following options.

Parameter	Type	Default value	Description
enabled	boolean	false	Either streaming or request-response
host	string		Name of the ingress host
rootPath	string	"/"	Root path for the ingress path, by default ingress path is rootPath + slug
config	config	{}	Additional ingress config customization

Configuring custom ingress class

By default, ingress resource will be created without any ingress class. If you want to use different class, you can set

ingress {
  enabled: true,
  config: {
    metadata: {
      annotations: {
        "kubernetes.io/ingress.class": "ingress-className"
      }
    }
  }
}

Configuring Prometheus metrics

Just like in Designer installation, you can attach JMX Exporter for Prometheus to your runtime pods. Pass PROMETHEUS_METRICS_PORT environment variable to enable agent, and simultaneously define port on which metrics will be exposed. By default, agent is configured to expose basic jvm metrics, but you can provide your own configuration file by setting PROMETHEUS_AGENT_CONFIG_FILE environment, which has to point to it.

Embedded Lite engine

Deployment Manager of type lite-embedded has the following configuration options:

Parameter	Type	Default value	Description
mode	string		Processing mode: either streaming-lite or request-response
http.interface	string	0.0.0.0	(Request-Response only) Interface on which REST API of scenarios will be exposed
http.port	int	8181	(Request-Response only) Port on which REST API of scenarios will be exposed
request-response.definitionMetadata.servers	string	[{"url": "./"}]	(Request-Response only) Configuration of exposed servers in scenario's OpenAPI definition. When not configured, will be used server with ./ relative url
request-response.definitionMetadata.servers[].url	string		(Request-Response only) Url of server in scenario's OpenAPI definition
request-response.definitionMetadata.servers[].description	string		(Request-Response only) (Optional) description of server in scenario's OpenAPI definition
request-response.security.basicAuth.user	string		(Request-Response only) (Optional) Basic auth user
request-response.security.basicAuth.password	string		(Request-Response only) (Optional) Basic auth password

Flink engine

Deployment Manager of type flinkStreaming has the following configuration options:

Parameter	Type	Default value	Description
restUrl	string		The only required parameter, REST API endpoint of the Flink cluster
jobManagerTimeout	duration	1 minute	Timeout for communication with FLink cluster. Consider extending if e.g. you have long savepoint times etc.
shouldVerifyBeforeDeploy	boolean	true	By default, before redeployment of scenario with state from savepoint, verification of savepoint compatibility is performed. There are some cases when it can be too time consuming or not possible. Use this flag to disable it.
shouldCheckAvailableSlots	boolean	true	When set to true, Nussknacker checks if there are free slots to run new job. This check should be disabled on Flink Kubernetes Native deployments, where Taskmanager is started on demand.
waitForDuringDeployFinish.enabled	boolean	true	When set to true, after Flink job execution, we check if tasks were successfully started on TaskMangers, before marking version as deployed. Otherwise version is marked as deployed immediately after successful response from JobManager.
waitForDuringDeployFinish.maxChecks	boolean	180	It works when `waitForDuringDeployFinish.enabled` option is set to `true`. This parameter describe how many times we should check if tasks were successfully started on TaskMangers before notifying about deployment failure.
waitForDuringDeployFinish.delay	boolean	1 second	It works when `waitForDuringDeployFinish.enabled` option is set to `true`. This parameter describe how long should be delay between checks.
scenarioStateCaching.enabled	boolean	true	Enables scenario state caching in scenario list view
scenarioStateCaching.cacheTTL	duration	10 seconds	TimeToLeave for scenario state cache entries
scenarioStateRequestTimeout	duration	3 seconds	Request timeout for fetching scenario state from Flink
jobConfigsCacheSize	int	1000	Maximum number of cached job configuration elements.

Scenario Deployment configuration

Kubernetes native Lite engine configuration​

Customizing K8s deployment resource definition​

Overriding configuration passed to runtime.​

Configuring replicas count​

Nussknacker instance name​

Configuring runtime logging​

Configuring runtime ingress​

Configuring custom ingress class​

Configuring Prometheus metrics​

Embedded Lite engine​

Flink engine​