Skip to content

Note: For more information about the type for each property refer to the corresponding CRD file.

Application Custom Resource (app.yaml) Reference🔗

This section provides a full example of the app.yaml custom resource along with a structured field reference table derived from the MLSysOpsApplication.yaml CRD.🔗

Example app.yaml🔗

MLSysOpsApp:
  name: test-application
  cluster_placement:
    cluster_id: ["uth-prod-cluster"]
  components:
    - metadata:
        name: server-app
        uid: a9jwduj9028uje
      node_placement:
        continuum_layer:
          - edge
        node: node-1 # Replace with proper hostname
        mobile: True
      runtime_class_name: nvidia
      sensors:
        - camera:
            model: d455
            camera_type: rgb
            minimum_framerate: 20
            resolution: 1024x768
      restart_policy: OnFailure
      node_type: virtualized
      os: ubuntu
      container_runtime: containerd
      containers:
        - image: harbor.nbfc.io/mlsysops/test-app:latest
          platform_requirements:
            cpu: 
              requests: "250m"
              limits: "500m"
              architecture:
                - amd64
              frequency: 1.4
              performance_indicator: 30 # BogoMIPS
            memory:
              requests: "64Mi"
              limits: "128Mi"
            disk: "120"
          image_pull_policy: IfNotPresent
          command: ["python", "TcpServer.py"]
          env:
            - name: OTEL_RESOURCE_ATTRIBUTES
              value: "service.name=server-app, service.version=0.0.0, service.experimentid=test"
            - name: OTEL_SERVICE_NAME
              value: "server-app"
            - name: NODE_IP
              value_from:
                field_ref:
                  field_path: status.hostIP
            - name: TELEMETRY_ENDPOINT # Add code for this to be done dynamically
              value: "$(NODE_IP):43170"
            - name: TCP_SERVER_IP 
              value: "0.0.0.0"
          ports:
            - container_port: 10000
              protocol: TCP
      qos_metrics:
        - application_metric_id: test_received_success_counter
          target: 20
          relation: lower_or_equal
          system_metrics_hints:
            - cpu_frequency
      host_network: False
    - metadata:
        name: client-app
        uid: jdaddwewed235uje
      node_placement:
        mobile: False
        continuum_layer:
          - edge
      sensors:
        - temperature:
            model: sdc30
      restart_policy: OnFailure
      node_type: native
      os: ubuntu # Just for demonstration purposes.
      container_runtime: containerd
      containers:
        - image: harbor.nbfc.io/mlsysops/test-app:latest
          platform_requirements:
            cpu: #change to cpu and merge with resources above.
              requests: "250m"
              limits: "500m"
              architecture:
                - arm64
              frequency: 1.4
            memory:
              requests: "64Mi"
              limits: "128Mi"
            disk: "100"
            gpu:
              model: k80
              memory: 2
              performance_indicator: 320 # BogoMIPS
          image_pull_policy: IfNotPresent
          command: ["python", "TcpClient.py"]
          env:
            - name: OTEL_RESOURCE_ATTRIBUTES
              value: "service.name=server-app, service.version=0.0.0, service.experimentid=test"
            - name: OTEL_SERVICE_NAME
              value: "server-app"
            - name: NODE_IP
              value_from:
                field_ref:
                  field_path: status.hostIP
            - name: TELEMETRY_ENDPOINT
              value: "$(NODE_IP):43170"
            - name: TCP_SERVER_IP
              value: "server-app"
      qos_metrics:
        - application_metric_id: test_sent_success_counter
          target: 30
          relation: equal  
  component_interactions:
    - component_name1: client-app
      type: egress
      component_name2: server-app
  global_satisfaction:
    threshold: 0.7
    relation: greater_than
    achievement_weights:
      - metric_id: test_received_success_counter
        weight: 0.5
      - metric_id: test_sent_success_counter
        weight: 0.5

Field Reference Table (Hierarchical View)🔗

Top-Level Fields🔗

Field Description Required Allowed Values
name The application name. Yes -
cluster_placement.cluster_id Array of clusters that can host the application. No -
components List of components of the application. Yes -
component_interactions Describes how components communicate. No -
global_satisfaction Global constraints for application satisfaction. No -

components[].metadata🔗

Field Description Required Allowed Values
name The unique name of the component. Yes -
uid Unique identifier (not user-defined). Yes -

components[].node_placement🔗

Field Description Required Allowed Values
continuum_layer Required component placement on the continuum. No cloud, far_edge, edge_infrastructure, edge, *
mobile Whether component is deployed on a mobile node. No True, False
labels Required labels for filtering. No -
node Required node name (optional). No -

components[].sensors[]🔗

Field Description Required Allowed Values
camera.model Camera sensor model. No d455, imx477, picamera-v2
camera.camera_type Type of camera sensor. No rgb, nir, thermal, monocular
camera.minimum_framerate Minimum framerate. No -
camera.resolution Camera resolution. No 1024x768, 4056x3040
temperature.model Temperature sensor model. No sdc30, ds18b20

components[].containers[]🔗

Field Description Required Allowed Values
image Name of the container image. Yes -
command Container startup command. No -
image_pull_policy Image pull policy. No Always, Never, IfNotPresent

components[].containers[].platform_requirements🔗

Field Description Required Allowed Values
cpu.requests CPU requests. No -
cpu.limits CPU limits. No -
cpu.architecture Supported architectures. No arm64, amd64
cpu.frequency Required CPU frequency in Hz. No -
cpu.performance_indicator CPU performance hint. No -
memory.requests Memory requests. No -
memory.limits Memory limits. No -
disk Required disk space in GB. No -
gpu.model GPU model. No k80, k40
gpu.memory GPU memory in GB. No -
gpu.performance_indicator GPU performance hint. No -

components[].containers[].ports[]🔗

Field Description Required Allowed Values
container_port Port exposed by the container. No (0, 65536)
protocol Protocol for the port. No UDP, TCP, SCTP

components[].containers[].env[]🔗

Field Description Required Allowed Values
name Env variable name. No -
value Env variable value. No -
value_from.field_ref.field_path Reference to Kubernetes field. No -

components[].qos_metrics[]🔗

Field Description Required Allowed Values
application_metric_id App metric id. No -
target Metric target value. No -
relation Desired relation (metric vs target). No lower_or_equal, greater_or_equal, equal, lower_than, greater_than

Other Component Fields🔗

Field Description Required Allowed Values
host_network Whether to use host network namespace. No True, False
runtime_class_name Runtime class to use. No nvidia, default, kata-fc, kata-dragon, urunc, crun, lunatic, nvidia-experimental, spin, wasmedge, slight
restart_policy Restart policy for the container. No Always, OnFailure, Never
os Operating system type. No ubuntu, kali, zephyr
node_type Type of the host node. No virtualized, native, bare_metal
container_runtime Container runtime. No containerd, docker, emb_serve

component_interactions[]🔗

Field Description Required Allowed Values
component_name1 Source component. No -
component_name2 Destination component. No -
type Type of interaction. No ingress, egress

global_satisfaction🔗

Field Description Required Allowed Values
threshold Minimum required satisfaction score. No [0.0, 1]
relation Satisfaction comparison. No greater_or_equal, equal, greater_than
achievement_weights[].metric_id Metric used for satisfaction. No -
achievement_weights[].weight Weight of each metric (total weight sum must be 1). No -

Continuum, Cluster, and Node Custom Resources Reference🔗

This documentation provides full example YAMLs and hierarchical field reference tables for the following custom resource definitions:

  • MLSysOpsContinuum
  • MLSysOpsCluster
  • MLSysOpsNode

Each section includes: - A sample YAML snippet. - A structured table of fields with descriptions, required/optional status, and allowed values (if defined).

MLSysOpsContinuum🔗

Example continuum.yaml🔗

MLSysOpsContinuum:
  name: demo-continuum
  continuum_id: demo-cont-id
  clusters:
    - uth-prod-cluster

Field Reference Table🔗

Field Description Required Allowed Values
name The continuum slice name. Yes -
continuum_id The unique continuum identifier. Yes -
clusters The set of registered clusters. Yes -
---

MLSysOpsCluster🔗

Example cluster.yaml🔗

MLSysOpsCluster:
  name: uth-prod-cluster
  cluster_id: uth-prod-cluster
  nodes:
    - node-1
    - node-2
    - node-3

Field Reference Table🔗

Field Description Required Allowed Values
name The cluster name. Yes -
cluster_id The unique continuum identifier. Yes -
nodes The set of registered nodes. Yes -
---

MLSysOpsNode🔗

Example node.yaml🔗

MLSysOpsNode:
  name: node-1
  labels:
    - gpu
    - edge-ready
  continuum_layer: edge
  cluster_id: uth-prod-cluster
  mobile: False
  location: [22.9576, 40.6401]  # [longitude, latitude] for stationary nodes
  sensors:
    - camera:
        model: d455
        camera_type: rgb
        framerate: 30
        supported_resolutions: ["1024x768"]
    - temperature:
        model: sdc30
  environment:
    node_type: virtualized
    os: ubuntu
    container_runtime: ["containerd"]
  hardware:
    cpu:
      model: Intel-i7
      architecture: amd64
      frequency: [2400000000, 3000000000]
      performance_indicator: 75 # BogoMIPS
    memory: 16
    disk: "256"
    gpu:
      model: k80
      memory: "4"
      performance_indicator: 100

Field Reference Table🔗

Field Description Required Allowed Values
name The name of the node. No -
labels The required labels for filtering. No -
continuum_layer Continuum placement level. Yes cloud, edge_infrastructure, edge, far_edge
cluster_id The unique cluster identifier that the node reports to. No -
mobile Specify if the node is mobile or stationary. No -
location Geolocation coordinates (lon, lat). Valid only for stationary nodes. For mobile ones, the respective information is collected using telemetry. No -
sensors[].camera.model The model name of the camera sensor. No imx415, imx219, d455, imx477, picamera-v2
sensors[].camera.camera_type The camera sensor type. No -
sensors[].camera.framerate Framerate. No -
sensors[].camera.supported_resolutions Supported camera resolutions. No 1024x768, 4056x3040
sensors[].temperature.model The model name of the temperature sensor. No sdc30, ds18b20
environment.node_type Node type. Yes virtualized, native, bare_metal
environment.os Operating system. Yes ubuntu, kali, zephyr
environment.container_runtime[] Supported runtimes. Yes containerd, docker, emb_serve
hardware.cpu.model CPU model name. No -
hardware.cpu.architecture CPU architecture. No amd64, arm64
hardware.cpu.frequency[] Possible CPU frequency values (Hz). No -
hardware.cpu.performance_indicator Quantifies processing capabilities (BogoMIPS). No -
hardware.memory Memory size (GB). No -
hardware.disk Disk space (GB). No -
hardware.gpu.model GPU model. No k80, k40
hardware.gpu.memory GPU memory size. No -
hardware.gpu.performance_indicator GPU performance score. No -
---