Skip to content

WorkloadScaling.md

Latest commit

 

History

History
90 lines (70 loc) · 3.01 KB

WorkloadScaling.md

File metadata and controls

90 lines (70 loc) · 3.01 KB
Raw

Workload scaling (Kubernetes-first, swappable)

Ashlar can dynamically change container replica counts through a first-class port:

Port Role
IWorkloadScaler Scale / inspect replicas (provider-specific)
IWorkloadScalePolicy Pure queue→replicas decision
IWorkloadDemandSignal Demand metrics feed (default: zeros; replace with mesh elastic status)

Providers are selected by config/env — swap without changing call sites.

Providers

Ashlar:WorkloadScaling:Provider / ASHLAR_WORKLOAD_SCALER Implementation
null (default) NullWorkloadScaler — safe no-op
kubernetes / k8s KubernetesWorkloadScaler — kubectl scale deployment/...
compose / docker-compose ComposeWorkloadScaler — docker compose up --scale

Add another adapter (ECS, Nomad, …) by implementing IWorkloadScaler and extending the switch in AddAshlarWorkloadScaling.

Configuration

{
  "Ashlar": {
    "WorkloadScaling": {
      "Provider": "kubernetes",
      "Enabled": true,
      "Kubernetes": {
        "KubectlPath": "kubectl",
        "Namespace": "ashlar",
        "Workloads": {
          "mesh-worker": {
            "Deployment": "ashlar-mesh-worker",
            "MinReplicas": 1,
            "MaxReplicas": 20,
            "DisplayName": "Mesh Worker"
          }
        }
      },
      "Autoscale": {
        "Enabled": false,
        "IntervalSeconds": 30,
        "ScaleUpQueueDepth": 5,
        "ScaleDownQueueDepth": 0,
        "WorkloadIds": [ "mesh-worker" ]
      }
    }
  }
}

Env shortcuts:

  • ASHLAR_WORKLOAD_SCALER=kubernetes|compose|null
  • ASHLAR_WORKLOAD_SCALING_ENABLED=true|false
  • ASHLAR_WORKLOAD_AUTOSCALE=true — starts ElasticWorkloadAutoscaleService

Sample Deployment: deploy/k8s/ashlar-mesh-worker-deployment.yaml.

HTTP API

Method Path Purpose
GET /api/workloads/provider Active provider + availability
GET /api/workloads Configured workloads
GET /api/workloads/{id}/replicas Desired / current / ready
PUT /api/workloads/{id}/replicas { "desiredReplicas": 3, "reason": "..." }

How this relates to mesh “elastic” scheduling

MeshPhase5ElasticScheduling.md places work onto existing workers and exposes GET /api/mesh/elastic/status as an operator signal.

This module creates/destroys capacity (replicas). Wire them together by implementing IWorkloadDemandSignal that reads mesh elastic queue depth and registering it instead of NullWorkloadDemandSignal.

Manual scale (kubectl)

kubectl -n ashlar scale deployment/ashlar-mesh-worker --replicas=3

Or via Ashlar:

curl -X PUT http://localhost:8088/api/workloads/mesh-worker/replicas \
  -H 'content-type: application/json' \
  -d '{"desiredReplicas":3,"reason":"load spike"}'