Resource Governance

This document describes the compute resource governance model of TFRS Operator, including ResourceQuota management, component resource defaults, and auto-fill mechanisms.

Governance Model Overview

ResourceQuota (Namespace-level)
├── TFRTenant Internal Components (12) ── User config > Operator defaults
└── TFRServer Containers (3)           ── User config > Operator defaults

Core logic: When a namespace has a ResourceQuota with CPU/Memory constraints, Kubernetes requires all containers to declare resources. The Operator uses a three-layer mechanism to prevent Pod rejection by the apiserver:

1. Component resource auto-fill: Components without explicit config use Operator defaults
2. LimitRange defaults: Provides fallback resources for third-party operator injected containers (e.g., MinIO sidecar)
3. Istio sidecar annotations: Declares resources for the istio-proxy container

ResourceQuota Management

Enabling

Controlled via the spec.resourceQuota field of the TFRTenant CR:

apiVersion: infra.tfrobot.io/v1
kind: TFRTenant
spec:
  resourceQuota:
    enabled: true
    spec:
      hard:
        requests.cpu: "8"
        requests.memory: "16Gi"
        limits.cpu: "16"
        limits.memory: "32Gi"
        pods: "50"

• CRD type definition: api/infra/v1/tfrtenant_types.go:316-329
• Deployable implementation: internal/controller/infra/tenant/resource_quota.go

Lifecycle

Operation	Behavior
enabled: true	Creates or updates the tfrs-quota-{tenantName} object
enabled: true + compute resources	Also creates tfrs-limit-{tenantName} LimitRange
enabled: false	Auto-deletes both ResourceQuota and LimitRange
TFRTenant deletion	Cascade-deleted via OwnerReference
Spec change	Spec drift detection auto-corrects (resource_quota.go:155-191)

LimitRange Auto-Creation

When ResourceQuota contains compute resources (CPU/Memory), the Operator automatically creates a LimitRange to provide default values for containers without resource declarations:

Type	Default (limits)	DefaultRequest (requests)
Container	50m / 64Mi	10m / 32Mi

Positioned as a "minimum safety net": only prevents containers without resource declarations from being rejected by Quota, not intended as reasonable defaults. Solves the problem of third-party operator injected containers (e.g., MinIO's sidecar and validate-arguments) lacking resource declarations and being rejected by ResourceQuota.

Implementation: internal/controller/infra/tenant/resource_quota.go

Istio Sidecar Resource Annotations

When a component has Resources set, the Operator automatically adds Istio sidecar resource annotations to the Pod:

Annotation	Value
sidecar.istio.io/proxyCPU	10m
sidecar.istio.io/proxyMemory	64Mi
sidecar.istio.io/proxyCPULimit	100m
sidecar.istio.io/proxyMemoryLimit	128Mi

Implementation: MergeIstioSidecarAnnotations() in internal/controller/infra/tenant/resources.go

Quota Detection

The Operator detects whether a namespace has compute resource quotas through two methods:

1. APIServer query: Checks all ResourceQuota objects in the namespace (resources.go:192-213)
2. TFRTenant Spec check: Handles first reconcile when the ResourceQuota object doesn't exist yet (resources.go:172-188)

Monitored fields: requests.cpu, requests.memory, limits.cpu, limits.memory — matching any one triggers auto-fill.

Entry point: tfrtenant_controller.go:276-277

TFRTenant Component Resources

Default Values

All defaults are defined in internal/controller/infra/tenant/resources.go:

Component	Requests (CPU/Mem)	Limits (CPU/Mem)
PostgreSQL	200m / 256Mi	1000m / 1Gi
MinIO	200m / 256Mi	1000m / 1Gi
RabbitMQ	200m / 256Mi	1000m / 1Gi
TimescaleDB	250m / 1Gi	500m / 2Gi
TFRSUtils Worker	250m / 256Mi	1000m / 2Gi
Prometheus	100m / 128Mi	500m / 512Mi
Redis Server	100m / 128Mi	500m / 512Mi
OtelCollector	100m / 128Mi	500m / 512Mi
Vector	100m / 128Mi	500m / 512Mi
Grafana	100m / 128Mi	500m / 512Mi
TFRSUtils Beat	100m / 128Mi	500m / 512Mi
Jaeger	50m / 128Mi	500m / 1Gi
TFRobotFront	50m / 64Mi	200m / 256Mi
Redis Insight	50m / 64Mi	200m / 256Mi

Total Requests for all components: ~1.85 CPU / 3Gi Memory

Total Limits for all components: ~8.4 CPU / 11.5Gi Memory

Resource Priority

User-specified in TFRTenant spec > Operator defaults

Implemented via EnsureResources() (resources.go:289-294): uses user config when non-nil, otherwise falls back to defaults.

User Configuration Example

apiVersion: infra.tfrobot.io/v1
kind: TFRTenant
spec:
  postgresql:
    resources:
      requests:
        cpu: "500m"
        memory: "512Mi"
      limits:
        cpu: "2000m"
        memory: "2Gi"
  redis:
    resources:
      requests:
        cpu: "200m"
        memory: "256Mi"

See component Options definitions for supported resource fields: internal/controller/infra/tenant/options.go:131-207

TFRServer Container Resources

Default Values

Defined in internal/controller/tfrobot/deployables/resources.go:22-46:

Container	Requests (CPU/Mem)	Limits (CPU/Mem)
API	250m / 256Mi	1000m / 1Gi
Worker (memory-worker)	500m / 512Mi	2000m / 2Gi
RobotWorker	500m / 512Mi	2000m / 2Gi

Total Requests per TFRServer: 1.25 CPU / 1.25Gi Memory

Total Limits per TFRServer: 5 CPU / 5Gi Memory

Auto-Fill Trigger Conditions

Triggered when both conditions are met (deployables/resources.go:48-60):

1. At least one container has nil resources
2. The namespace has a ResourceQuota with CPU/Memory constraints

Entry point: tfrserver_controller.go:286-295

User Configuration Example

apiVersion: tfrobot.tfrobot.io/v1
kind: TFRServer
spec:
  apiResources:
    requests:
      cpu: "500m"
      memory: "512Mi"
    limits:
      cpu: "2000m"
      memory: "2Gi"
  workerResources:
    requests:
      cpu: "1000m"
      memory: "1Gi"

CRD field definitions: api/tfrobot/v1/tfrserver_types.go:401-418

Quota Capacity Planning

Minimum quota for a single tenant (1 TFRTenant + 1 TFRServer) using default values:

Resource Type	Requests	Limits
CPU	3.1 (1.85 + 1.25)	13.4 (8.4 + 5)
Memory	4.3Gi (3 + 1.25)	16.5Gi (11.5 + 5)

Note

• The above covers only Operator-managed Pods, excluding Kubernetes system components and third-party Operator Pods
• For multiple TFRServers, scale proportionally: each additional TFRServer adds 1.25 CPU / 1.25Gi to Requests
• Recommend reserving 20-30% headroom for rolling updates and temporary Pods

Code Reference Index

Module	Path
TFRTenant component defaults	internal/controller/infra/tenant/resources.go
ResourceQuota Deployable	internal/controller/infra/tenant/resource_quota.go
TFRTenant auto-fill logic	internal/controller/infra/tfrtenant_controller.go:255-295
TFRServer container defaults	internal/controller/tfrobot/deployables/resources.go
TFRServer auto-fill logic	internal/controller/tfrobot/tfrserver_controller.go:286-295
ResourceQuota CRD types	api/infra/v1/tfrtenant_types.go:316-329
Container resource CRD types	api/tfrobot/v1/tfrserver_types.go:401-418
Options definitions	internal/controller/infra/tenant/options.go