Operators

Akash Provider uses Kubernetes operators to extend cluster functionality and manage custom resources. These operators run as separate services and communicate with the provider via Kubernetes Custom Resource Definitions (CRDs).

In This Section

Inventory Operator

Discovers and tracks cluster resources (CPU, GPU, memory, storage) in real-time.

IP Operator

Manages static IP address allocation for deployments via MetalLB integration.

Hostname Operator

Manages custom hostname assignments and creates ingress resources for deployments.

What Are Operators?

Kubernetes operators are applications that extend Kubernetes functionality by:

1. Watching Custom Resources - Monitor CRDs for changes
2. Reconciliation - Compare desired state with actual state
3. Taking Action - Create/update/delete Kubernetes resources
4. Continuous Monitoring - Ensure state consistency

Akash Operators

Inventory Operator

Purpose: Resource discovery and tracking

Key Features:

• Discovers cluster resources (CPU, GPU, storage)
• Tracks resource availability in real-time
• Supports hardware feature discovery (GPU models, storage classes)
• Integrates with NVIDIA Device Plugin for GPUs
• Updates Kubernetes node labels with capabilities
• Publishes inventory to provider service for bidding

When It Runs: Continuously, updates on hardware changes

IP Operator

Purpose: Static IP address management

Key Features:

• Allocates dedicated IP addresses for deployments
• Integrates with MetalLB for IP allocation
• Supports IP sharing across services (same tenant)
• Tracks IP pool usage and availability
• Handles IP lifecycle (create, update, delete)
• Exposes HTTP API for monitoring

When It Runs: Event-driven, responds to ProviderLeasedIP CRDs

Hostname Operator

Purpose: Custom hostname and ingress management

Key Features:

• Manages custom hostname assignments
• Creates Kubernetes Ingress resources automatically
• Integrates with ingress controllers (Ingress-NGINX, Traefik)
• Handles hostname conflicts and validation
• Supports wildcard hostnames
• Monitors hostname health

When It Runs: Event-driven, responds to ProviderHost CRDs

How Operators Integrate

+--------------------------------------------------+
|              Provider Service                    |
|  +-------------------------------------------+   |
|  |         Cluster Service                   |   |
|  |  - Creates CRDs (ProviderHost, etc.)      |   |
|  |  - Calls operator client interfaces       |   |
|  +-------------------+-----------------------+   |
+--------------------+--------------------------+
                     |
                     | Creates CRDs
                     v
            +--------------------+
            |   Kubernetes API   |
            |   (CRD Storage)    |
            +---------+----------+
                      |
                      | Watch Events
                      |
       +--------------+--------------+--------------+
       |              |              |              |
       v              v              v              v
+----------+   +----------+   +----------+   +----------+
|Inventory |   |    IP    |   | Hostname |   |  Other   |
|Operator  |   | Operator |   | Operator |   |Operators |
+----------+   +----------+   +----------+   +----------+
       |              |              |              |
       |              |              |              |
       v              v              v              v
+-----------------------------------------------------+
|           Kubernetes Resources                      |
|  - Nodes, Pods                                      |
|  - Services, Ingress                                |
|  - MetalLB IPAddressPools                           |
+-----------------------------------------------------+

Operator Communication

Via CRDs (Custom Resource Definitions)

Operators watch for CRD changes:

• ProviderHost - Hostname assignments
• ProviderLeasedIP - IP allocations
• Manifest - Deployment specifications

Via gRPC

Operators expose gRPC services for programmatic access to inventory, status, and state information.

Via Event Bus (PubSub)

Operators publish events:

• inventory.cluster - Resource availability updates
• inventory.nodes - Node hardware changes
• inventory.storage - Storage capacity updates

Related Documentation

• Provider Service Overview - High-level architecture
• Cluster Service - Resource management
• Bid Engine - Bidding logic
• Manifest Service - Manifest handling