Dashboard REST API

The Strata Coordinator provides a RESTful HTTP API primarily designed for the web dashboard, but it is also available for custom monitoring, CLI tools, and automation scripts. The API allows you to query the state of the distributed cluster, monitor performance metrics, and manage training tasks.

Base Configuration

By default, the REST API is served by the coordinator. In a local development environment, the base URL is:

http://localhost:3000/api

All responses are returned in JSON format.

System Endpoints

Health Check

GET /api/health

Returns the connectivity status of the coordinator.

Response:

{
  "status": "ok"
}

Coordinator Status

GET /api/status

Returns version information and uptime for the coordinator node.

Response:

{
  "connected": true,
  "address": "0.0.0.0:50051",
  "uptime": 3600,
  "version": "0.1.0"
}

Cluster & Data Monitoring

Worker Inventory

GET /api/workers

Returns a list of all workers currently registered with the coordinator, including their hardware specs and current training progress.

Response Schema (WorkerResponse): | Field | Type | Description | | :--- | :--- | :--- | | id | string | Unique worker identifier | | status | string | active, idle, failed, or recovering | | gpu_count | integer | Number of GPUs available on this node | | assigned_shards | integer | Number of data shards currently held | | current_step | integer | Last reported training step |

Dataset Registry

GET /api/datasets

Lists all datasets registered for sharding and distribution.

Response Schema (DatasetResponse): | Field | Type | Description | | :--- | :--- | :--- | | id | string | Dataset identifier | | total_samples | integer | Total number of records in the dataset | | shard_count | integer | Number of logical shards created | | format | string | e.g., parquet, tfrecord, webdataset |

Checkpoint History

GET /api/checkpoints

Returns a list of saved model checkpoints across all workers and storage backends.

Task Management

List Tasks

GET /api/tasks

Returns all active and completed training tasks.

Create Task

POST /api/tasks

Submit a new training job to the cluster.

Request Body:

{
  "name": "ResNet-50-Imagenet",
  "type": "training",
  "dataset_id": "imagenet-2024",
  "worker_count": 8,
  "config": {
    "learning_rate": 0.001,
    "batch_size": 32
  }
}

Stop Task

POST /api/tasks/:task_id/stop

Signals a running task to stop gracefully at the next synchronization barrier.

Metrics & Observability

System Metrics

GET /api/metrics

Provides real-time performance telemetry for the entire runtime.

Response:

{
  "checkpoint_throughput": 450,
  "coordinator_rps": 1250,
  "active_workers": 64,
  "total_workers": 64,
  "barrier_latency_p99": 42,
  "shard_assignment_time": 8
}

• checkpoint_throughput: Measured in MB/s.
• barrier_latency_p99: The 99th percentile latency for worker synchronization (ms).
• shard_assignment_time: Time taken to calculate consistent hashing for the cluster (ms).

Global Logs

GET /api/logs?limit=100

Retrieves the latest system-wide log entries from the coordinator and workers.

Task-Specific Logs

GET /api/tasks/:task_id/logs

Returns logs filtered specifically for a single training job.

Aggregate State

Dashboard State

GET /api/dashboard

A "fat" endpoint that returns the complete system state (workers, datasets, checkpoints, barriers, metrics, and tasks) in a single request. This is the recommended endpoint for high-frequency dashboard polling to reduce network overhead.

Response Structure:

{
  "coordinator": { ... },
  "workers": [ ... ],
  "datasets": [ ... ],
  "checkpoints": [ ... ],
  "barriers": [ ... ],
  "metrics": { ... },
  "tasks": [ ... ],
  "logs": [ ... ]
}