LogoLogo
WebsiteSlack
0.28
0.28
  • Get started
  • Clients
    • Install
    • CLI commands
    • Python API
    • Environments
    • Telemetry
    • Uninstall
  • Workloads
    • Realtime APIs
      • Example
      • Predictor
      • Configuration
      • Models
      • Parallelism
      • Server-side batching
      • Autoscaling
      • Statuses
      • Multi-model
        • Example
        • Configuration
        • Caching
      • Traffic Splitter
        • Example
        • Configuration
      • Troubleshooting
    • Batch APIs
      • Example
      • Predictor
      • Configuration
      • Jobs
      • Statuses
    • Task APIs
      • Example
      • Definition
      • Configuration
      • Jobs
      • Statuses
    • Dependencies
      • Example
      • Python packages
      • System packages
      • Custom images
  • Clusters
    • Cortex Cloud on AWS
      • Install
      • Update
      • Security
      • Logging
      • Spot instances
      • Networking
        • Custom domain
        • HTTPS (via API Gateway)
        • VPC peering
      • Setting up kubectl
      • Uninstall
    • Cortex Cloud on GCP
      • Install
      • Logging
      • Credentials
      • Setting up kubectl
      • Uninstall
    • Cortex Core on Kubernetes
      • Install
      • Uninstall
    • Private Docker registry
Powered by GitBook
On this page
  • Deploy APIs
  • Deploy a traffic splitter
  • Update the weights of the traffic splitter
  1. Workloads
  2. Realtime APIs
  3. Traffic Splitter

Example

Expose multiple RealtimeAPIs as a single endpoint for A/B tests, multi-armed bandits, or canary deployments.

Deploy APIs

import cortex

class PythonPredictor:
    def __init__(self, config):
        from transformers import pipeline
        self.model = pipeline(task="text-generation")

    def predict(self, payload):
        return self.model(payload["text"])[0]

requirements = ["tensorflow", "transformers"]

api_spec_cpu = {
    "name": "text-generator-cpu",
    "kind": "RealtimeAPI",
    "compute": {
        "cpu": 1,
    },
}

api_spec_gpu = {
    "name": "text-generator-gpu",
    "kind": "RealtimeAPI",
    "compute": {
        "gpu": 1,
    },
}

cx = cortex.client("aws")
cx.create_api(api_spec_cpu, predictor=PythonPredictor, requirements=requirements)
cx.create_api(api_spec_gpu, predictor=PythonPredictor, requirements=requirements)

Deploy a traffic splitter

traffic_splitter_spec = {
    "name": "text-generator",
    "kind": "TrafficSplitter",
    "apis": [
        {"name": "text-generator-cpu", "weight": 50},
        {"name": "text-generator-gpu", "weight": 50},
    ],
}

cx.create_api(traffic_splitter_spec)

Update the weights of the traffic splitter

traffic_splitter_spec = cx.get_api("text-generator")["spec"]["submitted_api_spec"]

# send 99% of the traffic to text-generator-gpu
traffic_splitter_spec["apis"][0]["weight"] = 1
traffic_splitter_spec["apis"][1]["weight"] = 99

cx.patch(traffic_splitter_spec)
PreviousTraffic SplitterNextConfiguration

Last updated 4 years ago