weather

Weather pulse Example

This example showcases the use of the various Pulse packages to build a simple weather forecast service that minimizes API requests and provides very fast response times. The system consists of two components:

• A poller service that leverages a Pulse worker pool and streaming to poll weather forecasts. The service exposes APIs to start poll jobs and subscribe to location specific weather forecast updates. The service also exposes "control" APIs to query statistics for the poll jobs, add or remove workers.

• A Forecaster service which provides HTTP APIs to query the latest forecast for a given location. The Forecaster service leverages the poller service to create new forecast poll jobs and to subscribe to updates.

The Forecaster service employs a cache system using a Pulse replicated map to store and retrieve the latest weather forecasts for different locations. When a forecast request is received, the service checks the cache and returns the cached forecast if available. In case of a cache miss, the forecaster service makes a request to the poller service which initiates a Pulse job to fetch the forecast from the weather API. The Forecaster service then subscribes to the replicated map and waits for the forecast to be available in the cache.

The poll job is executed by a dedicated worker that publishes the forecasts to a stream. The poller service subscribes to the stream and forwards forecasts to subscribers. The Forecaster service subscribes to the poller service and updates the cache with the latest forecasts.

This flow ensures efficient and timely retrieval of forecasts while minimizing API calls to the weather service. The following diagram shows the architecture of the weather system:

Running the Example

The scripts directory contains a couple of scripts that can be used to run the example:

• setup downloads build dependencies and builds the example.
• server runs the services using overmind. server also starts docker-compose with a configuration that runs Redis, the Grafana agent, cortex, tempo and dashboard locally.

Making a Request

The Forecaster service exposes a simple HTTP API that can be used to query the latest weather forecast for a given location. The following example shows how to query the weather forecast for ca/santa-barbara:

curl http://localhost:8080/forecast/ca/santa-barbara

The response is a JSON object that contains the latest weather forecast for santa-barbara:

{
    "location": {
        "city": "Santa Barbara",
        "lat": 34.4221319,
        "long": -119.702667,
        "state": "CA"
    },
    "periods": [
        {
            "endTime": "2023-05-31T06:00:00-07:00",
            "name": "Tonight",
            "startTime": "2023-05-30T20:00:00-07:00",
            "summary": "Mostly Cloudy",
            "temperature": 57,
            "temperatureUnit": "F"
        },
        ...
    ]
}

The poller service also expose a simple HTTP API that can be used to query status information.

curl http://localhost:8082/poller/status

The response is a JSON object that contains the list of workers and jobs that are currently being executed by the service:

{
    "jobs": [
        {
            "created_at": "2023-05-31T03:57:06Z",
            "key": "ca/santa-barbara",
            "payload": "AgAAAGNhDQAAAHNhbnRhLWJhcmJhcmE="
        }
    ],
    "workers": [
        {
            "created_at": "2023-05-30T20:57:05-07:00",
            "id": "01H1QZ7FNRQXPZ17KTWYDZ49SM",
            "jobs": [
                {
                    "created_at": "2023-05-31T03:57:06Z",
                    "key": "ca/santa-barbara",
                    "payload": "AgAAAGNhDQAAAHNhbnRhLWJhcmJhcmE="
                }
            ]
        }
    ]
}

Code Organization

The example consists of the following directories:

• forecaster contains the Forecaster service.
• poller contains the poller service.

Both the service and the worker are implemented using the Goa framework.

Each service contains a design package that defines the service API using the Goa DSL. The service directory itself contains the service implementation.

.
├── clients                         # HTTP clients for the poller service
├── cmd                                         # Service main
├── design                                # Service API definition
├── forecaster.go # Forecaster service implementation
├── gen                                         # Generated code
└── marshal.go                # Forecaster service implementation

The clients directory contains the downstream dependency clients used by the service implementation. In the case of the Forecaster service there is only one such client: the client used by the service to make requests to the poller service.

clients
└── poller
    ├── client.go
    └── mocks

The mocks package contains a mock implementation of the client used by the service. The mock is used by the service unit tests.

Critical Code & Reusable Patterns

The waitForForecast function in the forecaster.go file in the Forecaster service implements the logic to wait for a specific key to be available in a Pulse replicated map with a specific timeout. The function makes use of a select statement to wait for replicated map updates or a timeout:

// waitForecast waits for the forecast to be available in the forecasts map.
func (svc *Service) waitForecast(ctx context.Context, city, state string) (string, error) {
        // Wait up to 10 seconds for the forecast to be available in the map.
        ctx, cancel := context.WithDeadline(ctx, time.Now().Add(maxForecastWaitDelay))
        defer cancel()
        sub := svc.fmap.Subscribe()
        defer svc.fmap.Unsubscribe(sub)
        key := locationKey(city, state)
        for {
                select {
                case <-sub:
                        if marshaled, ok := svc.fmap.Get(key); ok {
                                return marshaled, nil
                        }
                case <-ctx.Done():
                        return "", ctx.Err()
                }
        }
}

The Subscribe method of the Poller service shows how to forward Pulse stream events to a WebSocket connection:

// Subscribe subscribes to forecasts for the given location.
func (svc *Service) Subscribe(ctx context.Context, location *genpoller.CityAndState, sub genpoller.SubscribeServerStream) error {
        // Create stream reader to retrieve events from the stream.
        r, err := svc.stream.NewReader(ctx,
                options.WithReaderTopic(locationKey(location.City, location.State)),
                options.WithReaderStartAt(time.Now().Add(-time.Hour)),
        )
        if err != nil {
                return fmt.Errorf("failed to subscribe to stream: %s", err)
        }
        // Don't forget to close the reader when done.
        defer r.Close()
        // Subscribe to the stream.
        ch := r.Subscribe()
        for {
                select {
                case <-ctx.Done():
                        // Client closed the connection
                        return nil
                case f, ok := <-ch:
                        if !ok {
                                // Service is shutting down.
                                return nil
                        }
                        // Got an event, unmarshal it and send it to the client.
                        ev, err := unmarshalForecastEvent(f.Payload)
                        if err != nil {
                                log.Errorf(ctx, err, "failed to unmarshal forecast event")
                                continue
                        }
                        if err := sub.Send(ev); err != nil {
                                return fmt.Errorf("failed to send forecast: %s", err)
                        }
                }
        }
}