The code for this project is in the skill-service directory of the book’s repository. It is implemented in Go. While this may not be a language you have used before, it is worth taking a look at the code to see how it works.
The actual code for storing and retrieving the data is in the internal/skill/autocomplete package and is not specific to Cloud Run.
The main.go file is the entry point for the application and is where the Cloud Run-specific code is located. In the init function, the configuration the service needs is loaded from environment variables and instances of Google Cloud-specific clients are created to interact with the Cloud Storage and Cloud Logging services. It is good to initialize these clients once and reuse them rather than creating them each time a request is received.
The main function sets up an HTTP server with three endpoints. This is using the popular Gorilla Mux library for routing requests. The three endpoints are:
GET /readiness
This is used by Cloud Run to determine if the service is healthy. It returns a 200 status. If it fails to return, Cloud Run will assume the instance has failed, terminate it, and start a new one.
GET /liveness
This is used by Cloud Run to determine if the service is ready to receive requests. It returns a 200 status code only after a flag is set when the trie data structure has been loaded into memory. This is to ensure the service is not sent requests by Cloud Run until it has the data it needs to respond to them.
GET /autocomplete
This is the endpoint that returns the suggestions to the user. It takes a query parameter query which is the text the user has typed so far. It returns a JSON array of strings that are the suggestions. This is the only endpoint that will ultimately be exposed to the user.
It is good to understand that Cloud Run just needs to provide an HTTP (or gRPC) endpoint running on port 8080 by default. There is no need to use a specific framework or library or even provide a specific entry point as with Cloud Functions. The point is, you can use whatever you want to build your service, and I am sure you will have your preferences.
Another piece of functionality in the main function of note is code to deal with a graceful shutdown. This is important, as Cloud Run will send a SIGTERM signal to the container when it is scaling down or terminating the instance. This is the disposability principle from the 12 factors. Applications should expect to be terminated at any time and should handle this gracefully. Here it is done by listening for the SIGTERM signal and then calling the Shutdown method on the HTTP server. This will allow any in-flight requests to be completed before the instance is terminated. In Chapter 3, I talked about services being largely stateless, but temporary state is acceptable. This is the time to clear up that temporary state.
The populate function retrieves the tags.csv file, created by the Tag Updater function from Chapter 5. It accesses Cloud Storage using the storage client. It then loads the data into a trie data structure. This is a treelike data structure that is optimized for searching for strings. The trie is stored in memory and is used to respond to requests.