2
2
mirror of https://github.com/octoleo/restic.git synced 2024-11-27 15:26:37 +00:00
restic/vendor/github.com/Azure/azure-sdk-for-go/arm/README.md
2017-08-06 21:47:04 +02:00

17 KiB

Introducing the Azure Resource Manager packages for Go

The github.com/Azure/azure-sdk-for-go/arm packages are used to perform operations using the Azure Resource Manager (ARM). Read more about Azure Resource Manager vs. classic deployment. Packages for Azure Service Manager or classic deployment are in the management folder.

How Did We Get Here?

Azure is growing rapidly, regularly adding new services and features. While rapid growth is good for users, it is hard on SDKs. Each new service and each new feature requires someone to learn the details and add the needed code to the SDK. As a result, the Azure SDK for Go has lagged behind Azure. It is missing entire services and has not kept current with features. There is simply too much change to maintain a hand-written SDK.

For this reason, the Azure SDK for Go, with the release of the Azure Resource Manager (ARM) packages, is transitioning to a generated-code model. Other Azure SDKs, notably the Azure SDK for .NET, have successfully adopted a generated-code strategy. Recently, Microsoft published the AutoRest tool used to create these SDKs and we have been adding support for Go. The ARM packages are the first set generated using this new toolchain. The input for AutoRest are the Azure REST API specs, files in Swagger JSON format.

There are a couple of items to note. First, since both the tooling and the underlying support packages are new, the code is not yet "production ready". Treat these packages as of beta quality. That's not to say we don't believe in the code, but we want to see what others think and how well they work in a variety of environments before settling down into an official, first release. If you find problems or have suggestions, please submit a pull request to document what you find. However, since the code is generated, we'll use your pull request to guide changes we make to the underlying generator versus merging the pull request itself.

The second item of note is that, to keep the generated code clean and reliable, it depends on another new package go-autorest. Though part of the SDK, we separated the code to better control versioning and maintain agility. Since go-autorest is hand-crafted, we will take pull requests in the same manner as for our other repositories.

We intend to rapidly improve these packages until they are "production ready". So, try them out and give us your thoughts.

What Have We Done?

Creating new frameworks is hard and often leads to "cliffs": The code is easy to use until some special case or tweak arises and then, well, then you're stuck. Often times small differences in requirements can lead to forking the code and investing a lot of time. Cliffs occur even more frequently in generated code. We wanted to avoid them and believe the new model does. Our initial goals were:

  • Easy-to-use out of the box. It should be "clone and go" for straight-forward use.
  • Easy composition to handle the majority of complex cases.
  • Easy to integrate with existing frameworks, fit nicely with channels, supporting fan-out / fan-in set ups.

These are best shown in a series of examples, all of which are included in the examples sub-folder.

How is the SDK tested?

Testing the SDK is currently a work in progress. It includes three different points:

  • Test the Azure REST API specs against the APIs themselves. This way we can find if the specs are reflecting correctly the API behavior. All Azure SDKs can benefit from this tests.
  • Add acceptance tests to AutoRest.
  • Test the generated SDK with code samples. This would catch bugs that escaped the previous tests, and provide some documentation.

First a Sidenote: Authentication and the Azure Resource Manager

Before using the Azure Resource Manager packages, you need to understand how it authenticates and authorizes requests. Azure Resource Manager requests can be authorized through OAuth2. While OAuth2 provides many advantages over certificates, programmatic use, such as for scripts on headless servers, requires understanding and creating one or more Service Principals.

The Azure-SDK-for-Node has an excellent tutorial that includes instructions for how to create Service Principals in the Portal and using the Azure CLI, both of which are applicable to Go. Find that documentation here: Authenticaion, Azure/azure-sdk-for-node

In addition, there are several good blog posts, such as Automating Azure on your CI server using a Service Principal and Microsoft Azure REST API + OAuth 2.0, that describe what this means. For details on creating and authorizing Service Principals, see the MSDN articles Azure API Management REST API Authentication and Create a new Azure Service Principal using the Azure portal. Dushyant Gill, a Senior Program Manager for Azure Active Directory, has written an extensive blog post, Developer's Guide to Auth with Azure Resource Manager API, that is also quite helpful.

Complete source code

Get code for a full example of authenticating to Azure via certificate or device authorization.

A Simple Example: Checking availability of name within Azure Storage

Each ARM provider, such as Azure Storage or Azure Compute, has its own package. Start by importing the packages for the providers you need. Next, most packages divide their APIs across multiple clients to avoid name collision and improve usability. For example, the Azure Storage package has two clients: storage.AccountsClient and storage.UsageOperationsClient. To check if a name is available, use the storage.AccountsClient.

Each ARM client composes with autorest.Client. autorest.Client enables altering the behavior of the API calls by leveraging the decorator pattern of go-autorest. For example, in the code above, the azure.ServicePrincipalToken includes a WithAuthorization autorest.PrepareDecorator that applies the OAuth2 authorization token to the request. It will, as needed, refresh the token using the supplied credentials.

Providing a decorated autorest.Sender or populating the autorest.Client with a custom autorest.PrepareDecorator or autorest.RespondDecorator enables more control. See the included example file check.go for more details. Through these you can modify the outgoing request, inspect the incoming response, or even go so far as to provide a circuit breaker to protect your service from unexpected latencies.

Lastly, all Azure ARM API calls return an instance of autorest.DetailedError. Not only DetailedError gives anonymous access to the original error, but provides the package type (e.g., storage.AccountsClient), the failing method (e.g., CheckNameAvailability), and a detailed error message.

Complete source code

Complete source code for this example can be found in check.go.

  1. Create a service principal. You will need the Tenant ID, Client ID and Client Secret for authentication, so keep them as soon as you get them.
  2. Get your Azure Subscription ID using either of the methods mentioned below:
  • Get it through the portal in the subscriptions section.
  • Get it using the Azure CLI with command azure account show.
  • Get it using Azure Powershell with cmdlet Get-AzureRmSubscription.
  1. Set environment variables AZURE_TENANT_ID = <TENANT_ID>, AZURE_CLIENT_ID = <CLIENT_ID>, AZURE_CLIENT_SECRET = <CLIENT_SECRET> and AZURE_SUBSCRIPTION_ID = <SUBSCRIPTION_ID>.
  2. Run the sample with commands:
$ cd arm/examples/check
$ go run check.go

Something a Bit More Complex: Creating a new Azure Storage account

Redundancy, both local and across regions, and service load affect service responsiveness. Some API calls will return before having completed the request. An Azure ARM API call indicates the request is incomplete (versus the request failed for some reason) by returning HTTP status code '202 Accepted.' The autorest.Client composed into all of the Azure ARM clients, provides support for basic request polling. The default is to poll until a specified duration has passed (with polling frequency determined by the HTTP Retry-After header in the response). By changing the autorest.Client settings, you can poll for a fixed number of attempts or elect to not poll at all.

Whether you elect to poll or not, all Azure ARM client responses compose with an instance of autorest.Response. At present, autorest.Response only composes over the standard http.Response object (that may change as we implement more features). When your code receives an error from an Azure ARM API call, you may find it useful to inspect the HTTP status code contained in the returned autorest.Response. If, for example, it is an HTTP 202, then you can use the GetPollingLocation response method to extract the URL at which to continue polling. Similarly, the GetPollingDelay response method returns, as a time.Duration, the service suggested minimum polling delay.

Creating a new Azure storage account is a straight-forward way to see these concepts.

autorest.Client portion of the storage.AccountsClient to poll for a fixed number of attempts versus polling for a set duration (which is the default). If an error occurs creating the storage account, the code inspects the HTTP status code and prints the URL the Azure Storage service returned for polling.

Complete source for the example

More details, including deleting the created account, are in the example code file create.go

  1. Create a service principal. You will need the Tenant ID, Client ID and Client Secret for authentication, so keep them as soon as you get them.
  2. Get your Azure Subscription ID using either of the methods mentioned below:
  • Get it through the portal in the subscriptions section.
  • Get it using the Azure CLI with command azure account show.
  • Get it using Azure Powershell with cmdlet Get-AzureRmSubscription.
  1. Set environment variables AZURE_TENANT_ID = <TENANT_ID>, AZURE_CLIENT_ID = <CLIENT_ID>, AZURE_CLIENT_SECRET = <CLIENT_SECRET> and AZURE_SUBSCRIPTION_ID = <SUBSCRIPTION_ID>.
  2. Create a resource group and add its name in the first line of the main function.
  3. Run the example with commands:
$ cd arm/examples/create
$ go run create.go

Making Asynchronous Requests

One of Go's many strong points is how natural it makes sending and managing asynchronous requests by means of goroutines. We wanted the ARM packages to fit naturally in the variety of asynchronous patterns used in Go code, but also be straight-forward for simple use cases. We accomplished both by adopting a pattern for all APIs. Each package API includes (at least) four methods (more if the API returns a paged result set). For example, for an API call named Foo the package defines:

  • FooPreparer: This method accepts the arguments for the API and returns a prepared http.Request.
  • FooSender: This method sends the prepared http.Request. It handles the possible status codes and will, unless the disabled in the autorest.Client, handling polling.
  • FooResponder: This method accepts and handles the http.Response returned by the sender and unmarshals the JSON, if any, into the result.
  • Foo: This method accepts the arguments for the API and returns the result. It is a wrapper around the FooPreparer, FooSender, and FooResponder.

By using the preparer, sender, and responder methods, package users can spread request and response handling across goroutines as needed. Further, adding a cancel channel to the http.Response (most easily through a PrepareDecorator), enables canceling sent requests (see the documentation on http.Request) for details.

Paged Result Sets

Some API calls return partial results. Typically, when they do, the result structure will include a Value array and a NextLink URL. The NextLink URL is used to retrieve the next page or block of results.

The packages add two methods to make working with and retrieving paged results natural. First, on paged result structures, the packages include a preparer method that returns an http.Request for the next set of results. For a result set returned in a structure named FooResults, the package will include a method named FooResultsPreparer. If the NextLink is nil or empty, the method returns nil.

The corresponding API (which typically includes "List" in the name) has a method to ease retrieving the next result set given a result set. For example, for an API named FooList, the package will include FooListNextResults that accepts the results of the last call and returns the next set.

Summing Up

The new Azure Resource Manager packages for the Azure SDK for Go are a big step toward keeping the SDK current with Azure's rapid growth. As mentioned, we intend to rapidly stabilize these packages for production use. We'll also add more examples, including some highlighting the Azure Resource Manager Templates and the other providers.

So, give the packages a try, explore the various ARM providers, and let us know what you think.

We look forward to hearing from you!

License

See the Azure SDK for Go LICENSE file.