MyGit

v22.0

graphql-java/graphql-java

版本发布时间: 2024-04-16 13:06:13

graphql-java/graphql-java最新发布版本:v22.3(2024-09-05 11:15:52)

We are pleased to announce the release of graphql-java v22.0.

Thanks to everyone in the community who contributed to the release, whether that was code, helping to report issues, or participating in discussions.

This is a breaking change release.

The graphql-java team takes breaking changes very seriously but in the name of performance we have made some significant breaking changes in this release.

Major Performance Changes

Past releases have been very much performance focused and this one is no different. However, this release is aiming to reduce memory pressure more than reduce pure CPU usage. When the graphql-java engine is running, if it produces less objects it will ultimately run faster because of reduced memory pressure and less garbage to collect.

The engine has been changed in two major ways that reduce memory pressure.

ExecutionResult wrapping

The first was that all values that come back got wrapped internally into a ExecutionResult object where the data attribute was the value. This was a carry over from some very early GraphQL code but was unnecessary and hence has been removed. The follow on effect of this is that some graphql.execution.ExecutionStrategy protected methods and graphql.execution.instrumentation.Instrumentation methods used to receive and return ExecutionResult values and no longer do, which is an API breaking change. We have made this breaking changes in the name of memory pressure performance.

CompletableFuture wrapping

The second major change is that the engine no longer exclusively uses java.util.concurrent.CompletableFutures when composing together results. Let us explain the past code first so we can discuss the new code.

CompletableFuture is a fantastic construct because it represents a promise to a value and it can also hold an exceptional state inside itself. Async programming with CompletableFuture is viral. If stage 1 of some compute process returns a CompletableFuture then stage 2 and 3 and 4 are likely to as well to ensure everything is asynchronous.

We use to take values that were not asynchronous (such as DataFetcher that returned a simple in memory object) and wrap them in a CompletableFuture so that all of the other code composed together using CompletableFuture methods like .thenApply() and .thenCompose and so on. The code is cleaner if you use all CompletableFuture code patterns.

However many objects in a GraphQL request are not asynchronous but rather materialised objects in memory. Scalars, enums and list are often just values immediately available to be used and allocating a CompletableFuture makes the code easier to write but it has a memory pressure cost.

So we have sacrificed cleaner code for more memory performant code.

graphql-java engine graphql.execution.ExecutionStrategy methods now just return Object where that object might be either a CompletableFuture or a materialised value.

    private Object /*CompletableFuture<FetchedValue> | FetchedValue>*/
       fetchField(GraphQLFieldDefinition fieldDef, ExecutionContext executionContext, ExecutionStrategyParameters parameters) {

Notice we have lost type safety here. In the past this would have only been CompletableFuture<FetchedValue> and hence been both type safe and async composable.

Now the caller of that method has to handle the async case where it might be a CompletableFuture<FetchedValue> AND the materialised value case where it might be a FetchedValue but as most simple fields in GraphQL are in fact materialised values, this is worth the less clean code.

DataFetchers can of course continue to return CompletableFuture values and they will be handled in a polymorphic manner.

The upshot of all this work is that the graphql-java engine allocated way less CompletableFuture values and hence reduces the amount of memory used.

Instrumentation Changes

The above changes now mean means that before graphql.execution.instrumentation.InstrumentationContext used to be given a CompletableFuture but now no longer does

    void onDispatched(CompletableFuture<T> result);

is now

    void onDispatched();

if you previously used the CompletableFuture to know when something was completed, well InstrumentationContext has the same semantics because it's completed method is similar in that it presents a value or an exception

    void onCompleted(T result, Throwable t);

graphql.execution.instrumentation.Instrumentation also had a lot of deprecated methods in place and these have been removed since the more performant shape of passing in graphql.execution.instrumentation.InstrumentationState has been in place for quite a few releases.

Some of the methods that received ExecutionResult wrapped values have also changed as mentioned above.

Instrumentation is probably the area that needs the most attention in terms of breaking changes. If you have not moved off the deprecated methods, your custom Instrumentations will no longer compile or run.

Interning key strings

Our friends at Netflix provided a PR that interns certain key strings like "field names" so that we create less String instances when processing field names in queries that we know must be previously interned in the schema.

https://github.com/graphql-java/graphql-java/pull/3504

The full list of performance related changes

https://github.com/graphql-java/graphql-java/pulls?q=is%3Apr+milestone%3A%2222.0%22+label%3Aperformance

Major Changes

@defer Experimental Support

Support for the @defer directive has been added in an experimental fashion. While @defer is still not in the released GraphQL specification, we have judged it mature enough to support at this stage and the graphql-js reference implementation has support for it.

https://github.com/graphql/graphql-wg/blob/main/rfcs/DeferStream.md

At this stage we have not yet supported @stream but this is likely to be included in a future release.

Breaking Changes

There are quite a few API breaking changes in this release. In fact there are 22 PRs containing breaking API changes.

Stricter parseValue coercion: Aligning with JS reference implementation

We have made changes to String, Boolean, Float, and Int parseValue coercion, to be consistent with the reference JS implementation. The key change is parseValue is now stricter on accepted inputs.

This is a breaking change. To help you migrate, in version 21.0, we introduced the InputInterceptor https://github.com/graphql-java/graphql-java/pull/3188 (and an implementation LegacyCoercingInputInterceptor https://github.com/graphql-java/graphql-java/pull/3218) to provide a migration pathway. You can use this interceptor to monitor and modify values.

For more, see https://github.com/graphql-java/graphql-java/pull/3553 JS reference implementation: https://github.com/graphql/graphql-js/blob/main/src/type/scalars.ts

Removing deprecated methods

Many methods that have been deprecated for a long time, sometimes even up to 6 years, have finally been removed.

The CompletableFuture unwrapping work mentioned above changed many methods in the graphql.execution.ExecutionStrategy classes but we don't expect this affect many people since very few people write their own engines.

That CompletableFuture unwrapping work also had knock on effects as mentioned above on graphql.execution.instrumentation.Instrumentation and hence this is the most likely place for there to be compatibility challenges in existing code.

DataLoaderDispatcherInstrumentation has been removed and is now built into the engine

The code to dispatch org.dataloader.DataLoaders used to be an instrumentation called DataLoadersDataLoaderDispatcherInstrumentation and it was automatically added at runtime. This approach has been removed.

The equivalent code is now built into the graphql-java engine itself. DataLoaders can now always be used without any special setup. This also allows the code to be more performant in how DataLoaders are dispatched.

SL4J logging has been removed

Previously, the graphql-java engine would log at DEBUG level certain errors or when fields get fetched. This has not proven used for from a support point to the graphql-java team and also has a compliance cost in that user generated content (UGC) and personally identifying information (PII) could end up being logged, which may not be allowed under regimes like European General Data Protection Regulation (GDPR).

If you want to log now we suggest you invest in a Instrumentation that logs key events and you can there for log what you want and in a compliant manner of your choosing.

https://github.com/graphql-java/graphql-java/pull/3403

The full list of API breaking changes

https://github.com/graphql-java/graphql-java/pulls?q=is%3Apr+milestone%3A%2222.0%22+label%3A%22breaking+change%22

Other changes

What's Changed

New Contributors

Full Changelog: https://github.com/graphql-java/graphql-java/compare/v21.5...v22.0

相关地址：原始地址 下载(tar) 下载(zip)

查看：2024-04-16发行的版本

微信小程序
MyGit:GitHub仓库更新&通知小工具