Fix concurrent problems when initializing multiple GroupedOpenApi parallelly

[bianjp]

Problem

Continuation of #1641

If a project has multiple GroupedOpenApi bean, they will be initialized parallelly. This lead to concurrent read and write of AbstractOpenApiResource#HIDDEN_REST_CONTROLLERS.

Since the list is not correctly synchronized, concurrent read and write might cause ConcurrentModificationException (Collections#synchronizedList isn't thread-safe on iteration):

Exception in thread "pool-3-thread-1" java.util.ConcurrentModificationException
  at java.util.ArrayList$ArrayListSpliterator.tryAdvance(ArrayList.java:1363)
  at java.util.stream.ReferencePipeline.forEachWithCancel(ReferencePipeline.java:126)
  at java.util.stream.AbstractPipeline.copyIntoWithCancel(AbstractPipeline.java:499)
  at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:486)
  at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:472)
  at java.util.stream.MatchOps$MatchOp.evaluateSequential(MatchOps.java:230)
  at java.util.stream.MatchOps$MatchOp.evaluateSequential(MatchOps.java:196)
  at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)
  at java.util.stream.ReferencePipeline.anyMatch(ReferencePipeline.java:516)
  at org.springdoc.api.AbstractOpenApiResource.isHiddenRestControllers(AbstractOpenApiResource.java:822)
  at org.springdoc.api.AbstractOpenApiResource.lambda$getOpenApi$3(AbstractOpenApiResource.java:309)
  at java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:174)
  at java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:175)
  at java.util.HashMap$EntrySpliterator.forEachRemaining(HashMap.java:1723)
  at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:482)
  at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:472)
  at java.util.stream.ReduceOps$ReduceOp.evaluateSequential(ReduceOps.java:708)
  at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)
  at java.util.stream.ReferencePipeline.collect(ReferencePipeline.java:566)
  at org.springdoc.api.AbstractOpenApiResource.getOpenApi(AbstractOpenApiResource.java:310)
  at org.springdoc.api.AbstractOpenApiResource.getOpenApi(AbstractOpenApiResource.java:256)
  at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
  at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
  at java.lang.Thread.run(Thread.java:750)

Solution

There are 3 possible solutions:

1. Synchronize on the list before every read and write:

   synchronized (HIDDEN_REST_CONTROLLERS) {
       return HIDDEN_REST_CONTROLLERS.stream().anyMatch(clazz -> clazz.isAssignableFrom(rawClass));
   }

   • Cons: The list is much more often used for read than write (in my project it's 1698 to 12), so synchronize on every read might hurt performance.

2. Use an fully thread-safe implementation for list, such as CopyOnWriteArrayList

   • Pros: Read operations don't need synchronization
   • Cons: Write operations are very expensive since they usually entail copying the entire underlying array. However, each GroupedOpenApi only triggers two write operations for HIDDEN_REST_CONTROLLERS, and the list is generally small, so the trade-off is acceptable.

3. Reactor the initialization to eliminate concurrent write
   Each GroupedOpenApi will try to add the same controllers to HIDDEN_REST_CONTROLLERS in OpenAPIService#initializeHiddenRestController, since the controllers are obtained from the same ApplicationContext.
   It's better to write to HIDDEN_REST_CONTROLLERS only once before parallel initialization. This can not only eliminate concurrent write but also avoid duplication.
   However, this refactoring requires more familiarity of Springdoc's code base, and it's beyond my powers at present.

This PR implements solution 2.

I also replaced ADDITIONAL_REST_CONTROLLERS, HIDDEN_REST_CONTROLLERS with set to avoid duplication. This will reduce the number of elements to iterate thus improving performance.

[bnasslahsen]

Thank you @bianjp for your contribution to the project.

[bianjp]

@bnasslahsen Can you release a new version?