Question: may I as why the default thread pool is a ScheduledThreadPoolExecutor?

[liufuyang]

If I am correct, from the code here
https://github.com/googleapis/gax-java/blob/master/gax/src/main/java/com/google/api/gax/core/InstantiatingExecutorProvider.java#L63

Means the default threadpool for grpc connections will be a ScheduledThreadPoolExecutor
https://github.com/AdoptOpenJDK/openjdk-jdk11/blob/master/src/java.base/share/classes/java/util/concurrent/ScheduledThreadPoolExecutor.java#L470

public ScheduledThreadPoolExecutor(int corePoolSize,
                                       ThreadFactory threadFactory) {
        super(corePoolSize, Integer.MAX_VALUE,
              DEFAULT_KEEPALIVE_MILLIS, MILLISECONDS,
              new DelayedWorkQueue(), threadFactory);
    }

Which uses some DelayedWorkQueue internally. Will this gives the in general the best performance or is there some specific thoughts after using this type of thread pool?

How it compares with a ExecutorService.newFixedThreadPool that uses LinkedBlockingQueue or a fork-join pool?

public static ExecutorService newFixedThreadPool(int nThreads) {
        return new ThreadPoolExecutor(nThreads, nThreads,
                                      0L, TimeUnit.MILLISECONDS,
                                      new LinkedBlockingQueue<Runnable>());
    }

I saw some earlier discussion here grpc/grpc-java#7381 but I am not sure whether those statement are still valid now?

Thank you.

[liufuyang]

And perhaps this #869 can sort of solves the issue so it allows me to switch to a different executor for the grpc channels? (For example my application uses Bigtable, and the Bigtable client seems uses gax-java?)

If I may ask, I would like to know, when I setup a Bigtable client connection to google's bigtable service, what is the actual threadpool/executor used to handle the grpc calls. And is there a way to allow me use a different threadpool/executor on it?

[mutianf]

Hi @liufuyang

when I setup a Bigtable client connection to google's bigtable service, what is the actual threadpool/executor used to handle the grpc calls

If you didn't set a TransportChannelProvider or ExecutorProvider, by default it'll use the ScheduledThreadPoolExecutor you mentioned.

is there a way to allow me use a different threadpool/executor on it?

Yes. For bigtable client, you could do:

BigtableDataSettings.Builder settings = BigtableDataSettings.newBuilder()
  .setProjectId("my-project")
  .setInstanceId("my-instance");

// Create your executor
Executor executor = ...

settings.stubSettings()
  .setTransportChannelProvider(
      settings.stubSettings().getTransportChannelProvider().withExecutor(executor));

We're also working on a fix to use the default gRPC executor.

[liufuyang]

I see. Thanks for the info 👍 This could be helpful.