grpc 库是否支持阻塞来自完成队列的 n 个事件的完成状态？

Question

We make multiple asynchronous requests in parallel with a unique tag for each. We want to block and wait until all those events have completed with success, failure, or have timed out.

We can accomplish this by counting the total number of outstanding calls and decrementing the total in-flight by 1 each time an event is popped off of the queue. I really don't like this.

1. This requires that the completion queue is either instantiated on a per-call basis (this has considerable overhead) or that we re-use a queue but ensure that there is only 1 active user of the queue at a given time. That means for m threads making n calls, we need m queues each servicing n calls.

2. We decrement in-flight on the assumption that no other events will come from the queue. I don't know if this is a guarantee.

    int inFlight = 10; // eg with dummy data while(cq->Next(&tag, &ok) { inFlight--; if (inFlight == 0) break; /// process... }

Is there a way to block and wait for n tagged async events to finish that is supported by the library?

Is it possible to block and wait for these events while the queue is being multiplexed?

Answer 1

gRPC doesn't have the mechanism to support your case. Completion queue is a low-level component which handles async operations so it mainly works as a simple queue. So you might have to manage its own way to do a bookkeeping. ClientInterceptor might be helpful to do the same thing across different rpcs.