Can ClientCall.Listener.onClose() ever run after asyncUnaryCall completes.

[catchpolecanva]

We use a ClientInterceptor with a SimpleForwardingClientCallListener to track call latency and other various metrics for gRPC calls. Under some circumstances, we notice that our latencies blow out from a typical 50 ms, up to 2000 ms. I accept that there are all sorts of reasons for delayed requests on a large busy system, especially with load balancers involved. The time of our actual server method call does not change, nor are there backed up requests etc, so we think it's related to the communications layer, not the call itself.

I put an extra timer around whole call, from the caller's perspective. I'm seeing initial metrics which suggest that the time observed by the caller of the method is in the expected range but sometimes the time recorded by ClientCall.Listener.onClose() is much higher.

If I put a sleep in onClose I can see it delays the whole call, so this suggests it's not running async, or at least not by default.

My question is simply, are their circumstances where gRPC internally will allow the asyncUnaryCall to complete before it has run ClientCall.Listener.onClose() from a ClientInterceptor?

My metrics suggest this can happen, but there is also the possibility my metrics are inaccurate.