xds: Allow FaultFilter's interceptor to be reused

This is the only usage of PickSubchannelArgs when creating a filter's
ClientInterceptor, and a follow-up commit will remove the argument and
actually reuse the interceptors. Other filter's interceptors can
already be reused.

There doesn't seem to be any significant loss of legibility by making
FaultFilter a more ordinary interceptor, but the change does cause the
ForwardingClientCall to be present when faultDelay is configured,
independent of whether the fault delay ends up being triggered.

Reusing interceptors will move more state management out of the RPC path
which will be more relevant with RLQS.

Author: ejona86

xds/src/main/java/io/grpc/xds/FaultFilter.java

@@ -190,94 +190,102 @@ public ClientInterceptor buildClientInterceptor(
       config = overrideConfig;
     }
     FaultConfig faultConfig = (FaultConfig) config;
-    Long delayNanos = null;
-    Status abortStatus = null;
-    if (faultConfig.maxActiveFaults() == null
-        || activeFaultCounter.get() < faultConfig.maxActiveFaults()) {
-      Metadata headers = args.getHeaders();
-      if (faultConfig.faultDelay() != null) {
-        delayNanos = determineFaultDelayNanos(faultConfig.faultDelay(), headers);
-      }
-      if (faultConfig.faultAbort() != null) {
-        abortStatus = determineFaultAbortStatus(faultConfig.faultAbort(), headers);
-      }
-    }
-    if (delayNanos == null && abortStatus == null) {
-      return null;
-    }
-    final Long finalDelayNanos = delayNanos;
-    final Status finalAbortStatus = getAbortStatusWithDescription(abortStatus);
 
     final class FaultInjectionInterceptor implements ClientInterceptor {
       @Override
       public <ReqT, RespT> ClientCall<ReqT, RespT> interceptCall(
           final MethodDescriptor<ReqT, RespT> method, final CallOptions callOptions,
           final Channel next) {
-        Executor callExecutor = callOptions.getExecutor();
-        if (callExecutor == null) { // This should never happen in practice because
-          // ManagedChannelImpl.ConfigSelectingClientCall always provides CallOptions with
-          // a callExecutor.
-          // TODO(https://github.com/grpc/grpc-java/issues/7868)
-          callExecutor = MoreExecutors.directExecutor();
+        boolean checkFault = false;
+        if (faultConfig.maxActiveFaults() == null
+            || activeFaultCounter.get() < faultConfig.maxActiveFaults()) {
+          checkFault = faultConfig.faultDelay() != null || faultConfig.faultAbort() != null;
         }
-        if (finalDelayNanos != null) {
-          Supplier<? extends ClientCall<ReqT, RespT>> callSupplier;
-          if (finalAbortStatus != null) {
-            callSupplier = Suppliers.ofInstance(
-                new FailingClientCall<ReqT, RespT>(finalAbortStatus, callExecutor));
-          } else {
-            callSupplier = new Supplier<ClientCall<ReqT, RespT>>() {
-              @Override
-              public ClientCall<ReqT, RespT> get() {
-                return next.newCall(method, callOptions);
-              }
-            };
+        if (!checkFault) {
+          return next.newCall(method, callOptions);
+        }
+        final class DeadlineInsightForwardingCall extends ForwardingClientCall<ReqT, RespT> {
+          private ClientCall<ReqT, RespT> delegate;
+
+          @Override
+          protected ClientCall<ReqT, RespT> delegate() {
+            return delegate;
           }
-          final DelayInjectedCall<ReqT, RespT> delayInjectedCall = new DelayInjectedCall<>(
-              finalDelayNanos, callExecutor, scheduler, callOptions.getDeadline(), callSupplier);
 
-          final class DeadlineInsightForwardingCall extends ForwardingClientCall<ReqT, RespT> {
-            @Override
-            protected ClientCall<ReqT, RespT> delegate() {
-              return delayInjectedCall;
+          @Override
+          public void start(Listener<RespT> listener, Metadata headers) {
+            Executor callExecutor = callOptions.getExecutor();
+            if (callExecutor == null) { // This should never happen in practice because
+              // ManagedChannelImpl.ConfigSelectingClientCall always provides CallOptions with
+              // a callExecutor.
+              // TODO(https://github.com/grpc/grpc-java/issues/7868)
+              callExecutor = MoreExecutors.directExecutor();
             }
 
-            @Override
-            public void start(Listener<RespT> listener, Metadata headers) {
-              Listener<RespT> finalListener =
-                  new SimpleForwardingClientCallListener<RespT>(listener) {
-                    @Override
-                    public void onClose(Status status, Metadata trailers) {
-                      if (status.getCode().equals(Code.DEADLINE_EXCEEDED)) {
-                        // TODO(zdapeng:) check effective deadline locally, and
-                        //   do the following only if the local deadline is exceeded.
-                        //   (If the server sends DEADLINE_EXCEEDED for its own deadline, then the
-                        //   injected delay does not contribute to the error, because the request is
-                        //   only sent out after the delay. There could be a race between local and
-                        //   remote, but it is rather rare.)
-                        String description = String.format(
-                            Locale.US,
-                            "Deadline exceeded after up to %d ns of fault-injected delay",
-                            finalDelayNanos);
-                        if (status.getDescription() != null) {
-                          description = description + ": " + status.getDescription();
-                        }
-                        status = Status.DEADLINE_EXCEEDED
-                            .withDescription(description).withCause(status.getCause());
-                        // Replace trailers to prevent mixing sources of status and trailers.
-                        trailers = new Metadata();
+            Long delayNanos;
+            Status abortStatus = null;
+            if (faultConfig.faultDelay() != null) {
+              delayNanos = determineFaultDelayNanos(faultConfig.faultDelay(), headers);
+            } else {
+              delayNanos = null;
+            }
+            if (faultConfig.faultAbort() != null) {
+              abortStatus = getAbortStatusWithDescription(
+                  determineFaultAbortStatus(faultConfig.faultAbort(), headers));
+            }
+
+            Supplier<? extends ClientCall<ReqT, RespT>> callSupplier;
+            if (abortStatus != null) {
+              callSupplier = Suppliers.ofInstance(
+                  new FailingClientCall<ReqT, RespT>(abortStatus, callExecutor));
+            } else {
+              callSupplier = new Supplier<ClientCall<ReqT, RespT>>() {
+                @Override
+                public ClientCall<ReqT, RespT> get() {
+                  return next.newCall(method, callOptions);
+                }
+              };
+            }
+            if (delayNanos == null) {
+              delegate = callSupplier.get();
+              delegate().start(listener, headers);
+              return;
+            }
+
+            delegate = new DelayInjectedCall<>(
+                delayNanos, callExecutor, scheduler, callOptions.getDeadline(), callSupplier);
+
+            Listener<RespT> finalListener =
+                new SimpleForwardingClientCallListener<RespT>(listener) {
+                  @Override
+                  public void onClose(Status status, Metadata trailers) {
+                    if (status.getCode().equals(Code.DEADLINE_EXCEEDED)) {
+                      // TODO(zdapeng:) check effective deadline locally, and
+                      //   do the following only if the local deadline is exceeded.
+                      //   (If the server sends DEADLINE_EXCEEDED for its own deadline, then the
+                      //   injected delay does not contribute to the error, because the request is
+                      //   only sent out after the delay. There could be a race between local and
+                      //   remote, but it is rather rare.)
+                      String description = String.format(
+                          Locale.US,
+                          "Deadline exceeded after up to %d ns of fault-injected delay",
+                          delayNanos);
+                      if (status.getDescription() != null) {
+                        description = description + ": " + status.getDescription();
                       }
-                      delegate().onClose(status, trailers);
+                      status = Status.DEADLINE_EXCEEDED
+                          .withDescription(description).withCause(status.getCause());
+                      // Replace trailers to prevent mixing sources of status and trailers.
+                      trailers = new Metadata();
                     }
-                  };
-              delegate().start(finalListener, headers);
-            }
+                    delegate().onClose(status, trailers);
+                  }
+                };
+            delegate().start(finalListener, headers);
           }
-
-          return new DeadlineInsightForwardingCall();
-        } else {
-          return new FailingClientCall<>(finalAbortStatus, callExecutor);
         }
+
+        return new DeadlineInsightForwardingCall();
       }
     }