core: use exponential backoff for name resolution

core/src/main/java/io/grpc/NameResolver.java

@@ -30,6 +30,10 @@
  * <p>The addresses and attributes of a target may be changed over time, thus the caller registers a
  * {@link Listener} to receive continuous updates.
  *
+ * <p>A {@code NameResolver} does not need to automatically re-resolve on failure. Instead, the
+ * {@link Listener} is responsible for eventually (after an appropriate backoff period) invoking
+ * {@link #refresh()}.
+ *
  * @since 1.0.0
  */
 @ExperimentalApi("https://github.com/grpc/grpc-java/issues/1770")
@@ -136,7 +140,8 @@ public interface Listener {
     void onAddresses(List<EquivalentAddressGroup> servers, Attributes attributes);
 
     /**
-     * Handles an error from the resolver.
+     * Handles an error from the resolver. The listener is responsible for eventually invoking
+     * {@link #refresh()} to re-attempt resolution.
      *
      * @param error a non-OK status
      * @since 1.0.0

core/src/main/java/io/grpc/internal/DnsNameResolver.java

@@ -36,9 +36,6 @@
 import java.util.Collections;
 import java.util.List;
 import java.util.concurrent.ExecutorService;
-import java.util.concurrent.ScheduledExecutorService;
-import java.util.concurrent.ScheduledFuture;
-import java.util.concurrent.TimeUnit;
 import java.util.logging.Level;
 import java.util.logging.Logger;
 import java.util.regex.Pattern;
@@ -79,29 +76,22 @@ final class DnsNameResolver extends NameResolver {
   private final String authority;
   private final String host;
   private final int port;
-  private final Resource<ScheduledExecutorService> timerServiceResource;
   private final Resource<ExecutorService> executorResource;
   private final ProxyDetector proxyDetector;
   @GuardedBy("this")
   private boolean shutdown;
   @GuardedBy("this")
-  private ScheduledExecutorService timerService;
-  @GuardedBy("this")
   private ExecutorService executor;
   @GuardedBy("this")
-  private ScheduledFuture<?> resolutionTask;
-  @GuardedBy("this")
   private boolean resolving;
   @GuardedBy("this")
   private Listener listener;
 
   DnsNameResolver(@Nullable String nsAuthority, String name, Attributes params,
-      Resource<ScheduledExecutorService> timerServiceResource,
       Resource<ExecutorService> executorResource,
       ProxyDetector proxyDetector) {
     // TODO: if a DNS server is provided as nsAuthority, use it.
     // https://www.captechconsulting.com/blogs/accessing-the-dusty-corners-of-dns-with-java
-    this.timerServiceResource = timerServiceResource;
     this.executorResource = executorResource;
     // Must prepend a "//" to the name when constructing a URI, otherwise it will be treated as an
     // opaque URI, thus the authority and host of the resulted URI would be null.
@@ -131,7 +121,6 @@ public final String getServiceAuthority() {
   @Override
   public final synchronized void start(Listener listener) {
     Preconditions.checkState(this.listener == null, "already started");
-    timerService = SharedResourceHolder.get(timerServiceResource);
     executor = SharedResourceHolder.get(executorResource);
     this.listener = Preconditions.checkNotNull(listener, "listener");
     resolve();
@@ -148,11 +137,6 @@ public final synchronized void refresh() {
       public void run() {
         Listener savedListener;
         synchronized (DnsNameResolver.this) {
-          // If this task is started by refresh(), there might already be a scheduled task.
-          if (resolutionTask != null) {
-            resolutionTask.cancel(false);
-            resolutionTask = null;
-          }
           if (shutdown) {
             return;
           }
@@ -171,16 +155,6 @@ public void run() {
           try {
             resolvedInetAddrs = delegateResolver.resolve(host);
           } catch (Exception e) {
-            synchronized (DnsNameResolver.this) {
-              if (shutdown) {
-                return;
-              }
-              // Because timerService is the single-threaded GrpcUtil.TIMER_SERVICE in production,
-              // we need to delegate the blocking work to the executor
-              resolutionTask =
-                  timerService.schedule(new LogExceptionRunnable(resolutionRunnableOnExecutor),
-                      1, TimeUnit.MINUTES);
-            }
             savedListener.onError(
                 Status.UNAVAILABLE.withDescription("Unable to resolve host " + host).withCause(e));
             return;
@@ -207,17 +181,6 @@ public void run() {
       }
     };
 
-  private final Runnable resolutionRunnableOnExecutor = new Runnable() {
-      @Override
-      public void run() {
-        synchronized (DnsNameResolver.this) {
-          if (!shutdown) {
-            executor.execute(resolutionRunnable);
-          }
-        }
-      }
-    };
-
   @GuardedBy("this")
   private void resolve() {
     if (resolving || shutdown) {
@@ -232,12 +195,6 @@ public final synchronized void shutdown() {
       return;
     }
     shutdown = true;
-    if (resolutionTask != null) {
-      resolutionTask.cancel(false);
-    }
-    if (timerService != null) {
-      timerService = SharedResourceHolder.release(timerServiceResource, timerService);
-    }
     if (executor != null) {
       executor = SharedResourceHolder.release(executorResource, executor);
     }

core/src/main/java/io/grpc/internal/DnsNameResolverProvider.java

@@ -47,8 +47,12 @@ public DnsNameResolver newNameResolver(URI targetUri, Attributes params) {
       Preconditions.checkArgument(targetPath.startsWith("/"),
           "the path component (%s) of the target (%s) must start with '/'", targetPath, targetUri);
       String name = targetPath.substring(1);
-      return new DnsNameResolver(targetUri.getAuthority(), name, params, GrpcUtil.TIMER_SERVICE,
-          GrpcUtil.SHARED_CHANNEL_EXECUTOR, GrpcUtil.getProxyDetector());
+      return new DnsNameResolver(
+          targetUri.getAuthority(),
+          name,
+          params,
+          GrpcUtil.SHARED_CHANNEL_EXECUTOR,
+          GrpcUtil.getProxyDetector());
     } else {
       return null;
     }

core/src/main/java/io/grpc/internal/ManagedChannelImpl.java

@@ -244,6 +244,7 @@ public void transportTerminated() {
             lbHelper = null;
           }
           if (nameResolver != null) {
+            cancelNameResolverBackoff();
             nameResolver.shutdown();
             nameResolver = null;
             nameResolverStarted = false;
@@ -366,6 +367,7 @@ private void enterIdleMode() {
     // either the idleModeTimer ran twice without exiting the idle mode, or the task in shutdown()
     // did not cancel idleModeTimer, or prepareToLoseNetwork() ran while shutdown or in idle, all of
     // which are bugs.
+    cancelNameResolverBackoff();
     nameResolver.shutdown();
     nameResolverStarted = false;
     nameResolver = getNameResolver(target, nameResolverFactory, nameResolverParams);
@@ -402,6 +404,46 @@ public void run() {
         idleTimeoutMillis, TimeUnit.MILLISECONDS);
   }
 
+  // Run from channelExecutor
+  @VisibleForTesting
+  class NameResolverRefresh implements Runnable {
+    // Only mutated from channelExecutor
+    boolean cancelled;
+
+    @Override
+    public void run() {
+      if (cancelled) {
+        // Race detected: this task was scheduled on channelExecutor before
+        // cancelNameResolverBackoff() could cancel the timer.
+        return;
+      }
+      nameResolverRefreshFuture = null;
+      nameResolverRefresh = null;
+      if (nameResolver != null) {
+        nameResolver.refresh();
+      }
+    }
+  }
+
+  // Must be used from channelExecutor
+  @Nullable private ScheduledFuture<?> nameResolverRefreshFuture;
+  // Must be used from channelExecutor
+  @Nullable private NameResolverRefresh nameResolverRefresh;
+  // The policy to control backoff between name resolution attempts. Non-null when an attempt is
+  // scheduled. Must be used from channelExecutor
+  @Nullable private BackoffPolicy nameResolverBackoffPolicy;
+
+  // Must be run from channelExecutor
+  private void cancelNameResolverBackoff() {
+    if (nameResolverRefreshFuture != null) {
+      nameResolverRefreshFuture.cancel(false);
+      nameResolverRefresh.cancelled = true;
+      nameResolverRefreshFuture = null;
+      nameResolverRefresh = null;
+      nameResolverBackoffPolicy = null;
+    }
+  }
+
   private final ClientTransportProvider transportProvider = new ClientTransportProvider() {
     @Override
     public ClientTransport get(PickSubchannelArgs args) {
@@ -745,24 +787,28 @@ public void run() {
 
   @Override
   public void resetConnectBackoff() {
-    channelExecutor.executeLater(
-        new Runnable() {
-          @Override
-          public void run() {
-            if (shutdown.get()) {
-              return;
-            }
-            if (nameResolverStarted) {
-              nameResolver.refresh();
-            }
-            for (InternalSubchannel subchannel : subchannels) {
-              subchannel.resetConnectBackoff();
-            }
-            for (InternalSubchannel oobChannel : oobChannels) {
-              oobChannel.resetConnectBackoff();
-            }
-          }
-        }).drain();
+    channelExecutor
+        .executeLater(
+            new Runnable() {
+              @Override
+              public void run() {
+                if (shutdown.get()) {
+                  return;
+                }
+                if (nameResolverRefreshFuture != null) {
+                  checkState(nameResolverStarted, "name resolver must be started");
+                  cancelNameResolverBackoff();
+                  nameResolver.refresh();
+                }
+                for (InternalSubchannel subchannel : subchannels) {
+                  subchannel.resetConnectBackoff();
+                }
+                for (InternalSubchannel oobChannel : oobChannels) {
+                  oobChannel.resetConnectBackoff();
+                }
+              }
+            })
+        .drain();
   }
 
   @Override
@@ -1072,6 +1118,8 @@ public void run() {
             return;
           }
 
+          nameResolverBackoffPolicy = null;
+
           try {
             if (retryEnabled) {
               retryPolicies = getRetryPolicies(config);
@@ -1105,16 +1153,41 @@ public void onError(final Status error) {
       checkArgument(!error.isOk(), "the error status must not be OK");
       logger.log(Level.WARNING, "[{0}] Failed to resolve name. status={1}",
           new Object[] {getLogId(), error});
-      channelExecutor.executeLater(new Runnable() {
-          @Override
-          public void run() {
-            // Call LB only if it's not shutdown.  If LB is shutdown, lbHelper won't match.
-            if (NameResolverListenerImpl.this.helper != ManagedChannelImpl.this.lbHelper) {
-              return;
-            }
-            balancer.handleNameResolutionError(error);
-          }
-        }).drain();
+      channelExecutor
+          .executeLater(
+              new Runnable() {
+                @Override
+                public void run() {
+                  // Call LB only if it's not shutdown.  If LB is shutdown, lbHelper won't match.
+                  if (NameResolverListenerImpl.this.helper != ManagedChannelImpl.this.lbHelper) {
+                    return;
+                  }
+                  balancer.handleNameResolutionError(error);
+                  if (nameResolverRefreshFuture != null) {
+                    // The name resolver may invoke onError multiple times, but we only want to
+                    // schedule one backoff attempt
+                    // TODO(ericgribkoff) Update contract of NameResolver.Listener or decide if we
+                    // want to reset the backoff interval upon repeated onError() calls
+                    return;
+                  }
+                  if (nameResolverBackoffPolicy == null) {
+                    nameResolverBackoffPolicy = backoffPolicyProvider.get();
+                  }
+                  long delayNanos = nameResolverBackoffPolicy.nextBackoffNanos();
+                  if (logger.isLoggable(Level.FINE)) {
+                    logger.log(
+                        Level.FINE,
+                        "[{0}] Scheduling DNS resolution backoff for {1} ns",
+                        new Object[] {logId, delayNanos});
+                  }
+                  nameResolverRefresh = new NameResolverRefresh();
+                  nameResolverRefreshFuture =
+                      transportFactory
+                          .getScheduledExecutorService()
+                          .schedule(nameResolverRefresh, delayNanos, TimeUnit.NANOSECONDS);
+                }
+              })
+          .drain();
     }
   }