Introduce new config to randomly pick the xDS server host (#972)

* Introduce new config to randomly pick the xDS server host

Instead of using the default "pick_first" policy, which always picks the first
IP address returned by DNS, "round_robin" will pick a random starting point in
the list of IPs. This will fix the very lopsided load seen on the xDS servers.

* Make policy configurable

* Bump version, update changelog

* Add log, update changelog

CHANGELOG.md

@@ -14,6 +14,9 @@ and what APIs have changed, if applicable.
 
 ## [Unreleased]
 
+## [29.49.9] - 2024-01-26
+- Introduce new config to randomly pick the xDS server host
+
 ## [29.49.8] - 2024-01-19
 - add WIRE_COMPATIBLE compatility checker mode.
 
@@ -5618,7 +5621,8 @@ patch operations can re-use these classes for generating patch messages.
 
 ## [0.14.1]
 
-[Unreleased]: https://github.com/linkedin/rest.li/compare/v29.49.8...master
+[Unreleased]: https://github.com/linkedin/rest.li/compare/v29.49.9...master
+[29.49.9]: https://github.com/linkedin/rest.li/compare/v29.49.8...v29.49.9
 [29.49.8]: https://github.com/linkedin/rest.li/compare/v29.49.7...v29.49.8
 [29.49.7]: https://github.com/linkedin/rest.li/compare/v29.49.6...v29.49.7
 [29.49.6]: https://github.com/linkedin/rest.li/compare/v29.49.5...v29.49.6

d2/src/main/java/com/linkedin/d2/xds/XdsChannelFactory.java

@@ -21,6 +21,7 @@
 import io.grpc.netty.shaded.io.grpc.netty.NettyChannelBuilder;
 import io.grpc.netty.shaded.io.netty.handler.ssl.SslContext;
 import java.util.concurrent.TimeUnit;
+import javax.annotation.Nullable;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
@@ -29,12 +30,40 @@ public class XdsChannelFactory
 {
   private static final Logger _log = LoggerFactory.getLogger(XdsChannelFactory.class);
 
+  public static final String ROUND_ROBIN_POLICY = "round_robin";
+
   private final SslContext _sslContext;
   private final String _xdsServerUri;
+  @Nullable
+  private final String _defaultLoadBalancingPolicy;
+
+  /**
+   * Invokes alternative constructor with {@code defaultLoadBalancingPolicy} as {@value ROUND_ROBIN_POLICY}.
+   */
+  public XdsChannelFactory(SslContext sslContext, String xdsServerUri)
+  {
+    this(sslContext, xdsServerUri, ROUND_ROBIN_POLICY);
+  }
 
-  public XdsChannelFactory(SslContext sslContext, String xdsServerUri) {
+  /**
+   * @param sslContext                 The sslContext to use. If {@code null}, SSL will not be used when connecting to
+   *                                   the xDS server.
+   * @param xdsServerUri               The address of the xDS server. Can be an IP address or a domain with multiple
+   *                                   underlying A/AAAA records.
+   * @param defaultLoadBalancingPolicy If provided, changes the default load balancing policy on the builder to the
+   *                                   given policy (see
+   *                                   {@link io.grpc.ManagedChannelBuilder#defaultLoadBalancingPolicy(String)}).
+   * @see <a href="https://daniel.haxx.se/blog/2012/01/03/getaddrinfo-with-round-robin-dns-and-happy-eyeballs/"/>
+   * Details on IPv6 routing.
+   */
+  public XdsChannelFactory(
+      @Nullable SslContext sslContext,
+      String xdsServerUri,
+      @Nullable String defaultLoadBalancingPolicy)
+  {
     _sslContext = sslContext;
     _xdsServerUri = xdsServerUri;
+    _defaultLoadBalancingPolicy = defaultLoadBalancingPolicy;
   }
 
   public ManagedChannel createChannel()
@@ -46,13 +75,22 @@ public ManagedChannel createChannel()
     }
 
     NettyChannelBuilder builder = NettyChannelBuilder.forTarget(_xdsServerUri);
+    if (_defaultLoadBalancingPolicy != null)
+    {
+      _log.info("Applying custom load balancing policy for xDS channel: {}", _defaultLoadBalancingPolicy);
+      builder = builder.defaultLoadBalancingPolicy(_defaultLoadBalancingPolicy);
+    }
 
-    if (_sslContext != null) {
+    if (_sslContext != null)
+    {
       builder.sslContext(_sslContext);
-    } else {
+    }
+    else
+    {
       builder.usePlaintext();
     }
 
+
     return builder.keepAliveTime(5, TimeUnit.MINUTES)
         // No proxy wanted here; the default proxy detector can mistakenly detect forwarded ports as proxies.
         .proxyDetector(GrpcUtil.NOOP_PROXY_DETECTOR)

gradle.properties

@@ -1,4 +1,4 @@
-version=29.49.8
+version=29.49.9
 group=com.linkedin.pegasus
 org.gradle.configureondemand=true
 org.gradle.parallel=true