Make HTTP/2 channel creation timeout configurable (#1162)

* Make HTTP/2 parent channel creation timeout configurable

The HTTP/2 parent channel creation timeout was hardcoded at 10,000ms
(with a TODO to expose via cfg2). When a downstream host becomes
unreachable, all concurrent requests to that host queue behind the
parent channel bootstrap and block for the full 10s timeout before
failing — destroying p95 latency for callers.

This change threads a configurable channelCreationTimeoutMs through:
  HttpClientFactory.Builder → ChannelPoolManagerFactoryImpl →
  Http2ChannelPoolFactory → Http2ChannelLifecycle

Services can now tune this via HttpClientFactory.Builder
  .setHttp2ChannelCreationTimeout(3000) to fail faster and let D2's
degrader remove flaky hosts sooner.

Default remains 10,000ms for backwards compatibility. All existing
constructors are preserved as @deprecated overloads.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Address review: keep Constructor A delegation unchanged

Constructor A (without udsAddress) keeps delegating to Constructor B
(with udsAddress=null) as before — B now delegates to C with the
default timeout, so changing A's delegation was redundant.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Fix CI: Http2ChannelLifecycle is package-private

Http2ChannelLifecycle has package-private visibility so it cannot be
imported from HttpClientFactory or ChannelPoolManagerFactoryImpl which
are in different packages. Define the default value (10000ms) directly
in HttpClientFactory and reference it from ChannelPoolManagerFactoryImpl.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* Add multipart-mime to conditional testing modules

The multipart-mime module has flaky tests (TestMIMEReaderDrain) that are
unrelated to most PRs and cause spurious CI failures. Adding it to
CONDITIONAL_TESTING_MODULES ensures its tests are only run when its
dependencies (r2-core, data) are actually touched.

* Revert "Add multipart-mime to conditional testing modules"

This reverts commit e4e1ed1.

* Add CHANGELOG entry for HTTP/2 timeout config

Record the new configurable channel creation timeout under the
Unreleased heading, per the Pegasus Git Workflow.

* Bump version to 29.85.13 and roll changelog

Bump gradle.properties to 29.85.13 (patch) and run
./scripts/update-changelog to move the HTTP/2 timeout
entry under the new version heading.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: kt2205

CHANGELOG.md

@@ -14,6 +14,11 @@ and what APIs have changed, if applicable.
 
 ## [Unreleased]
 
+## [29.85.13] - 2026-06-04
+- Make the HTTP/2 parent channel creation timeout configurable via
+  `HttpClientFactory.Builder.setHttp2ChannelCreationTimeout(...)`, allowing callers to
+  fail faster on unreachable hosts (default unchanged at 10,000ms).
+
 ## [29.85.12] - 2026-05-14
 - Avoid per-request HashSet allocation in RelativeLoadBalancerStrategy
 - Make `recordRequestSizeBytes` a `default` no-op method on `XdsClientOtelMetricsProvider` to restore binary compatibility with older implementations (e.g., container <= 38.31.0) that predate the method.
@@ -5994,7 +5999,8 @@ patch operations can re-use these classes for generating patch messages.
 
 ## [0.14.1]
 
-[Unreleased]: https://github.com/linkedin/rest.li/compare/v29.85.12...master
+[Unreleased]: https://github.com/linkedin/rest.li/compare/v29.85.13...master
+[29.85.13]: https://github.com/linkedin/rest.li/compare/v29.85.12...v29.85.13
 [29.85.12]: https://github.com/linkedin/rest.li/compare/v29.85.11...v29.85.12
 [29.85.11]: https://github.com/linkedin/rest.li/compare/v29.85.10...29.85.11
 [29.85.10]: https://github.com/linkedin/rest.li/compare/v29.85.9...v29.85.10

gradle.properties

@@ -1,4 +1,4 @@
-version=29.85.12
+version=29.85.13
 group=com.linkedin.pegasus
 org.gradle.configureondemand=true
 org.gradle.parallel=true

r2-netty/src/main/java/com/linkedin/r2/netty/client/http2/Http2ChannelLifecycle.java

@@ -76,7 +76,8 @@ class Http2ChannelLifecycle implements AsyncPool.Lifecycle<Channel>
   private long _lastActiveTime;
 
   Http2ChannelLifecycle(SocketAddress address, ScheduledExecutorService scheduler, Clock clock,
-      ChannelGroup channelGroup, boolean ssl, long maxContentLength, long idleTimeout, AsyncPool.Lifecycle<Channel> parentChannelLifecycle)
+      ChannelGroup channelGroup, boolean ssl, long maxContentLength, long idleTimeout,
+      AsyncPool.Lifecycle<Channel> parentChannelLifecycle, long channelCreationTimeoutMs)
   {
     _address = address;
     _scheduler = scheduler;
@@ -87,7 +88,7 @@ class Http2ChannelLifecycle implements AsyncPool.Lifecycle<Channel>
     _idleTimeout = idleTimeout;
     _parentChannelLifecycle = parentChannelLifecycle;
     _childChannelCount = 0;
-    _channelCreationTimeoutMs = DEFAULT_CHANNEL_CREATION_TIMEOUT_MS; // TODO: expose this through cfg2
+    _channelCreationTimeoutMs = channelCreationTimeoutMs > 0 ? channelCreationTimeoutMs : DEFAULT_CHANNEL_CREATION_TIMEOUT_MS;
 
     _lastActiveTime = _clock.currentTimeMillis();
     _scheduler.scheduleAtFixedRate(this::closeParentIfIdle, idleTimeout, idleTimeout, TimeUnit.MILLISECONDS);

r2-netty/src/main/java/com/linkedin/r2/netty/client/http2/Http2ChannelPoolFactory.java

@@ -51,6 +51,7 @@ public class Http2ChannelPoolFactory implements ChannelPoolFactory
   private final int _minPoolSize;
   private final boolean _tcpNoDelay;
   private final boolean _ssl;
+  private final long _channelCreationTimeoutMs;
   private final Bootstrap _bootstrap;
   private final ChannelGroup _allChannels;
   private final ScheduledExecutorService _scheduler;
@@ -81,6 +82,33 @@ public Http2ChannelPoolFactory(
         enableSSLSessionResumption, connectTimeout, sslHandShakeTimeout, null);
   }
 
+  @Deprecated
+  public Http2ChannelPoolFactory(
+      ScheduledExecutorService scheduler,
+      EventLoopGroup eventLoopGroup,
+      ChannelGroup channelGroup,
+      AsyncPoolImpl.Strategy strategy,
+      SSLContext sslContext,
+      SSLParameters sslParameters,
+      int maxPoolSize,
+      int minPoolSize,
+      int maxPoolWaiterSize,
+      int maxInitialLineLength,
+      int maxHeaderSize,
+      int maxChunkSize,
+      long idleTimeout,
+      long maxContentLength,
+      boolean tcpNoDelay,
+      boolean enableSSLSessionResumption,
+      int connectTimeout,
+      int sslHandShakeTimeout,
+      String udsAddress) {
+    this(scheduler, eventLoopGroup, channelGroup, strategy, sslContext, sslParameters, maxPoolSize, minPoolSize,
+        maxPoolWaiterSize, maxInitialLineLength, maxHeaderSize, maxChunkSize, idleTimeout, maxContentLength, tcpNoDelay,
+        enableSSLSessionResumption, connectTimeout, sslHandShakeTimeout, udsAddress,
+        Http2ChannelLifecycle.DEFAULT_CHANNEL_CREATION_TIMEOUT_MS);
+  }
+
   public Http2ChannelPoolFactory(
       ScheduledExecutorService scheduler,
       EventLoopGroup eventLoopGroup,
@@ -100,7 +128,8 @@ public Http2ChannelPoolFactory(
       boolean enableSSLSessionResumption,
       int connectTimeout,
       int sslHandShakeTimeout,
-      String udsAddress)
+      String udsAddress,
+      long channelCreationTimeoutMs)
   {
     final ChannelInitializer<Channel> initializer = new Http2ChannelInitializer(
         sslContext, sslParameters, maxInitialLineLength, maxHeaderSize, maxChunkSize, maxContentLength,
@@ -115,6 +144,7 @@ public Http2ChannelPoolFactory(
     _idleTimeout = idleTimeout;
     _maxContentLength = maxContentLength;
     _tcpNoDelay = tcpNoDelay;
+    _channelCreationTimeoutMs = channelCreationTimeoutMs;
 
     Bootstrap bootstrap = !StringUtils.isEmpty(udsAddress) ?
         new Bootstrap().channel(getDomainSocketClass()) : new Bootstrap().channel(NioSocketChannel.class);
@@ -144,7 +174,8 @@ public AsyncPool<Channel> getPool(SocketAddress address)
                 _bootstrap,
                 _allChannels,
                 _tcpNoDelay
-            )),
+            ),
+            _channelCreationTimeoutMs),
         _maxPoolSize,
         _idleTimeout,
         _scheduler,

r2-netty/src/main/java/com/linkedin/r2/transport/http/client/HttpClientFactory.java

@@ -159,6 +159,7 @@ public class HttpClientFactory implements TransportClientFactory
   public static final int DEFAULT_CONNECT_TIMEOUT = 30000;
   public static final int DEFAULT_SSL_HANDSHAKE_TIMEOUT = 10000;
   public static final int DEFAULT_CHANNELPOOL_WAITER_TIMEOUT = Integer.MAX_VALUE;
+  public static final long DEFAULT_HTTP2_CHANNEL_CREATION_TIMEOUT_MS = 10000;
   public static final double DEFAULT_MAX_CLIENT_REQUEST_RETRY_RATIO = 0.2;
   public static final double UNLIMITED_CLIENT_REQUEST_RETRY_RATIO = 1.0;
   /**
@@ -203,6 +204,7 @@ public class HttpClientFactory implements TransportClientFactory
   private final int _connectTimeout;
   private final int _sslHandShakeTimeout;
   private final int _channelPoolWaiterTimeout;
+  private final long _http2ChannelCreationTimeoutMs;
   private final String _udsAddress;
   /** Request compression config for each http service. */
   private final Map<String, CompressionConfig> _requestCompressionConfigs;
@@ -657,7 +659,8 @@ private HttpClientFactory(FilterChain filters,
         shutdownCallbackExecutor, jmxManager, requestCompressionThresholdDefault, requestCompressionConfigs,
         responseCompressionConfigs, compressionExecutor, defaultHttpVersion, shareConnection, eventProviderRegistry,
         enableSSLSessionResumption, usePipelineV2, executorsToShutDown, DEFAULT_CONNECT_TIMEOUT,
-        DEFAULT_SSL_HANDSHAKE_TIMEOUT, DEFAULT_CHANNELPOOL_WAITER_TIMEOUT, udsAddress, null);
+        DEFAULT_SSL_HANDSHAKE_TIMEOUT, DEFAULT_CHANNELPOOL_WAITER_TIMEOUT, udsAddress, null,
+        DEFAULT_HTTP2_CHANNEL_CREATION_TIMEOUT_MS);
   }
 
   private HttpClientFactory(FilterChain filters,
@@ -682,7 +685,8 @@ private HttpClientFactory(FilterChain filters,
                             int sslHandShakeTimeout,
                             int channelPoolWaiterTimeout,
                             String udsAddress,
-                            DnsMetricsCallback dnsMetricsCallback)
+                            DnsMetricsCallback dnsMetricsCallback,
+                            long http2ChannelCreationTimeoutMs)
   {
     _filters = filters;
     _eventLoopGroup = eventLoopGroup;
@@ -698,6 +702,7 @@ private HttpClientFactory(FilterChain filters,
     _connectTimeout = connectTimeout;
     _sslHandShakeTimeout = sslHandShakeTimeout;
     _channelPoolWaiterTimeout = channelPoolWaiterTimeout;
+    _http2ChannelCreationTimeoutMs = http2ChannelCreationTimeoutMs;
     _udsAddress = udsAddress;
     _dnsMetricsCallback = dnsMetricsCallback;
     if (requestCompressionConfigs == null)
@@ -715,7 +720,7 @@ private HttpClientFactory(FilterChain filters,
     _defaultHttpVersion = defaultHttpVersion;
     _channelPoolManagerFactory = new ChannelPoolManagerFactoryImpl(
         _eventLoopGroup, _executor, enableSSLSessionResumption,_usePipelineV2, _channelPoolWaiterTimeout,
-        _connectTimeout, _sslHandShakeTimeout);
+        _connectTimeout, _sslHandShakeTimeout, _http2ChannelCreationTimeoutMs);
 
     if (eventProviderRegistry != null)
     {
@@ -760,6 +765,7 @@ public static class Builder
     private int _connectTimeout = DEFAULT_CONNECT_TIMEOUT;
     private int _sslHandShakeTimeout = DEFAULT_SSL_HANDSHAKE_TIMEOUT;
     private int _channelPoolWaiterTimeout = DEFAULT_CHANNELPOOL_WAITER_TIMEOUT;
+    private long _http2ChannelCreationTimeoutMs = DEFAULT_HTTP2_CHANNEL_CREATION_TIMEOUT_MS;
     private DnsMetricsCallback _dnsMetricsCallback;
 
     /**
@@ -931,6 +937,20 @@ public Builder setSslHandShakeTimeout(int sslHandShakeTimeout)
       return this;
     }
 
+    /**
+     * Sets the timeout for HTTP/2 parent channel creation. When establishing a new HTTP/2
+     * connection, if the TCP + TLS handshake does not complete within this timeout, an
+     * {@link ObjectCreationTimeoutException} is thrown. Lower values cause faster fail-over
+     * to healthy hosts but may trigger false positives on slow networks.
+     *
+     * @param http2ChannelCreationTimeoutMs timeout in milliseconds (default: 10000)
+     */
+    public Builder setHttp2ChannelCreationTimeout(long http2ChannelCreationTimeoutMs)
+    {
+      _http2ChannelCreationTimeoutMs = http2ChannelCreationTimeoutMs;
+      return this;
+    }
+
     public Builder setChannelPoolWaiterTimeout(int channelPoolWaiterTimeout)
     {
       _channelPoolWaiterTimeout = channelPoolWaiterTimeout;
@@ -1007,7 +1027,7 @@ public HttpClientFactory build()
         _requestCompressionThresholdDefault, _requestCompressionConfigs, _responseCompressionConfigs,
         compressionExecutor, _defaultHttpVersion, _shareConnection, eventProviderRegistry, _enableSSLSessionResumption,
           _usePipelineV2, executorsToShutDown, _connectTimeout, _sslHandShakeTimeout, _channelPoolWaiterTimeout,
-          _udsAddress, _dnsMetricsCallback);
+          _udsAddress, _dnsMetricsCallback, _http2ChannelCreationTimeoutMs);
     }
 
   }

r2-netty/src/main/java/com/linkedin/r2/transport/http/client/common/ChannelPoolManagerFactoryImpl.java

@@ -20,6 +20,7 @@
 import com.linkedin.common.util.None;
 import com.linkedin.r2.netty.client.http.HttpChannelPoolFactory;
 import com.linkedin.r2.netty.client.http2.Http2ChannelPoolFactory;
+import com.linkedin.r2.transport.http.client.HttpClientFactory;
 import com.linkedin.r2.transport.http.client.rest.HttpNettyChannelPoolFactory;
 import com.linkedin.r2.transport.http.client.stream.http.HttpNettyStreamChannelPoolFactory;
 import com.linkedin.r2.transport.http.client.stream.http2.Http2NettyStreamChannelPoolFactory;
@@ -56,6 +57,7 @@ public class ChannelPoolManagerFactoryImpl implements ChannelPoolManagerFactory
   private final int _channelPoolWaiterTimeout;
   private final int _connectTimeout;
   private final int _sslHandShakeTimeout;
+  private final long _http2ChannelCreationTimeoutMs;
 
   /**
    * @param eventLoopGroup The EventLoopGroup; it is the caller's responsibility to shut
@@ -65,9 +67,19 @@ public class ChannelPoolManagerFactoryImpl implements ChannelPoolManagerFactory
    * @param enableSSLSessionResumption Enable reuse of Ssl Session.
    * @param usePipelineV2 Use unified new code.
    */
+  @Deprecated
   public ChannelPoolManagerFactoryImpl(EventLoopGroup eventLoopGroup, ScheduledExecutorService scheduler,
       boolean enableSSLSessionResumption, boolean usePipelineV2, int channelPoolWaiterTimeout,
       int connectTimeout, int sslHandShakeTimeout)
+  {
+    this(eventLoopGroup, scheduler, enableSSLSessionResumption, usePipelineV2, channelPoolWaiterTimeout,
+        connectTimeout, sslHandShakeTimeout,
+        HttpClientFactory.DEFAULT_HTTP2_CHANNEL_CREATION_TIMEOUT_MS);
+  }
+
+  public ChannelPoolManagerFactoryImpl(EventLoopGroup eventLoopGroup, ScheduledExecutorService scheduler,
+      boolean enableSSLSessionResumption, boolean usePipelineV2, int channelPoolWaiterTimeout,
+      int connectTimeout, int sslHandShakeTimeout, long http2ChannelCreationTimeoutMs)
   {
     _eventLoopGroup = eventLoopGroup;
     _scheduler = scheduler;
@@ -76,6 +88,7 @@ public ChannelPoolManagerFactoryImpl(EventLoopGroup eventLoopGroup, ScheduledExe
     _channelPoolWaiterTimeout = channelPoolWaiterTimeout;
     _connectTimeout = connectTimeout;
     _sslHandShakeTimeout = sslHandShakeTimeout;
+    _http2ChannelCreationTimeoutMs = http2ChannelCreationTimeoutMs;
   }
 
   @Override
@@ -204,7 +217,8 @@ public ChannelPoolManager buildHttp2Stream(ChannelPoolManagerKey channelPoolMana
           _enableSSLSessionResumption,
           _connectTimeout,
           _sslHandShakeTimeout,
-          channelPoolManagerKey.getUdsAddress());
+          channelPoolManagerKey.getUdsAddress(),
+          _http2ChannelCreationTimeoutMs);
     }
     else
     {