[SPARK-4563][CORE] Allow driver to advertise a different network address.

The goal of this feature is to allow the Spark driver to run in an
isolated environment, such as a docker container, and be able to use
the host's port forwarding mechanism to be able to accept connections
from the outside world.

The change is restricted to the driver: there is no support for achieving
the same thing on executors (or the YARN AM for that matter). Those still
need full access to the outside world so that, for example, connections
can be made to an executor's block manager.

The core of the change is simple: add a new configuration that tells what's
the address the driver should bind to, which can be different than the address
it advertises to executors (spark.driver.host). Everything else is plumbing
the new configuration where it's needed.

To use the feature, the host starting the container needs to set up the
driver's port range to fall into a range that is being forwarded; this
required the block manager port to need a special configuration just for
the driver, which falls back to the existing spark.blockManager.port when
not set. This way, users can modify the driver settings without affecting
the executors; it would theoretically be nice to also have different
retry counts for driver and executors, but given that docker (at least)
allows forwarding port ranges, we can probably live without that for now.

Because of the nature of the feature it's kinda hard to add unit tests;
I just added a simple one to make sure the configuration works.

This was tested with a docker image running spark-shell with the following
command:

 docker blah blah blah \
   -p 38000-38100:38000-38100 \
   [image] \
   spark-shell \
     --num-executors 3 \
     --conf spark.shuffle.service.enabled=false \
     --conf spark.dynamicAllocation.enabled=false \
     --conf spark.driver.host=[host's address] \
     --conf spark.driver.port=38000 \
     --conf spark.driver.blockManager.port=38020 \
     --conf spark.ui.port=38040

Running on YARN; verified the driver works, executors start up and listen
on ephemeral ports (instead of using the driver's config), and that caching
and shuffling (without the shuffle service) works. Clicked through the UI
to make sure all pages (including executor thread dumps) worked. Also tested
apps without docker, and ran unit tests.

Author: Marcelo Vanzin <vanzin@cloudera.com>

Closes #15120 from vanzin/SPARK-4563.

5 files changed, 24 insertions, 9 deletions

diff --git a/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferSecuritySuite.scala b/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferSecuritySuite.scala
index ed15e77ff1..022fe91eda 100644
--- a/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferSecuritySuite.scala
+++ b/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferSecuritySuite.scala
@@ -108,11 +108,13 @@ class NettyBlockTransferSecuritySuite extends SparkFunSuite with MockitoSugar wi
     when(blockManager.getBlockData(blockId)).thenReturn(blockBuffer)
 
     val securityManager0 = new SecurityManager(conf0)
-    val exec0 = new NettyBlockTransferService(conf0, securityManager0, "localhost", numCores = 1)
+    val exec0 = new NettyBlockTransferService(conf0, securityManager0, "localhost", "localhost", 0,
+      1)
     exec0.init(blockManager)
 
     val securityManager1 = new SecurityManager(conf1)
-    val exec1 = new NettyBlockTransferService(conf1, securityManager1, "localhost", numCores = 1)
+    val exec1 = new NettyBlockTransferService(conf1, securityManager1, "localhost", "localhost", 0,
+      1)
     exec1.init(blockManager)
 
     val result = fetchBlock(exec0, exec1, "1", blockId) match {
diff --git a/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferServiceSuite.scala b/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferServiceSuite.scala
index e7df7cb419..121447a965 100644
--- a/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferServiceSuite.scala
+++ b/core/src/test/scala/org/apache/spark/network/netty/NettyBlockTransferServiceSuite.scala
@@ -23,6 +23,7 @@ import org.mockito.Mockito.mock
 import org.scalatest._
 
 import org.apache.spark.{SecurityManager, SparkConf, SparkFunSuite}
+import org.apache.spark.internal.config._
 import org.apache.spark.network.BlockDataManager
 
 class NettyBlockTransferServiceSuite
@@ -86,10 +87,10 @@ class NettyBlockTransferServiceSuite
   private def createService(port: Int): NettyBlockTransferService = {
     val conf = new SparkConf()
       .set("spark.app.id", s"test-${getClass.getName}")
-      .set("spark.blockManager.port", port.toString)
     val securityManager = new SecurityManager(conf)
     val blockDataManager = mock(classOf[BlockDataManager])
-    val service = new NettyBlockTransferService(conf, securityManager, "localhost", numCores = 1)
+    val service = new NettyBlockTransferService(conf, securityManager, "localhost", "localhost",
+      port, 1)
     service.init(blockDataManager)
     service
   }
diff --git a/core/src/test/scala/org/apache/spark/rpc/netty/NettyRpcEnvSuite.scala b/core/src/test/scala/org/apache/spark/rpc/netty/NettyRpcEnvSuite.scala
index 2d6543d328..0409aa3a5d 100644
--- a/core/src/test/scala/org/apache/spark/rpc/netty/NettyRpcEnvSuite.scala
+++ b/core/src/test/scala/org/apache/spark/rpc/netty/NettyRpcEnvSuite.scala
@@ -27,8 +27,8 @@ class NettyRpcEnvSuite extends RpcEnvSuite {
       name: String,
       port: Int,
       clientMode: Boolean = false): RpcEnv = {
-    val config = RpcEnvConfig(conf, "test", "localhost", port, new SecurityManager(conf),
-      clientMode)
+    val config = RpcEnvConfig(conf, "test", "localhost", "localhost", port,
+      new SecurityManager(conf), clientMode)
     new NettyRpcEnvFactory().create(config)
   }
 
@@ -41,4 +41,16 @@ class NettyRpcEnvSuite extends RpcEnvSuite {
     assert(e.getCause.getMessage.contains(uri))
   }
 
+  test("advertise address different from bind address") {
+    val sparkConf = new SparkConf()
+    val config = RpcEnvConfig(sparkConf, "test", "localhost", "example.com", 0,
+      new SecurityManager(sparkConf), false)
+    val env = new NettyRpcEnvFactory().create(config)
+    try {
+      assert(env.address.hostPort.startsWith("example.com:"))
+    } finally {
+      env.shutdown()
+    }
+  }
+
 }
diff --git a/core/src/test/scala/org/apache/spark/storage/BlockManagerReplicationSuite.scala b/core/src/test/scala/org/apache/spark/storage/BlockManagerReplicationSuite.scala
index b9e3a364ee..e1c1787cbd 100644
--- a/core/src/test/scala/org/apache/spark/storage/BlockManagerReplicationSuite.scala
+++ b/core/src/test/scala/org/apache/spark/storage/BlockManagerReplicationSuite.scala
@@ -67,7 +67,7 @@ class BlockManagerReplicationSuite extends SparkFunSuite
       name: String = SparkContext.DRIVER_IDENTIFIER): BlockManager = {
     conf.set("spark.testing.memory", maxMem.toString)
     conf.set("spark.memory.offHeap.size", maxMem.toString)
-    val transfer = new NettyBlockTransferService(conf, securityMgr, "localhost", numCores = 1)
+    val transfer = new NettyBlockTransferService(conf, securityMgr, "localhost", "localhost", 0, 1)
     val memManager = UnifiedMemoryManager(conf, numCores = 1)
     val serializerManager = new SerializerManager(serializer, conf)
     val store = new BlockManager(name, rpcEnv, master, serializerManager, conf,
diff --git a/core/src/test/scala/org/apache/spark/storage/BlockManagerSuite.scala b/core/src/test/scala/org/apache/spark/storage/BlockManagerSuite.scala
index 6d53d2e5f0..1652fcdb96 100644
--- a/core/src/test/scala/org/apache/spark/storage/BlockManagerSuite.scala
+++ b/core/src/test/scala/org/apache/spark/storage/BlockManagerSuite.scala
@@ -80,7 +80,7 @@ class BlockManagerSuite extends SparkFunSuite with Matchers with BeforeAndAfterE
     conf.set("spark.memory.offHeap.size", maxMem.toString)
     val serializer = new KryoSerializer(conf)
     val transfer = transferService
-      .getOrElse(new NettyBlockTransferService(conf, securityMgr, "localhost", numCores = 1))
+      .getOrElse(new NettyBlockTransferService(conf, securityMgr, "localhost", "localhost", 0, 1))
     val memManager = UnifiedMemoryManager(conf, numCores = 1)
     val serializerManager = new SerializerManager(serializer, conf)
     val blockManager = new BlockManager(name, rpcEnv, master, serializerManager, conf,
@@ -854,7 +854,7 @@ class BlockManagerSuite extends SparkFunSuite with Matchers with BeforeAndAfterE
   test("block store put failure") {
     // Use Java serializer so we can create an unserializable error.
     conf.set("spark.testing.memory", "1200")
-    val transfer = new NettyBlockTransferService(conf, securityMgr, "localhost", numCores = 1)
+    val transfer = new NettyBlockTransferService(conf, securityMgr, "localhost", "localhost", 0, 1)
     val memoryManager = UnifiedMemoryManager(conf, numCores = 1)
     val serializerManager = new SerializerManager(new JavaSerializer(conf), conf)
     store = new BlockManager(SparkContext.DRIVER_IDENTIFIER, rpcEnv, master,