[SPARK-20343][BUILD] Avoid Unidoc build only if Hadoop 2.6 is explicitly set in SBT build

## What changes were proposed in this pull request?

This PR proposes two things as below:

- Avoid Unidoc build only if Hadoop 2.6 is explicitly set in SBT build

  Due to a different dependency resolution in SBT & Unidoc by an unknown reason, the documentation build fails on a specific machine & environment in Jenkins but it was unable to reproduce.

  So, this PR just checks an environment variable `AMPLAB_JENKINS_BUILD_PROFILE` that is set in Hadoop 2.6 SBT build against branches on Jenkins, and then disables Unidoc build. **Note that PR builder will still build it with Hadoop 2.6 & SBT.**

  ```
  ========================================================================
  Building Unidoc API Documentation
  ========================================================================
  [info] Building Spark unidoc (w/Hive 1.2.1) using SBT with these arguments:  -Phadoop-2.6 -Pmesos -Pkinesis-asl -Pyarn -Phive-thriftserver -Phive unidoc
  Using /usr/java/jdk1.8.0_60 as default JAVA_HOME.
  ...
  ```

  I checked the environment variables from the logs (first bit) as below:

  - **spark-master-test-sbt-hadoop-2.6** (this one is being failed) - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-master-test-sbt-hadoop-2.6/lastBuild/consoleFull

  ```
  JAVA_HOME=/usr/java/jdk1.8.0_60
  JAVA_7_HOME=/usr/java/jdk1.7.0_79
  SPARK_BRANCH=master
  AMPLAB_JENKINS_BUILD_PROFILE=hadoop2.6   <- I use this variable
  AMPLAB_JENKINS="true"
  ```
  - spark-master-test-sbt-hadoop-2.7 - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-master-test-sbt-hadoop-2.7/lastBuild/consoleFull

  ```
  JAVA_HOME=/usr/java/jdk1.8.0_60
  JAVA_7_HOME=/usr/java/jdk1.7.0_79
  SPARK_BRANCH=master
  AMPLAB_JENKINS_BUILD_PROFILE=hadoop2.7
  AMPLAB_JENKINS="true"
  ```

  - spark-master-test-maven-hadoop-2.6 - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-master-test-maven-hadoop-2.6/lastBuild/consoleFull

  ```
  JAVA_HOME=/usr/java/jdk1.8.0_60
  JAVA_7_HOME=/usr/java/jdk1.7.0_79
  HADOOP_PROFILE=hadoop-2.6
  HADOOP_VERSION=
  SPARK_BRANCH=master
  AMPLAB_JENKINS="true"
  ```

  - spark-master-test-maven-hadoop-2.7 - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-master-test-maven-hadoop-2.7/lastBuild/consoleFull

  ```
  JAVA_HOME=/usr/java/jdk1.8.0_60
  JAVA_7_HOME=/usr/java/jdk1.7.0_79
  HADOOP_PROFILE=hadoop-2.7
  HADOOP_VERSION=
  SPARK_BRANCH=master
  AMPLAB_JENKINS="true"
  ```

  - PR builder - https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/75843/consoleFull

  ```
  JENKINS_MASTER_HOSTNAME=amp-jenkins-master
  JAVA_HOME=/usr/java/jdk1.8.0_60
  JAVA_7_HOME=/usr/java/jdk1.7.0_79
  ```

  Assuming from other logs in branch-2.1

    - SBT & Hadoop 2.6 against branch-2.1 https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-branch-2.1-test-sbt-hadoop-2.6/lastBuild/consoleFull

      ```
      JAVA_HOME=/usr/java/jdk1.8.0_60
      JAVA_7_HOME=/usr/java/jdk1.7.0_79
      SPARK_BRANCH=branch-2.1
      AMPLAB_JENKINS_BUILD_PROFILE=hadoop2.6
      AMPLAB_JENKINS="true"
      ```

    - Maven & Hadoop 2.6 against branch-2.1 https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-branch-2.1-test-maven-hadoop-2.6/lastBuild/consoleFull

      ```
      JAVA_HOME=/usr/java/jdk1.8.0_60
      JAVA_7_HOME=/usr/java/jdk1.7.0_79
      HADOOP_PROFILE=hadoop-2.6
      HADOOP_VERSION=
      SPARK_BRANCH=branch-2.1
      AMPLAB_JENKINS="true"
      ```

  We have been using the same convention for those variables. These are actually being used in `run-tests.py` script - here https://github.com/apache/spark/blob/master/dev/run-tests.py#L519-L520

- Revert the previous try

  After https://github.com/apache/spark/pull/17651, it seems the build still fails on SBT Hadoop 2.6 master.

  I am unable to reproduce this - https://github.com/apache/spark/pull/17477#issuecomment-294094092 and the reviewer was too. So, this got merged as it looks the only way to verify this is to merge it currently (as no one seems able to reproduce this).

## How was this patch tested?

I only checked `is_hadoop_version_2_6 = os.environ.get("AMPLAB_JENKINS_BUILD_PROFILE") == "hadoop2.6"` is working fine as expected as below:

```python
>>> import collections
>>> os = collections.namedtuple('os', 'environ')(environ={"AMPLAB_JENKINS_BUILD_PROFILE": "hadoop2.6"})
>>> print(not os.environ.get("AMPLAB_JENKINS_BUILD_PROFILE") == "hadoop2.6")
False
>>> os = collections.namedtuple('os', 'environ')(environ={"AMPLAB_JENKINS_BUILD_PROFILE": "hadoop2.7"})
>>> print(not os.environ.get("AMPLAB_JENKINS_BUILD_PROFILE") == "hadoop2.6")
True
>>> os = collections.namedtuple('os', 'environ')(environ={})
>>> print(not os.environ.get("AMPLAB_JENKINS_BUILD_PROFILE") == "hadoop2.6")
True
```

I tried many ways but I was unable to reproduce this in my local. Sean also tried the way I did but he was also unable to reproduce this.

Please refer the comments in https://github.com/apache/spark/pull/17477#issuecomment-294094092

Author: hyukjinkwon <gurwls223@gmail.com>

Closes #17669 from HyukjinKwon/revert-SPARK-20343.

3 files changed, 12 insertions, 15 deletions

diff --git a/dev/run-tests.py b/dev/run-tests.py
index 450b68123e..818a0c9f48 100755
--- a/dev/run-tests.py
+++ b/dev/run-tests.py
@@ -365,8 +365,16 @@ def build_spark_assembly_sbt(hadoop_version):
     print("[info] Building Spark assembly (w/Hive 1.2.1) using SBT with these arguments: ",
           " ".join(profiles_and_goals))
     exec_sbt(profiles_and_goals)
-    # Make sure that Java and Scala API documentation can be generated
-    build_spark_unidoc_sbt(hadoop_version)
+
+    # Note that we skip Unidoc build only if Hadoop 2.6 is explicitly set in this SBT build.
+    # Due to a different dependency resolution in SBT & Unidoc by an unknown reason, the
+    # documentation build fails on a specific machine & environment in Jenkins but it was unable
+    # to reproduce. Please see SPARK-20343. This is a band-aid fix that should be removed in
+    # the future.
+    is_hadoop_version_2_6 = os.environ.get("AMPLAB_JENKINS_BUILD_PROFILE") == "hadoop2.6"
+    if not is_hadoop_version_2_6:
+        # Make sure that Java and Scala API documentation can be generated
+        build_spark_unidoc_sbt(hadoop_version)
 
 
 def build_apache_spark(build_tool, hadoop_version):
diff --git a/pom.xml b/pom.xml
index 14370d92a9..c1174593c1 100644
--- a/pom.xml
+++ b/pom.xml
@@ -142,7 +142,6 @@
     <ivy.version>2.4.0</ivy.version>
     <oro.version>2.0.8</oro.version>
     <codahale.metrics.version>3.1.2</codahale.metrics.version>
-    <!-- Keep consistent with Avro vesion in SBT build for SPARK-20343 -->
     <avro.version>1.7.7</avro.version>
     <avro.mapred.classifier>hadoop2</avro.mapred.classifier>
     <jets3t.version>0.9.3</jets3t.version>
diff --git a/project/SparkBuild.scala b/project/SparkBuild.scala
index 77dae289f7..e52baf51ae 100644
--- a/project/SparkBuild.scala
+++ b/project/SparkBuild.scala
@@ -318,8 +318,8 @@ object SparkBuild extends PomBuild {
     enable(MimaBuild.mimaSettings(sparkHome, x))(x)
   }
 
-  /* Generate and pick the spark build info from extra-resources and override a dependency */
-  enable(Core.settings ++ CoreDependencyOverrides.settings)(core)
+  /* Generate and pick the spark build info from extra-resources */
+  enable(Core.settings)(core)
 
   /* Unsafe settings */
   enable(Unsafe.settings)(unsafe)
@@ -444,16 +444,6 @@ object DockerIntegrationTests {
 }
 
 /**
- * Overrides to work around sbt's dependency resolution being different from Maven's in Unidoc.
- *
- * Note that, this is a hack that should be removed in the future. See SPARK-20343
- */
-object CoreDependencyOverrides {
-  lazy val settings = Seq(
-    dependencyOverrides += "org.apache.avro" % "avro" % "1.7.7")
-}
-
-/**
  * Overrides to work around sbt's dependency resolution being different from Maven's.
  */
 object DependencyOverrides {