diff --git a/dev-support/bin/hadoop.sh b/dev-support/bin/hadoop.sh
index 3343014aae8bb..c92942b86160b 100755
--- a/dev-support/bin/hadoop.sh
+++ b/dev-support/bin/hadoop.sh
@@ -35,6 +35,8 @@ function personality_globals
   JIRA_ISSUE_RE='^(HADOOP|YARN|MAPREDUCE|HDFS)-[0-9]+$'
   #shellcheck disable=SC2034
   GITHUB_REPO_DEFAULT="apache/hadoop"
+  # mount urandom to increase entropy
+  DOCKER_EXTRAARGS=("-v" "/dev/urandom:/dev/random")
 
   HADOOP_HOMEBREW_DIR=${HADOOP_HOMEBREW_DIR:-$(brew --prefix 2>/dev/null)}
   if [[ -z "${HADOOP_HOMEBREW_DIR}" ]]; then
diff --git a/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/mapred/TestMRIntermediateDataEncryption.java b/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/mapred/TestMRIntermediateDataEncryption.java
index fa8dacf6dd507..ce4a6998360c8 100644
--- a/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/mapred/TestMRIntermediateDataEncryption.java
+++ b/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-jobclient/src/test/java/org/apache/hadoop/mapred/TestMRIntermediateDataEncryption.java
@@ -58,12 +58,7 @@
 public class TestMRIntermediateDataEncryption {
   private static final Logger LOG =
       LoggerFactory.getLogger(TestMRIntermediateDataEncryption.class);
-  /**
-   * Use urandom to avoid the YarnChild  process from hanging on low entropy
-   * systems.
-   */
-  private static final String JVM_SECURITY_EGD_OPT =
-      "-Djava.security.egd=file:/dev/./urandom";
+
   // Where MR job's input will reside.
   private static final Path INPUT_DIR = new Path("/test/input");
   // Where output goes.
@@ -115,14 +110,6 @@ public static void setupClass() throws Exception {
     Configuration conf = new Configuration();
     conf.setBoolean(MRJobConfig.MR_ENCRYPTED_INTERMEDIATE_DATA, true);
 
-    // Set the jvm arguments.
-    conf.set(MRJobConfig.MR_AM_ADMIN_COMMAND_OPTS,
-        JVM_SECURITY_EGD_OPT);
-    final String childJVMOpts = JVM_SECURITY_EGD_OPT
-        + " " + conf.get("mapred.child.java.opts", " ");
-    conf.set("mapred.child.java.opts", childJVMOpts);
-
-
     // Start the mini-MR and mini-DFS clusters.
     dfsCluster = new MiniDFSCluster.Builder(conf)
         .numDataNodes(NUM_NODES).build();
diff --git a/start-build-env.sh b/start-build-env.sh
index 2ee44de41aa6e..1d3c94a325ff4 100755
--- a/start-build-env.sh
+++ b/start-build-env.sh
@@ -87,10 +87,20 @@ DOCKER_INTERACTIVE_RUN=${DOCKER_INTERACTIVE_RUN-"-i -t"}
 # within the container and use the result on your normal
 # system.  And this also is a significant speedup in subsequent
 # builds because the dependencies are downloaded only once.
-docker run --rm=true $DOCKER_INTERACTIVE_RUN \
-  -v "${PWD}:${DOCKER_HOME_DIR}/hadoop${V_OPTS:-}" \
-  -w "${DOCKER_HOME_DIR}/hadoop" \
-  -v "${HOME}/.m2:${DOCKER_HOME_DIR}/.m2${V_OPTS:-}" \
-  -v "${HOME}/.gnupg:${DOCKER_HOME_DIR}/.gnupg${V_OPTS:-}" \
-  -u "${USER_ID}" \
+dockerargs=(--rm=true)
+dockerargs+=($DOCKER_INTERACTIVE_RUN)
+# use urandom to increase entropy
+dockerargs+=(-v "/dev/urandom:/dev/random${V_OPTS:-}")
+# mount current directory
+dockerargs+=(-v "${PWD}:${DOCKER_HOME_DIR}/hadoop${V_OPTS:-}")
+# mount maven directory
+dockerargs+=(-v "${HOME}/.m2:${DOCKER_HOME_DIR}/.m2${V_OPTS:-}")
+# mount gnu
+dockerargs+=(-v "${HOME}/.gnupg:${DOCKER_HOME_DIR}/.gnupg${V_OPTS:-}")
+# set work directory
+dockerargs+=(-w "${DOCKER_HOME_DIR}/hadoop")
+# set user
+dockerargs+=(-u "${USER_ID}")
+
+docker run "${dockerargs[@]}" \
   "hadoop-build-${USER_ID}" "$@"