[HADOOP-17266] Sudo in hadoop-functions.sh should preserve environment variables

• Type: Bug
• Status: Patch Available
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.3.0
• Component/s: scripts
• Assignee: Unassigned
• Reporter: Chengbing Liu
• Created: 17/Sep/20 03:55
• Updated: 17/Sep/20 14:27

Description

Steps to reproduce:
1. Set HDFS_NAMENODE_USER=hdfs in /etc/default/hadoop-hdfs-namenode to enable user check (and switch to hdfs to start/stop NameNode daemon)
2. Stop NameNode with: service hadoop-hdfs-namenode stop
3. Got an error and NameNode is not stopped

ERROR: Cannot execute /usr/lib/hadoop-hdfs/bin/../libexec/hdfs-config.sh.
Failed to stop Hadoop namenode. Return value: 1. [FAILED]

The root cause is that after sudo, HADOOP_HOME=/usr/lib/hadoop is not preserved, and /usr/lib/hadoop-hdfs/bin/hdfs locates libexec by the following logic:

# let's locate libexec...
if [[ -n "${HADOOP_HOME}" ]]; then
  HADOOP_DEFAULT_LIBEXEC_DIR="${HADOOP_HOME}/libexec"
else
  bin=$(cd -P -- "$(dirname -- "${MYNAME}")" >/dev/null && pwd -P)
  HADOOP_DEFAULT_LIBEXEC_DIR="${bin}/../libexec"
fi

I believe the key point here is that we should preserve environment variables when doing sudo.

Note that this bug is not introduced by HDFS-15353, before which su -l is used, which will also discard environment variables.