[HAWQ-1504] Namenode hangs during restart of docker environment configured using incubator-hawq/contrib/hawq-docker/

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 2.3.0.0-incubating
• Component/s: Command Line Tools
• Assignee: Radar Da Lei
• Reporter: Shubham Sharma
• Created: 18/Jul/17 01:20
• Updated: 04/Sep/17 17:47
• Resolved: 04/Sep/17 07:04

Description

After setting up an environment using instructions provided under incubator-hawq/contrib/hawq-docker/, while trying to restart docker containers namenode hangs and tries a namenode -format during every start.

Steps to reproduce this issue -

• Navigate to incubator-hawq/contrib/hawq-docker
• make stop
• make start
• docker exec -it centos7-namenode bash
• ps -ef | grep java

You can see namenode -format running.

[gpadmin@centos7-namenode data]$ ps -ef | grep java
hdfs        11    10  1 00:56 ?        00:00:06 /etc/alternatives/java_sdk/bin/java -Dproc_namenode -Xmx1000m -Dhdfs.namenode=centos7-namenode -Dhadoop.log.dir=/var/log/hadoop/hdfs -Dhadoop.log.file=hadoop.log -Dhadoop.home.dir=/usr/hdp/2.5.0.0-1245/hadoop -Dhadoop.id.str= -Dhadoop.root.logger=INFO,console -Djava.library.path=:/usr/hdp/2.5.0.0-1245/hadoop/lib/native/Linux-amd64-64:/usr/hdp/2.5.0.0-1245/hadoop/lib/native -Dhadoop.policy.file=hadoop-policy.xml -Djava.net.preferIPv4Stack=true -Dhadoop.security.logger=INFO,NullAppender org.apache.hadoop.hdfs.server.namenode.NameNode -format

Since namenode -format runs in interactive mode and at this stage it is waiting for a (Yes/No) response, the namenode will remain stuck forever. This makes hdfs unavailable.

Root cause of the problem -

In the dockerfiles present under incubator-hawq/contrib/hawq-docker/centos6-docker/hawq-test and incubator-hawq/contrib/hawq-docker/centos7-docker/hawq-test, the docker directive ENTRYPOINT executes entrypoin.sh during startup.

The entrypoint.sh in turn executes start-hdfs.sh. start-dfs.sh checks for the following -

if [ ! -d /tmp/hdfs/name/current ]; then
    su -l hdfs -c "hdfs namenode -format"
  fi

My assumption is it looks for fsimage and edit logs. If they are not present the script assumes that this a first time initialization and namenode format should be done. However, path /tmp/hdfs/name/current does not exist on namenode.

From namenode logs it is clear that fsimage and edit logs are written under /tmp/hadoop-hdfs/dfs/name/current.

2017-07-18 00:55:20,892 INFO org.apache.hadoop.hdfs.server.namenode.FSImage: No edit log streams selected.
2017-07-18 00:55:20,893 INFO org.apache.hadoop.hdfs.server.namenode.FSImage: Planning to load image: FSImageFile(file=/tmp/hadoop-hdfs/dfs/name/current/fsimage_0000000000000000000, cpktTxId=0000000000000000000)
2017-07-18 00:55:20,995 INFO org.apache.hadoop.hdfs.server.namenode.FSImageFormatPBINode: Loading 1 INodes.
2017-07-18 00:55:21,064 INFO org.apache.hadoop.hdfs.server.namenode.FSImageFormatProtobuf: Loaded FSImage in 0 seconds.
2017-07-18 00:55:21,065 INFO org.apache.hadoop.hdfs.server.namenode.FSImage: Loaded image for txid 0 from /tmp/hadoop-hdfs/dfs/name/current/fsimage_0000000000000000000
2017-07-18 00:55:21,084 INFO org.apache.hadoop.hdfs.server.namenode.FSNamesystem: Need to save fs image? false (staleImage=false, haEnabled=false, isRollingUpgrade=false)
2017-07-18 00:55:21,084 INFO org.apache.hadoop.hdfs.server.namenode.FSEditLog: Starting log segment at 1

Thus wrong path in incubator-hawq/contrib/hawq-docker/centos*-docker/hawq-test/start-hdfs.sh causes namenode to hang during each restart of the containers making hdfs unavailable.

Issue Links

• links to: GitHub Pull Request #1267