On Non-secure Hadoop Using Derby |

This page shows how to operate Hive on MR3 in a non-secure Hadoop cluster without Kerberos. The same user, not necessarily an administrator of the Hadoop cluster, will run all the components of Hive on MR3, such as Metastore, HiveServer2, and Beeline. For running Metastore, we will use a Derby database which is included in the MR3 release. By following the instruction, the user will learn:

how to install and configure Hive on MR3 in a non-secure Hadoop cluster without Kerberos
how to start and stop Metastore with a Derby database
how to start and stop HiveServer2
how to create Beeline connections and send queries to HiveServer2
how to generate TPC-DS datasets and execute Hive queries in the TPC-DS benchmark
how to run multiple Map/Reduce Tasks concurrently in a single ContainerWorker

This scenario has the following prerequisites:

Java 1.8 is available.
A non-secure Hadoop cluster 2.7 or higher is available.
The user has access to the home directory and /tmp directory on HDFS.
mvn, gcc, and javac should be available in order to generate TPC-DS datasets.

This scenario should take less than 30 minutes to complete, not including the time for downloading an MR3 release, generating a TPC-DS dataset, loading Hive tables, and computing column statistics. This page has been tested with MR3 release 1.0 on Hadoop 2.7 (HDP 2.6.4) and Hadoop 3.1 (HDP 3.0.1).

Installation

Download a pre-built MR3 release and uncompress it. We choose the pre-built MR3 release based on Hive 3.1.2 which corresponds to --hivesrc3 option to be used later.

$ wget https://github.com/mr3project/mr3-release/releases/download/v1.0/hivemr3-1.0-hive3.1.2.tar.gz
$ gunzip -c hivemr3-1.0-hive3.1.2.tar.gz | tar xvf -;

Rename the new directory to mr3-run and change the working directory. Renaming the new directory is not strictly necessary, but it is recommended because the sample configuration file hive-site.xml included in the MR3 release uses the same directory name.

$ mv hivemr3-1.0-hive3.1.2/ mr3-run
$ cd mr3-run/

Configuring

Open env.sh and set JAVA_HOME and PATH if necessary. Set HADOOP_HOME to the Hadoop installation directory.

$ vi env.sh

export JAVA_HOME=/usr/apps/java/default
export PATH=$JAVA_HOME/bin:$PATH
export HADOOP_HOME=/usr/hdp/2.6.4.0-91/hadoop

Set the following environment variables to adjust the memory size (in MB) to be allocated to each component:

HIVE_METASTORE_HEAPSIZE specifies the memory size for Metastore.
HIVE_SERVER2_HEAPSIZE specifies the memory size for HiveServer2.
HIVE_CLIENT_HEAPSIZE specifies the memory size of HiveCLI (hive command) and Beeline (beeline command).
MR3_AM_HEAPSIZE specifies the memory size of MR3 DAGAppMaster.

In our example, we use the following values. We increase the value for HIVE_CLIENT_HEAPSIZE to 16384 (16GB) because we will run hive command when generating TPC-DS datasets.

HIVE_METASTORE_HEAPSIZE=4096
HIVE_SERVER2_HEAPSIZE=16384
HIVE_CLIENT_HEAPSIZE=16384
MR3_AM_HEAPSIZE=10240

HIVE3_HDFS_WAREHOUSE specifies the warehouse directory on HDFS. Update it to use the current user name.

HIVE3_HDFS_WAREHOUSE=/user/gla/warehouse

Open conf/cluster/hive3/hive-site.xml and set the following four configuration keys according to the current user name (instead of the default user hive) and the working directory (instead of the default directory /home/hive).

$ vi conf/cluster/hive3/hive-site.xml 

<property>
  <name>hive.users.in.admin.role</name>
  <value>gla</value>
</property>

<property>
  <name>hive.aux.jars.path</name>
  <value>/home/gla/mr3-run/hive/hivejar/apache-hive-3.1.2-bin/lib/hive-llap-common-3.1.2.jar,/home/gla/mr3-run/hive/hivejar/apache-hive-3.1.2-bin/lib/hive-llap-server-3.1.2.jar,/home/gla/mr3-run/hive/hivejar/apache-hive-3.1.2-bin/lib/hive-llap-tez-3.1.2.jar</value>
</property>

<property>
  <name>hive.exec.scratchdir</name>
  <value>/tmp/gla</value>
</property>

<property>
  <name>hive.server2.logging.operation.log.location</name>
  <value>/tmp/gla/operation_logs</value>
</property>

The following configuration keys specify the resource to be allocated to a Map Task, a Reduce Task, or a ContainerWorker. By default, we allocate 4GB and a single core to a Map Task, a Reduce Task, and a ContainerWorker.

<property>
  <name>hive.mr3.map.task.memory.mb</name>
  <value>4096</value>
</property>

<property>
  <name>hive.mr3.map.task.vcores</name>
  <value>1</value>
</property>

<property>
  <name>hive.mr3.reduce.task.memory.mb</name>
  <value>4096</value>
</property>

<property>
  <name>hive.mr3.reduce.task.vcores</name>
  <value>1</value>
</property>

<property>
  <name>hive.mr3.all-in-one.containergroup.memory.mb</name>
  <value>4096</value>
</property>

<property>
  <name>hive.mr3.all-in-one.containergroup.vcores</name>
  <value>1</value>
</property>

When updating these configuration keys, we should meet the following requirements:

hive.mr3.map.task.memory.mb <= hive.mr3.all-in-one.containergroup.memory.mb
hive.mr3.map.task.vcores <= hive.mr3.all-in-one.containergroup.vcores
hive.mr3.reduce.task.memory.mb <= hive.mr3.all-in-one.containergroup.memory.mb
hive.mr3.reduce.task.vcores <= hive.mr3.all-in-one.containergroup.vcores

For simplicity, disable impersonation by setting hive.server2.enable.doAs to false.

<property>
  <name>hive.server2.enable.doAs</name>
  <value>false</value>
</property>

Creating directories on HDFS

Create the warehouse directory specified in env.sh.

$ grep HIVE3_HDFS_WAREHOUSE env.sh
HIVE3_HDFS_WAREHOUSE=/user/gla/warehouse
$ hdfs dfs -mkdir -p /user/gla/warehouse

Create a directory for storing MR3 and Tez jar files.

$ hdfs dfs -mkdir /user/gla/lib

Load MR3 jar files.

$ mr3/upload-hdfslib-mr3.sh

Load Tez jar files.

$ tez/upload-hdfslib-tez.sh

Make sure that /tmp/gla does NOT exist on HDFS.

$ hdfs dfs -ls /tmp/gla
ls: `/tmp/gla': No such file or directory

If the directory already exists (e.g., when running Hive on MR3 for the second time), make sure that its permission is set to 733. HiveServer2 automatically creates a new directory with permission 733 if it does not exist.

$ hdfs dfs -ls /tmp/ | grep gla
drwx-wx-wx   - gla           hdfs          0 2020-02-03 00:05 /tmp/gla

Creating temporary directories

Create a new directory specified by hive.server2.logging.operation.log.location.

$ ls -alt /tmp/gla/operation_logs
ls: cannot access /tmp/gla/operation_logs: No such file or directory
$ mkdir -p /tmp/gla/operation_logs

Running Metastore

Metastore uses the port specified by the environment variable HIVE3_METASTORE_PORT in env.sh. Make sure that the port is not in use.

$ grep HIVE3_METASTORE_PORT env.sh
HIVE3_METASTORE_PORT=9830
$ ss -anpt | grep -E "LISTEN.*:9830"
$

Run Metastore with a Derby database using --cluster option, and initialize it using --init-schema option.

$ hive/metastore-service.sh start --cluster --hivesrc3 --init-schema

After a while, check if Metastore has successfully started.

$ tail /home/gla/mr3-run/hive/metastore-service-result/hive-mr3--2020-02-02--23-59-55-6775428c/out-metastore.txt 


Initialization script completed
schemaTool completed
2020-02-03 00:00:04: Starting Hive Metastore Server
...

When the user restarts Metastore, do not use --init-schema option in order to reuse existing Hive databases. For example, the user can kill Metastore and restart it as follows. (If the following command fails to kill Metastore for some reason, manually kill with kill -9.)

$ hive/metastore-service.sh stop --cluster --hivesrc3
$ hive/metastore-service.sh start --cluster --hivesrc3

The log file for Metastore is found under /tmp/gla.

$ ls /tmp/gla/hive.log
/tmp/gla/hive.log

Running HiveServer2

Run HiveServer2 using --cluster option. In order to use LocalProcess mode for MR3 DAGAppMaster, use --amprocess option. In our example, we use --amprocess option.

$ hive/hiveserver2-service.sh start --cluster --hivesrc3 --amprocess

After a while, check if HiveServer2 has successfully started by inspecting its log file.

$ grep -e "New MR3Session created" /home/gla/mr3-run/hive/hiveserver2-service-result/hive-mr3--2020-02-03--00-07-12-5c0c29fb/hive-logs/hive.log 
2020-02-03T00:07:31,805  INFO [main] session.MR3SessionManagerImpl: New MR3Session created: 76cc9479-0b74-4d17-a7f2-64d671f803d4, gla

The user can find a new Yarn application of type mr3 submitted by the user.

$ yarn application -list
20/02/03 00:08:06 INFO client.RMProxy: Connecting to ResourceManager at gold0/10.1.90.9:8050
20/02/03 00:08:06 INFO client.AHSProxy: Connecting to Application History server at gold0/10.1.90.9:10200
Total number of applications (application-types: [] and states: [SUBMITTED, ACCEPTED, RUNNING]):1
                Application-Id      Application-Name      Application-Type        User       Queue               State         Final-State         Progress                        Tracking-URL
application_1548163665378_0142  76cc9479-0b74-4d17-a7f2-64d671f803d4                   mr3         gla     default             RUNNING           UNDEFINED               0%                                 N/A

The user can also find the process for MR3 DAGAppMaster.

$ ps -ef | grep DAGAppMaster | grep mr3
gla       6708  6528 10 00:07 pts/2    00:00:11 /usr/jdk64/jdk1.8.0_112/jre/bin/java -server -Djava.net.preferIPv4Stack=true -Dhadoop.metrics.log.level=WARN -XX:+PrintGCDetails -verbose:gc -XX:+PrintGCTimeStamps -XX:+UseNUMA -XX:+UseG1GC -XX:+ResizeTLAB -Xmx8192m -Dlog4j.configurationFile=mr3-container-log4j2.properties -Dlog4j.configuratorClass=com.datamonad.mr3.MR3Log4jConfigurator -Dmr3.root.logger=INFO -Dyarn.app.container.log.dir=/home/gla/mr3-run/hive/hiveserver2-service-result/hive-mr3--2020-02-03--00-07-12-5c0c29fb/application_1548163665378_0142 -Dsun.nio.ch.bugLevel='' com.datamonad.mr3.master.DAGAppMaster --session

The user can find the log file for MR3 DAGAppMaster.

$ ls /home/gla/mr3-run/hive/hiveserver2-service-result/hive-mr3--2020-02-03--00-07-12-5c0c29fb/application_1548163665378_0142/run.log 
/home/gla/mr3-run/hive/hiveserver2-service-result/hive-mr3--2020-02-03--00-07-12-5c0c29fb/application_1548163665378_0142/run.log

Running queries

Download a sample dataset.

$ wget https://github.com/mr3project/mr3-release/releases/download/v1.0/pokemon.csv

Run Beeline using --cluster option.

$ hive/run-beeline.sh --cluster --hivesrc3

Use the default database.

0: jdbc:hive2://gold0:9832/> use default;

Create a table called pokemon.

0: jdbc:hive2://gold0:9832/> CREATE TABLE pokemon (Number Int,Name String,Type1 String,Type2 String,Total Int,HP Int,Attack Int,Defense Int,Sp_Atk Int,Sp_Def Int,Speed Int) row format delimited fields terminated BY ',' lines terminated BY '\n' tblproperties("skip.header.line.count"="1");

Import the sample dataset.

0: jdbc:hive2://gold0:9832/> load data local inpath './pokemon.csv' INTO table pokemon;

Execute queries.

0: jdbc:hive2://gold0:9832/> select avg(HP) from pokemon;
...

0: jdbc:hive2://gold0:9832/> create table pokemon1 as select *, IF(HP>160.0,'strong',IF(HP>140.0,'moderate','weak')) AS power_rate from pokemon;
...

0: jdbc:hive2://gold0:9832/> select COUNT(name), power_rate from pokemon1 group by power_rate;
...

Exit Beeline. The warehouse directory on HDFS has now two sub-directories.

$ hdfs dfs -ls /user/gla/warehouse/
Found 2 items
drwxr-xr-x   - gla hdfs          0 2020-02-03 00:40 /user/gla/warehouse/pokemon
drwxr-xr-x   - gla hdfs          0 2020-02-03 00:41 /user/gla/warehouse/pokemon1

Stop HiveServer2. MR3 DAGAppMaster also stops.

$ hive/hiveserver2-service.sh stop --cluster --hivesrc3

Generating a TPC-DS dataset

Open env.sh and set HIVE_DS_SCALE_FACTOR to specify a scale factor for the TPC-DS benchmark. In our example, we use 100 for a dataset of 100GB.

$ vi env.sh

HIVE_DS_SCALE_FACTOR=100

The script for generating TPC-DS datasets creates sub-directories under /tmp/tpcds-generate/ on HDFS. Make sure that the directory does not exist.

$ hdfs dfs -ls /tmp/tpcds-generate
ls: `/tmp/tpcds-generate': No such file or directory

It it already exists, make sure that the user has write permission so that the script can create sub-directories. If a sub-directory /tmp/tpcds-generate/100 (corresponding to a scale factor of 100) already exists, it should be created by the same script.

Generate a TPC-DS dataset and load Hive tables. Depending on the scale factor and the size of the cluster, it may take up to a few hours.

$ hive/gen-tpcds.sh --cluster --hivesrc3

The raw data is created in the directory /tmp/tpcds-generate/100 on HDFS.

$ hdfs dfs -du -h /tmp/tpcds-generate
96.3 G  /tmp/tpcds-generate/100

The warehouse directory on HDFS now has two new sub-directories. /user/gla/warehouse/tpcds_text_100.db is empty because it corresponds to external Hive tables pointing to the raw data in /tmp/tpcds-generate/100.

$ hdfs dfs -du -h /user/gla/warehouse
44.2 K  /user/gla/warehouse/pokemon
36.9 K  /user/gla/warehouse/pokemon1
30.1 G  /user/gla/warehouse/tpcds_bin_partitioned_orc_100.db
0       /user/gla/warehouse/tpcds_text_100.db

Running TPC-DS queries

Run HiveServer2 using --cluster option as well as --amprocess option.

$ hive/hiveserver2-service.sh start --cluster --hivesrc3 --amprocess

Run Beeline using --cluster option.

$ hive/run-beeline.sh --cluster --hivesrc3

The user can now find two new databases available.

0: jdbc:hive2://gold0:9832/> show databases;
INFO  : Compiling command(queryId=gla_20190507030245_8d1e5560-f4c4-4901-8153-f01759aedfc9): show databases
INFO  : Semantic Analysis Completed (retrial = false)
INFO  : Returning Hive schema: Schema(fieldSchemas:[FieldSchema(name:database_name, type:string, comment:from deserializer)], properties:null)
INFO  : Completed compiling command(queryId=gla_20190507030245_8d1e5560-f4c4-4901-8153-f01759aedfc9); Time taken: 1.487 seconds
INFO  : Executing command(queryId=gla_20190507030245_8d1e5560-f4c4-4901-8153-f01759aedfc9): show databases
INFO  : Starting task [Stage-0:DDL] in serial mode
INFO  : Completed executing command(queryId=gla_20190507030245_8d1e5560-f4c4-4901-8153-f01759aedfc9); Time taken: 0.158 seconds
INFO  : OK
+--------------------------------+
|         database_name          |
+--------------------------------+
| default                        |
| tpcds_bin_partitioned_orc_100  |
| tpcds_text_100                 |
+--------------------------------+
3 rows selected (2.004 seconds)

Use tpcds_bin_partitioned_orc_100.

0: jdbc:hive2://gold0:9832/> use tpcds_bin_partitioned_orc_100;

Execute queries. The complete set of 99 TPC-DS queries is found under hive/benchmarks/hive-testbench/sample-queries-tpcds-hive2. Most of the queries should run okay. There are, however, a few queries that may fail occasionally. As an example, try query 4 (hive/benchmarks/hive-testbench/sample-queries-tpcds-hive2/query4.sql). The execution may fail with java.lang.OutOfMemoryError, even with the mechanism of query re-execution in Hive 3. This is because column statistics of Hive tables are not available yet.

Compute column statistics of Hive tables. Depending on the scale factor and the size of the cluster, it may take a long time (but much less than generating the TPC-DS dataset).

0: jdbc:hive2://gold0:9832/> analyze table call_center compute statistics for columns; analyze table catalog_page compute statistics for columns; analyze table catalog_returns compute statistics for columns; analyze table catalog_sales compute statistics for columns; analyze table customer compute statistics for columns; analyze table customer_address compute statistics for columns; analyze table customer_demographics compute statistics for columns; analyze table date_dim compute statistics for columns; analyze table household_demographics compute statistics for columns; analyze table income_band compute statistics for columns; analyze table inventory compute statistics for columns; analyze table item compute statistics for columns; analyze table promotion compute statistics for columns; analyze table reason compute statistics for columns; analyze table ship_mode compute statistics for columns; analyze table store compute statistics for columns; analyze table store_returns compute statistics for columns; analyze table store_sales compute statistics for columns; analyze table time_dim compute statistics for columns; analyze table warehouse compute statistics for columns; analyze table web_page compute statistics for columns; analyze table web_returns compute statistics for columns; analyze table web_sales compute statistics for columns; analyze table web_site compute statistics for columns;
...

Execute queries. With column statistics available, every query should run more reliably.

Stop HiveServer2.

$ hive/hiveserver2-service.sh stop --cluster --hivesrc3

Running multiple Map/Reduce Tasks concurrently in a single ContainerWorker

With the default configuration, a ContainerWorker can run one Map or Reduce Task at a time. By allocating more resources, however, we can run multiple Map and Reduce Tasks concurrently inside the same ContainerWorker.

Open conf/cluster/hive3/hive-site.xml. Increase the memory size and the number of cores to be allocated to each ContainerWorker. In our example, we increase the memory size to 80GB for each ContainerWorker. Make sure that the configuration key hive.mr3.container.combine.taskattempts is set to true.

$ vi conf/cluster/hive3/hive-site.xml 

<property>
  <name>hive.mr3.container.combine.taskattempts</name>
  <value>true</value>
</property>

<property>
  <name>hive.mr3.all-in-one.containergroup.memory.mb</name>
  <value>81920</value>
</property>

<property>
  <name>hive.mr3.all-in-one.containergroup.vcores</name>
  <value>20</value>
</property>

Run HiveServer2 using --cluster option as well as --amprocess option.

$ hive/hiveserver2-service.sh start --cluster --hivesrc3 --amprocess

Run Beeline using --cluster option.

$ hive/run-beeline.sh --cluster --hivesrc3

Use tpcds_bin_partitioned_orc_100.

0: jdbc:hive2://gold0:9832/> use tpcds_bin_partitioned_orc_100;

Execute queries. The user can check that Yarn has created much fewer containers.

Stopping HiveServer2 and Metastore

Stop HiveServer2.

$ hive/hiveserver2-service.sh stop --cluster --hivesrc3

Stop Metastore.

$ hive/metastore-service.sh stop --cluster --hivesrc3

The user can check if Metastore has successfully stopped by reading its log file.

$ tail -n3 /tmp/gla/hive.log
/************************************************************
SHUTDOWN_MSG: Shutting down HiveMetaStore at gold0/10.1.90.9
************************************************************/