An experimental VoltDB to Hive export conduit that takes advantage of Hive's new streaming hcatalog API. It allows export table writers to push data directly into correspoding Hive tables.

• Install Gradle

On a Mac if you have Homebrew setup then simply install the gradle bottle

brew install gradle

On Linux setup GVM, and install gradle as follows

gvm install gradle

• Create gradle.properties file and set the voltdbhome property to the base directory where your VoltDB is installed

echo voltdbhome=/voltdb/home/dirname > gradle.properties

• Invoke gradle to compile artifacts

gradle shadowJar

• To setup an eclipse project run gradle as follows

gradle cleanEclipse eclipse

then import it into your eclipse workspace by using File->Import projects menu option

• Copy the built jar from build/libs to lib/extension under your VoltDB installation directory

• Edit your deployment file and use the following export XML stanza as a template

<?xml version="1.0"?>
<deployment>
    <cluster hostcount="1" sitesperhost="4" kfactor="0" />
    <httpd enabled="true">
        <jsonapi enabled="true" />
    </httpd>
    <export>
        <configuration stream="default" enabled="true" type="custom"
            exportconnectorclass="org.voltdb.exportclient.hive.HiveExportClient">
            <property name="hive.uri">thrift://hive-host:9083</property>
            <property name="hive.db">meco</property>
            <property name="hive.table">alerts</property>
            <property name="hive.partition.columns">ALERTS:CONTINENT|COUNTRY</property>
        </configuration>
    </export>
</deployment>

This tells VoltDB to write to the alerts table on Hive, via the homonymous export table in VoltDB, using columns CONTINENT and COUNTRY as value providers for Hive partitions discerners. For example the alerts table is defined in Hive as:

create table alerts ( id int , msg string )
     partitioned by (continent string, country string)
     clustered by (id) into 5 buckets
     stored as orc; // currently ORC is required for streaming

while the VoltDB export table is defined as:

FILE -inlinebatch END_OF_EXPORT

create table alerts (
  id integer not null,
  msg varchar(128),
  continent varchar(64),
  country varchar(64)
)
;
partition table alerts on column id
;
export table alerts
;
END_OF_EXPORT

When a row is inserted into the export table

INSERT INTO ALERTS (ID,MSG,CONTINENT,COUNTRY) VALUES (1,'fab-02 inoperable','EU','IT');

The continent ('EU') and country ('IT') column values are used to specify the Hive table partition.

• hive.uri (mandatory) thrift URI to the Hive host
• hive.db (mandatory) Hive database
• hive.table (mandatory) Hive table
• hive.partition.columns (mandatory if the hive table is partitioned) format: table-1:column-1|column-2|...|column-n,table-2:column-1|column-2|...|column-n,...,table-n:column-1|column-2|...|column-n
• timezone (optional, default: local timezone) timezone used to format timestamp values

Partition columns must be of type VARCHAR. Any empty or null partition column values are converted to __VoltDB_unspecified__