HBase Connector

Overview

The HBase Connector allows querying and creating tables on an external Apache HBase instance. Users can create a table in HBase connector, mapping it to an existing table in HBase Cluster, and support insert, select, or delete.

The HBase Connector maintains a Metastore to persist HBase metadata, currently support Metastore: openLooKeng Metastore.

Note: Apache HBase 2.2.3 version or ealier versions are supported in HBase connector

Connector Configuration

1 . Create and edit etc/catalog/hbase.properties and replacing the xxx as required; connector.name is fixed. If the value of hbase.zookeeper.quorum contains multiple IP addresses, use commas (,) as separators.

connector.name=hbase-connector
hbase.zookeeper.quorum=xxx.xxx.xxx.xxx,xxx.xxx.xxx.xxx
hbase.zookeeper.property.clientPort=xxxx

2 . Use openlookeng metadata storage to store hbase metadata

You must create etc/hetu-metastore.properties to configure the metadata storage platform. The storage platform can be RDBMS or HDFS. Configuration example:

vi etc/hetu-metastore.properties

Option 1: Set hetu-metastore to RDBMS.

hetu.metastore.type=jdbc
hetu.metastore.db.url=jdbc:mysql://xx.xx.xx.xx:3306/dbName....
hetu.metastore.db.user=root
hetu.metastore.db.password=123456

Option 2：Set hetu-metastore to HDFS：

hetu.metastore.type=hetufilesystem
# profile name of hetu file system
hetu.metastore.hetufilesystem.profile-name=hdfs-config-metastore
#the path of metastore storage in the hetu file system
hetu.metastore.hetufilesystem.path=/etc/openlookeng/metastore

If HDFS is used, configure hetu filesystem. The configuration file name must be the same as that of profile-name, and the suffix is properties.

vi etc/filesystem/hdfs-config-metastore.properties

fs.client.type=hdfs
hdfs.config.resources=/opt/openlookeng/xxx/core-site.xml,/opt/openlookeng/xxx/hdfs-site.xml
hdfs.authentication.type=NONE

For details about HDFS Kerberos configurations, see filesystem.

Note: Either the hbase.properties and hetu-metastore.properties configuration files are mandatory. If HDFS is configured, hdfs-config-metastore.properties is required.

Possible problem: After the configuration is complete, an error message indicating that the xxx path (configuration file A) is not in the white path list is displayed when the service is started.
Solution: Select a directory in the white path list recorded in the log, copy the file in the directory where the error occurs, and modify the corresponding path in configuration file A.

Kerberos Configuration:

If HBase/Zookeeper is a security cluster，so we should add the configuration about kerberos.

hbase.jaas.conf.path=/opt/openlookeng/xxx/jaas.conf

hbase.hbase.site.path=opt/openlookeng/xxx/hbase-site.xml

hbase.krb5.conf.path=opt/openlookeng/xxx/krb5.conf

hbase.kerberos.keytab=opt/openlookeng/xxx/user.keytab

hbase.kerberos.principal=lk_username@HADOOP.COM

hbase.authentication.type=KERBEROS

vi jaas.conf

Client {
com.sun.security.auth.module.Krb5LoginModule required
useKeyTab=true
keyTab="opt/openlookeng/xxx/user.keytab"
principal="lk_username@HADOOP.COM"
useTicketCache=false
storeKey=true
debug=true;
};

Configuration Properties

Table Properties

Data Types

Currently, following 10 data type are supported: VARCHAR, TINYINT, SMALLINT, INTEGER, BIGINT, DOUBLE, BOOLEAN, TIME, DATE and TIMESTAMP.

Push Down

The HBase Connector supports push down most of operators, such as rowkey-based point query, rowkey-based range query. Besides, those predicate conditions are supported to push down: =, >=, >, <, <=, !=, in, not in, between and.

Performance Optimization Configuration

1. Specify sharding rules during table creation to improve the query performance of a single table. Create table xxx() with(split_by_char='0-9, a~z, A~Z'),
split_by_char indicates the range of the first character of a rowKey, which is the basis for splitting a segment. If the first character of the RowKey consists of digits, the rowKey can be segmented based on different digits to improve the query concurrency. Different types of symbols are separated by commas.
If this parameter is set incorrectly, the query result may be incomplete. Set this parameter based on the actual situation. If there is no special requirement, you do not need to change the value. By default, all digits and letters are contained.
If rowKey is a Chinese character, create table xxx() with(split_by_char='一~锯').
In addition, the splitKey of the pre-partition is specified based on split_by_char during table creation and data is distributed to each region as much as possible. In this way, performance is improved during HBase read/write.

2. The client side mode can be configured to read data, improving the multi-concurrent query performance.
The working mechanism of ClientSide is to create snapshots for HBase tables in HDFS and record the address of the region where each data file is located. When data is read, the region is directly accessed without going through the HBase region server. This reduces the pressure on the region server in high concurrency.

Add the following configuration to the etc/catalog/hbase.properties file:
    hbase.client.side.enable=true
    hbase.core.site.path=/opt/openlookeng/xxx/core-site.xml
    hbase.hdfs.site.path=/opt/openlookeng/xxx/hdfs-site.xml

[If HDFS is a security cluster, add the following configuration:
    hbase.authentication.type=KERBEROS
    hbase.krb5.conf.path=/opt/openlookeng/xxx/krb5.conf
    hbase.kerberos.keytab=/opt/openlookeng/xxx/user.keytab
    hbase.kerberos.principal=lk_username@HADOOP.COM

Note: 
1. Currently, the snapshot lifecycle in client side mode is not maintained. If the number of snapshots exceeds the limit of HBase, you need to manually clear the snapshots in HDFS.
2. Operators push down is not supported in client side mode.
3. Snapshots cannot be created for HBase system tables (for example, the schema name is hbase).

Usage Examples

The HBase Connector basically supports all SQL statements, includes creating, querying, and deleting schemas; Adding, dropping, and modifying tables; Inserting data, deleting rows and so on.

Here are some example:

Create Schema

CREATE SCHEMA schemaName;

Drop Schema

It can only drop empty schema.

DROP SCHEMA schemaName;

Create Table

HBase Connector supports two forms of table creation:

1. Create a table and directly link to a existing table in the HBase data source, you must specify hbase_table_name.

2. Create a new table that does not exist in the HBase data source, there is no need to specify hbase_table_name. We must specify ’external = false'.

Below is an example of how to create a table schemaName.tableName and link it to an existing table named hbaseNamespace:hbaseTable :

Column mapping format: ‘column_name:family:qualifier’

CREATE TABLE schemaName.tableName (
    rowId		VARCHAR,
    qualifier1	TINYINT,
    qualifier2	SMALLINT,
    qualifier3	INTEGER,
    qualifier4	BIGINT,
    qualifier5	DOUBLE,
    qualifier6	BOOLEAN,
    qualifier7	TIME,
    qualifier8	DATE,
    qualifier9	TIMESTAMP
)
WITH (
    column_mapping = 'qualifier1:f1:q1, qualifier2:f1:q2, 
    qualifier3:f2:q3, qualifier4:f2:q4, qualifier5:f2:q5, qualifier6:f3:q1, 
    qualifier7:f3:q2, qualifier8:f3:q3, qualifier9:f3:q4',
    row_id = 'rowId',
    hbase_table_name = 'hbaseNamespace:hbaseTable',
    external = false
);

Note

If user specifies hbase_table_name, then the schemaName must be same as the namespace in hbase_table_name when creating a table. Otherwise, the creation will be failed.

Example (failure to create table):

CREATE TABLE default.typeMapping (
    rowId 		VARCHAR,
    qualifier1 	INTEGER
)
WITH (
    column_mapping = 'qualifier2:f1:q2',
    row_id = 'rowId',
    hbase_table_name = 'hello:type4'
);

When creating a table to link with HBase cluster, you must ensure that the schema name in HBase connector and HBase data source are the same and achieve the same isolation level.

Drop Table

DROP TABLE schemaName.tableName;

Rename Table

ALTER TABLE schemaName.tableName RENAME TO schemaName.tableNameNew;

Note

Currently, it does not support renaming the tables which are created through HBase.

Add Column

ALTER TABLE schemaName.tableName
ADD COLUMN column_name INTEGER 
WITH (family = 'f1', qualifier = 'q1');

Note

Currently, deleting columns is not supported

Delete row/rows

DELETE FROM schemeName.tableName where rowId = xxx;
DELETE FROM schemeName.tableName where qualifier1 = xxx;
DELETE FROM schemeName.tableName;

Limitations

Statement show tables can only display the tables that the user has established the associations with HBase data source. Because HBase does not provide an interface to retrieve metadata of tables.

When a large amount of data is inserted into HBase using Openlk, the insertion success depends on the processing capability of the HBase server. If an error occurs during data insertion, you can increase the related parameters of HBase server. You are advised to use a third-party(Bulk) tool rather than openlk to export HBase data.

vi hbase-site.xml

<!-- default is 2 -->
  <property>
    <name>hbase.hregion.memstore.block.multiplier</name>
    <value>4</value>
  </property>

  <!-- default is 64MB 67108864 -->
  <property>
    <name>hbase.hregion.memstore.flush.size</name>
    <value>134217728</value>
  </property>
  <!-- default is 7, should be at least 2x compactionThreshold -->
  <property>
    <name>hbase.hstore.blockingStoreFiles</name>
    <value>200</value>
  </property>
  <property>
    <name>hbase.hstore.compaction.max</name>
    <value>20</value>
    <description>Max number of HStoreFiles to compact per 'minor' compaction.</description>
  </property>