Using HBase Dual-Read

Scenario

The HBase client application loads the configuration items of the active and standby clusters by customization to implement the dual-read capability. HBase dual-read is a key feature that improves the high availability of the HBase cluster system. It applies to four query scenarios: reading data using Get, reading data in batches using Get, reading data using Scan, and querying data using a secondary index. HBase can read data from the active and standby clusters at the same time, reducing the query glitch time. The advantages are as follows:

• High success rate: The concurrent dual-read mechanism ensures a high success rate of read requests.
• High availability: When a single cluster is faulty, the query service is not interrupted. A short network jitter does not prolong the query time.
• High generality: The dual-read feature does not support dual-write, but does not affect the original real-time write scenario.
• Ease-of-use: Client encapsulation is performed, which is not sensed by services.

Restrictions on HBase dual-read:

• The HBase dual-read feature is implemented based on replication. Data read from the standby cluster may be different from that from the active cluster. Therefore, only eventual consistency can be achieved.
• Currently, the HBase dual-read feature is used only for query. When the active cluster breaks down, the latest data cannot be synchronized. As a result, the latest data cannot be queried in the standby cluster.
• A Scan operation of HBase may be split into multiple RPC operations. Data may not be completely the same because related session information is not synchronized between different clusters. Therefore, the dual-read feature takes effect only when an RPC operation is performed for the first time. Requests before ResultScanner close access the cluster used for the first RPC operation.
• The HBase Admin API and real-time write API access only the active cluster. Therefore, after the active cluster breaks down, the Admin API and real-time write API are unavailable, and only the Get and Scan query services are available.

Procedure

1. Obtain the client configuration files core-site.xml, hbase-site.xml, and hdfs-site.xml of the HBase active cluster and save them to the src/main/resources/conf/active directory. This directory needs to be created by yourself. For details, see Preparing for Development and Operating Environment.
2. Obtain the client configuration files core-site.xml, hbase-site.xml, and hdfs-site.xml of the standby cluster and save them to the src/main/resources/conf/standby directory. This directory needs to be created by yourself. For details, see Preparing for Development and Operating Environment.
3. Create the hbase-dual.xml configuration file and save it to the src/main/resources/conf/ directory. For details about the configuration items in the configuration file, see Table 1.

   <?xml version="1.0" encoding="UTF-8"?>
   <configuration>
   <!--Configuration file directory of the active cluster-->
       <property>
           <name>hbase.dualclient.active.cluster.configuration.path</name>
           <value>{Sample code directory}\\src\\main\\resources\\conf\\active</value>
           </property>
   <!--Configuration file directory of the standby cluster-->
       <property>
           <name>hbase.dualclient.standby.cluster.configuration.path</name>
           <value>{Sample code directory}\\src\\main\\resources\\standby</value>
       </property>
   <!--Connection implementation of the dual-read mode-->
       <property>
           <name>hbase.client.connection.impl</name>
           <value>org.apache.hadoop.hbase.client.HBaseMultiClusterConnectionImpl</value>
       </property>
   <!--Normal mode-->
       <property>
           <name>hbase.security.authentication</name>
           <value>Simple</value>
       </property>
   <!--Normal mode-->
       <property>
           <name>hadoop.security.authentication</name>
           <value>Simple</value>
       </property>

Sample Code

• Creating a dual-read configuration. The following code snippet belongs to the init method in TestMain class of the com.huawei.bigdata.hbase.examples packet.

  private static void init() throws IOException {
      // Default load from conf directory
      conf = HBaseConfiguration.create();
      //In Windows environment
      String userdir = TestMain.class.getClassLoader().getResource("conf").getPath() + File.separator;
      //In Linux environment
      //String userdir = System.getProperty("user.dir") + File.separator + "conf" + File.separator;
      conf.addResource(new Path(userdir + "hbase-dual.xml"), false);
    }

• Determining the data source cluster
  • GET request. The following code snippet belongs to the testGet method in HBaseSample class of the com.huawei.bigdata.hbase.examples packet.

    Result result = table.get(get); 
    if (result instanceof DualResult) {
         LOG.info(((DualResult)result).getClusterId()); 
    }

  • Scan request. The following code snippet belongs to the testScanData method in HBaseSample class of the com.huawei.bigdata.hbase.examples packet.

    ResultScanner rScanner = table.getScanner(scan);  
    if (rScanner instanceof HBaseMultiScanner) {
         LOG.info(((HBaseMultiScanner)rScanner).getClusterId()); 
    }

• The client can print metric information.

  Add the following content to the log4j.properties file so that the client can export metric information to the specified file: For details about the metrics, see Printing Metric Information

  log4j.logger.DUAL=debug,DUAL 
  log4j.appender.DUAL=org.apache.log4j.RollingFileAppender 
  log4j.appender.DUAL.File=/var/log/dual.log //Local dual-read log path on the client. Change the value to the actual directory, but ensure that the directory has the write permission.
  log4j.additivity.DUAL=false 
  log4j.appender.DUAL.MaxFileSize=${hbase.log.maxfilesize} 
  log4j.appender.DUAL.MaxBackupIndex=${hbase.log.maxbackupindex} 
  log4j.appender.DUAL.layout=org.apache.log4j.PatternLayout 
  log4j.appender.DUAL.layout.ConversionPattern=%d{ISO8601} %-5p [%t] %c{2}: %m%n