Help Center> MapReduce Service> Component Operation Guide (LTS)> Using HBase> Common Issues About HBase> Why Other Services Become Unstable If HBase Sets up A Large Number of Connections over the Network Port?
Updated on 2023-04-28 GMT+08:00

Why Other Services Become Unstable If HBase Sets up A Large Number of Connections over the Network Port?

Question

Why other services become unstable if HBase sets up a large number of connections over the network port?

Answer

When the OS command lsof or netstat is run, it is found that many TCP connections are in the CLOSE_WAIT state and the owner of the connections is HBase RegionServer. This can cause exhaustion of network ports or limit exceeding of HDFS connections, resulting in instability of other services. The HBase CLOSE_WAIT phenomenon is the HBase mechanism.

The reason why HBase CLOSE_WAIT occurs is as follows: HBase data is stored in the HDFS as HFile, which can be called StoreFiles. HBase functions as the client of the HDFS. When HBase creates a StoreFile or starts loading a StoreFile, it creates an HDFS connection. When the StoreFile is created or loaded successfully, the HDFS considers that the task is completed and transfers the connection close permission to HBase. However, HBase may choose not to close the connection to ensure real-time response; that is, HBase may maintain the connection so that it can quickly access the corresponding data file upon request. In this case, the connection is in the CLOSE_WAIT, which indicates that the connection needs to be closed by the client.

When a StoreFile will be created: HBase executes the Flush operation.

When Flush is executed: The data written by HBase is first stored in memstore. The Flush operation is performed only when the usage of memstore reaches the threshold or the flush command is run to write data into the HDFS.

To resolve the issue, use either of the following methods:

Because of the HBase connection mechanism, the number of StoreFiles must be restricted to reduce the occupation of HBase ports. This can be achieved by triggering HBase's the compaction action, that is, HBase file merging.

Method 1: On HBase shell client, run major_compact.

Method 2: Compile HBase client code to invoke the compact method of the HBaseAdmin class to trigger HBase's compaction action.

If the HBase port occupation issue cannot be resolved through compact, it indicates that the HBase usage has reached the bottleneck. In such a case, you are advised to perform the following:

  • Check whether the initial number of Regions configured in the table is appropriate.
  • Check whether useless data exists.

If useless data exists, delete the data to reduce the number of storage files for the HBase. If the preceding conditions are not met, then you need to consider a capacity expansion.