Help Center> MapReduce Service> Component Operation Guide (LTS)> Using HBase> Common Issues About HBase> Why Does the HBase BulkLoad Task (One Table Has 26 TB Data) Consisting of 210,000 Map Tasks and 10,000 Reduce Tasks Fail?
Updated on 2023-04-28 GMT+08:00

Why Does the HBase BulkLoad Task (One Table Has 26 TB Data) Consisting of 210,000 Map Tasks and 10,000 Reduce Tasks Fail?

Question

The HBase bulkLoad task (a single table contains 26 TB data) has 210,000 maps and 10,000 reduce tasks, and the task fails.

Answer

ZooKeeper I/O bottleneck observation methods:

  1. On the monitoring page of Manager, check whether the number of ZooKeeper requests on a single node exceeds the upper limit.
  2. View ZooKeeper and HBase logs to check whether a large number of I/O Exception Timeout or SocketTimeout Exception exceptions occur.

Optimization suggestions:

  1. Change the number of ZooKeeper instances to 5 or more. You are advised to set peerType to observer to increase the number of observers.
  2. Control the number of concurrent maps of a single task or reduce the memory for running tasks on each node to lighten the node load.
  3. Upgrade ZooKeeper data disks, such as SSDs.