All Products
Search

This Product

    E-MapReduce:Impala

    Document Center

    E-MapReduce:Impala

    Last Updated:Jun 23, 2026

    Impala provides high-performance, low-latency SQL queries for data stored in Apache Hadoop. It supports real-time queries with SELECT, JOIN, and aggregate functions on data in HDFS or HBase.

    Background information

    Impala uses the same metadata, SQL syntax (Hive SQL), and ODBC drivers as Apache Hive, providing a familiar and unified platform for batch and real-time queries.

    Note

    If you use the Impala component, delete Hive table partitions using Impala or Hive commands. Do not delete partition directories directly from the file system. This action makes the table unavailable.

    Advantages

    Instead of MapReduce, Impala uses a distributed query engine to access data directly, similar to a relational database management system (RDBMS). Depending on the query type and configuration, Impala can be several orders of magnitude faster than Hive.

    Impala offers the following advantages over other SQL-on-Hadoop solutions:

    • Local processing on data nodes avoids network bottlenecks.

    • Eliminates overhead from expensive data format conversions.

    • Uses a single, open, and unified metastore.

    • All data is immediately queryable, with no need to wait for an extract, transform, and load (ETL) process.

    • Uses all hardware for both Impala queries and MapReduce tasks.

    • Scales using a single pool of computers.

    For more information about Impala, see Apache Impala.

    Architecture

    The following figure shows the Impala architecture in E-MapReduce.Impala

    Impala consists of the following components:

    • Impalad

      Deployed on core and task nodes. Supports scale-out and scale-in.

      The Impala Daemon (Impalad) is the core component of Impala. It runs on each node, reads and writes data files, and accepts queries from the impala-shell command, Hue, Java Database Connectivity (JDBC), or ODBC. Impalad parallelizes queries, distributes work across the cluster, and returns local results to the coordinator node.

    • Statestored

      Deployed on the master-1-1 machine of the master node.

      The Statestore service runs as the Statestored process. It monitors the health of all Impalad processes in the cluster and broadcasts their status. If an Impalad process becomes unavailable due to a node failure, network issue, or software problem, the Statestore notifies all other Impalad processes so that no new queries are sent to the unavailable node.

    • Catalogd

      Deployed on the master-1-1 machine of the master node.

      Catalogd synchronizes metadata changes from any Impalad process to all other Impalad processes in the cluster. Because all requests pass through Statestore, run Statestore and Catalogd on the same node.