Additional Impala Configuration Options

You can modify the env.sh file located in /opt/mapr/impala/impala-<version>/conf/ to set certain Impala start up options.

Modifying Startup Options

The env.sh file contains values that the Impala server, statestore, and catalog services use during start up. The file also has information about resources allocated for Impala. Most of the default values in the env.sh file should work effectively, however there are some values that you can modify. You can check the current value of all the settings through the Impala web interface, available by default at http://<impala-node-hostname>:25000/varz.

You may want to modify the following content in the env.sh file:

  • Statestore address
  • Amount of memory available to Impala
  • Core dump enablement
  • Session and query idle time

To modify the values, edit the env.sh file and then restart the Impala server, Impala statestore, and Impala catalog to implement the changes.

Example of some file content that you may want to modify:

HIVE_METASTORE_URI=thrift://localhost:9083
  # not needed if /opt/mapr/hive is configured
IMPALA_STATE_STORE_HOST=127.0.0.1
IMPALA_STATE_STORE_PORT=24000
IMPALA_BACKEND_PORT=22000
IMPALA_LOG_DIR=/opt/mapr/impala/impala-1.1.1/log
export IMPALA_STATE_STORE_ARGS=${IMPALA_STATE_STORE_ARGS:- \
-log_dir=${IMPALA_LOG_DIR}} \
-state_store_port=${IMPALA_STATE_STORE_PORT}
export IMPALA_SERVER_ARGS=${IMPALA_SERVER_ARGS:- -log_dir=${IMPALA_LOG_DIR} \   
-state_store_port=${IMPALA_STATE_STORE_PORT} \
-use_statestore 
-state_store_host=${IMPALA_STATE_STORE_HOST} \
-be_port=${IMPALA_BACKEND_PORT}}
export ENABLE_CORE_DUMPS=false
The following table contains a list of settings that you can edit in /opt/mapr/impala/impala-<version>/conf/env.sh with descriptions for how to change them:
Setting Description
Statestore address

You can modify this setting to change the statestore IP address or hostname.

Example:

If a machine with an IP address of 192.168.0.28 is hosting statestore, you can change

IMPALA_STATE_STORE_HOST=127.0.0.1
to 
IMPALA_STATE_STORE_HOST=192.168.0.28.
Memory limits

If you run Impala on nodes that also run MapReduce, both frameworks may compete for memory. Configure memory based on your job requirements and SLAs to ensure that each framework has enough memory to avoid conflicts.

You can include the mem_limit parameter in IMPALA_SERVER_ARGS to limit the amount of memory available to Impala. Use absolute notation, such as 500M or 2G, or a percentage of physical memory, such as 50% to specify the memory limit. Impala aborts queries that exceeds the specified memory limit. Percentage limits are based on the physical memory of the machine.

You can also include the num_threads_per_disk parameter in IMPALA_SERVER_ARGS to limit the number of threads that each disk processes per impala server daemon. Limiting the number of threads per disk can reduce the overall amount of memory consumed.

Example:

To limit Impala to 50% of system memory, modify:
export IMPALA_SERVER_ARGS=${IMPALA_SERVER_ARGS:- \          
    -log_dir=${IMPALA_LOG_DIR} \     
    -state_store_port=${IMPALA_STATE_STORE_PORT} \      
    -use_statestore -state_store_host=${IMPALA_STATE_STORE_HOST} \    
    -be_port=${IMPALA_BACKEND_PORT}}
to
export IMPALA_SERVER_ARGS=${IMPALA_SERVER_ARGS:- \
    -log_dir=${IMPALA_LOG_DIR} \
    -state_store_port=${IMPALA_STATE_STORE_PORT} \
    -use_statestore -state_store_host=${IMPALA_STATE_STORE_HOST} \
    -be_port=${IMPALA_BACKEND_PORT} \
    -mem_limit=50% \
    -num_threads_per_disk=2}
Core dump enablement

Core dump file locations can vary depending on your operating system configuration. Other security settings may prevent Impala from writing core dumps when you enable this option.

To enable core dumps, change export ENABLE_CORE_DUMPS=false to export ENABLE_CORE_DUMPS=true.
Session and query idle time
You can modify the time for which sessions and queries can remain idle by adding the following options to env.sh:
--idle_session_timeout
--idle_query_timeout
Example:
IMPALA_SERVER_ARGS=" \
 -log_dir=${IMPALA_LOG_DIR} \
 -state_store_port=${IMPALA_STATE_STORE_PORT} \
 -idle_query_timeout=<value in seconds> \
 -idle_session_timeout=<value in seconds> \
 -use_statestore \
 -state_store_host=${IMPALA_STATE_STORE_HOST} \
 -catalog_service_host=${CATALOG_SERVICE_HOST} \
 -be_port=${IMPALA_BACKEND_PORT}"
Use background threads to loadand cache metadata

The following option controls the parallelism of metadata loading during start up for the catalogd daemon and makes Impala use background threads after start up to load and cache metadata.

--load_catalog_in_background

The default setting is true.
Determine how much parallelism Impala devotes to loading metadata in background

The following option controls the parallelism of metadata loading during start up for the catalogd daemon and determines how much parallelism Impala devotes to loading metadata in the background:

--num_metadata_loading_threads

The default is 16, but you can increase this value for systems with a huge number of databases, tables, or partitions. You can lower this value for busy systems that have CPU-constraints due to jobs from other components running in the cluster.
INSERT to create new partition and inherit permissions

The following option causes Impala INSERT statements to create each new partition with the same MapR-FS permissions as its parent directory:

--insert_inherit_permissions

By default, INSERT statements create directories for new partitions using default MapR-FS permissions.

After you edit /opt/mapr/impala/impala-<version>/conf/env.sh, use the following commands to restart the Impala server and services:

  • Issue the following command to restart the Impala server:

    $ sudo maprcli node services -name impalaserver -action restart -nodes <IP address where impala server is installed>

  • Issue the following command to restart the Impala statestore:

    $ sudo maprcli node services -name impalastore -action restart -nodes <IP address where impala statestore is installed>

  • Issue the following command to restart the Impala catalog:

    $ sudo maprcli node services -name impalacatalog -action restart -nodes <IP address where impala catalog is installed>