Configuring LLAP involves setting properties in YARN and Hive. After you configure
the llap queue, you need to go back to the Hive configuration to continue setting up
low-latency analytical processing (LLAP).
In this task, you accept the Ambari configuration of a number of properties or
reconfigure the properties. Ambari generally configures the number of nodes that run
an LLAP daemon and total concurrent queries depending on the size of the llap queue.
Ambari also attempts to correctly configure the following properties for your
particular cluster:
Memory per Daemon
YARN container size for each daemon (MB)
In-Memory Cache per Daemon
Size of the cache in each container (MB)
Number of executors per LLAP Daemon
The number of fragments that can execute in parallel on a daemon
Use the slider controls to change or restore settings: To set the value
outside the slider range, you move your pointer over the field to enable the hover
actions, and select Override.
Accept or change the Number of Nodes Used By Hive LLAP
(num_llap_nodes property). For example,
accept using 2 nodes for LLAP.
Accept the Maximum Total Concurrent Queries
(hive.server2.tez.sessions.per.default.queue property), or
make changes.
Check Memory per Daemon (hive.llap.daemon.yarn.container.mb
property) and In-Memory Cache per Daemon
(hive.llap.io.memory.size property).
This memory (hive.llap.daemon.yarn.container.mb) plus the
cache (hive.llap.io.memory.size) must fit within the container
size specified for the YARN container. The YARN container configuration setting
appears in Services > YARN > Configs > Settings.
Accept the Number of Executors per LLAP Daemon
(hive.llap.daemon.num.executors), or change this setting if
you know what you are doing, and check that the
hive.llap.io.threadpool.size is the same value.
Save any Hive configuration changes, and in Services > YARN > Settings > Memory - Node, check that the Minimum Container Size (Memory) for YARN is
low.
The value should rarely exceed 1024 MB.
Set the Maximum Container Size (Memory) to the same value as the
Memory Allocated for All YARN Containers on a Node.
In Ambari, select Services > YARN > Configs > Advanced.
In Custom yarn-site, add the following properties unless, upon attempting to
add these properties, Ambari indicates the properties are already
added:yarn.resourcemanager.monitor.capacity.preemption.natural_termination_factor
(value = 1) and
yarn.resourcemanager.monitor.capacity.preemption.total_preemption_per_round
(as described below).
Calculate the value of the total preemption per round by dividing 1 by the
number of cluster nodes. Enter the value as a decimal.
For example, if your cluster has 20 nodes, then divide 1 by 20 and enter
0.05 as the value of this property setting.
Save YARN changes, and go back to the Hive configuration.