Migrate Hive workloads from HDP (LLAP) to CDW (LLAP)

LLAP in HDP

LLAP on HDP runs on YARN with a persistent LLAP daemon that provides execution and caching of data. You can adjust many aspects of the LLAP deployment, such as:

• Size of the LLAP daemons (Memory / Executors)
• Number of daemons created to scale up and handle large workloads
• Number of Apache Tez ApplicationMasters (coordinators) to establish query concurrency
• Ratio of memory used for processing and cache

The YARN configurations in HDP are complex and are not usually optimal, leading to poor experiences.

In HDP, you could only have one LLAP instance running and the instance was sized to handle workloads at peak intervals. The LLAP instances in HDP could not autoscale and consumed a finite amount of YARN resources regardless of whether workloads were running or not.

When Hive LLAP is used for large complex ETL queries and without a robust workload management in place, large workloads can block BI-type workloads thereby leading to poor user experience for BI users.

Some advanced LLAP on YARN implementations may have used ‘Hive Workload Management’ in LLAP to help manage query isolation and deal with query outliers. ‘Hive Workload Management’ in LLAP is not supported in CDW, however, query isolation in CDW ensures that individual warehouses are completely isolated and ensures that instances have sufficient compute resources for their workloads.

LLAP in CDW

In CDW Private Cloud, Hive LLAP runs in Docker containers on Kubernetes instead of YARN. The Virtual Warehouse instances in CDW are preconfigured to handle the LLAP configurations described above and are optimized for your workloads thereby enabling a predictable and stable deployment.

CDW offers the following benefits:

• Provides isolation by having more than one LLAP instance, which was not easily obtained in LLAP on HDP
• Enables Virtual Warehouse instances to AutoScale (scale up and down) to address varying workload demands
• Automatically suspends a Virtual Warehouse instance if workloads cease and Executors are left idle for a period of time

While setting up your Virtual Warehouse, consider setting up multiple instances to leverage the full use of LLAP that you have enjoyed on HDP. The Virtual Warehouse instances should be configured based on the characteristics of the workloads.

Query concurrency in CDW is controlled by the number of Coordinators in an Executor Group. HiveServer locates an available query coordinator in the Virtual Warehouse to handle the query. The query coordinator generates the final query plan that distributes query tasks across available Executors for execution. Each query coordinator can send query tasks to all query Executors in the Executor Group. There is a 1:1 ratio of Coordinators to Executor Groups. When the query load increases, auto-scaling increases concurrency by adding additional query Executor Groups to the Virtual Warehouse instance.

The following table lists the size of the Executor Groups and Coordinators based on the Virtual Warehouse size:

HDP and CDW LLAP terminology map

When migrating HDP LLAP configurations to CDW, you need to familiarize yourself with the differences in terminology and default configurations that are available in CDW.