Admission Control and Query Queuing

On a busy CDH cluster, you might find there is an optimal number of Impala queries that run concurrently. For example, when the I/O capacity is fully utilized by I/O-intensive queries, you might not find any throughput benefit in running more concurrent queries. By allowing some queries to run at full speed while others wait, rather than having all queries contend for resources and run slowly, admission control can result in higher overall throughput.

For another example, consider a memory-bound workload such as many large joins or aggregation queries. Each such query could briefly use many gigabytes of memory to process intermediate results. Because Impala by default cancels queries that exceed the specified memory limit, running multiple large-scale queries at once might require re-running some queries that are cancelled. In this case, admission control improves the reliability and stability of the overall workload by only allowing as many concurrent queries as the overall memory of the cluster can accommodate.

The admission control feature lets you set an upper limit on the number of concurrent Impala queries and on the memory used by those queries. Any additional queries are queued until the earlier ones finish, rather than being cancelled or running slowly and causing contention. As other queries finish, the queued queries are allowed to proceed.

In CDH 5.7 / Impala 2.5 and higher, you can specify these limits and thresholds for each pool rather than globally. That way, you can balance the resource usage and throughput between steady well-defined workloads, rare resource-intensive queries, and ad hoc exploratory queries.

For more details on the internal workings of admission control, see How Impala Schedules and Enforces Limits on Concurrent Queries.

One way to limit resource usage through admission control is to set an upper limit on the number of concurrent queries. This is the initial technique you might use when you do not have extensive information about memory usage for your workload. This setting can be specified separately for each dynamic resource pool.

You can combine this setting with the memory-based approach described in Memory Limits and Admission Control. If either the maximum number of or the expected memory usage of the concurrent queries is exceeded, subsequent queries are queued until the concurrent workload falls below the threshold again.

See Managing Dynamic Resource Pools for information about all these dynamic resource pool settings, how to use them together, and how to divide different parts of your workload among different pools.

Each dynamic resource pool can have an upper limit on the cluster-wide memory used by queries executing in that pool. This is the technique to use once you have a stable workload with well-understood memory requirements.

Always specify the Default Query Memory Limit for the expected maximum amount of RAM that a query might require on each host, which is equivalent to setting the MEM_LIMIT query option for every query run in that pool. That value affects the execution of each query, preventing it from overallocating memory on each host, and potentially activating the spill-to-disk mechanism or cancelling the query when necessary.

Optionally, specify the Max Memory setting, a cluster-wide limit that determines how many queries can be safely run concurrently, based on the upper memory limit per host multiplied by the number of Impala nodes in the cluster.

You can combine the memory-based settings with the upper limit on concurrent queries described in Concurrent Queries and Admission Control. If either the maximum number of or the expected memory usage of the concurrent queries is exceeded, subsequent queries are queued until the concurrent workload falls below the threshold again.

See Managing Dynamic Resource Pools for information about all these dynamic resource pool settings, how to use them together, and how to divide different parts of your workload among different pools.

The admission control feature is similar in some ways to the Cloudera Manager static partitioning feature, as well as the YARN resource management framework. These features can be used separately or together. This section describes some similarities and differences, to help you decide which combination of resource management features to use for Impala.

Admission control is a lightweight, decentralized system that is suitable for workloads consisting primarily of Impala queries and other SQL statements. It sets "soft" limits that smooth out Impala memory usage during times of heavy load, rather than taking an all-or-nothing approach that cancels jobs that are too resource-intensive.

Because the admission control system does not interact with other Hadoop workloads such as MapReduce jobs, you might use YARN with static service pools on CDH clusters where resources are shared between Impala and other Hadoop components. This configuration is recommended when using Impala in a multitenant cluster. Devote a percentage of cluster resources to Impala, and allocate another percentage for MapReduce and other batch-style workloads. Let admission control handle the concurrency and memory usage for the Impala work within the cluster, and let YARN manage the work for other components within the cluster. In this scenario, Impala's resources are not managed by YARN.

The Impala admission control feature uses the same configuration mechanism as the YARN resource manager to map users to pools and authenticate them.

Although the Impala admission control feature uses a fair-scheduler.xml configuration file behind the scenes, this file does not depend on which scheduler is used for YARN. You still use this file, and Cloudera Manager can generate it for you, even when YARN is using the capacity scheduler.

The admission control system is decentralized, embedded in each Impala daemon and communicating through the statestore mechanism. Although the limits you set for memory usage and number of concurrent queries apply cluster-wide, each Impala daemon makes its own decisions about whether to allow each query to run immediately or to queue it for a less-busy time. These decisions are fast, meaning the admission control mechanism is low-overhead, but might be imprecise during times of heavy load across many coordinators. There could be times when the more queries were queued (in aggregate across the cluster) than the specified limit, or when number of admitted queries exceeds the expected number. Thus, you typically err on the high side for the size of the queue, because there is not a big penalty for having a large number of queued queries; and you typically err on the low side for configuring memory resources, to leave some headroom in case more queries are admitted than expected, without running out of memory and being cancelled as a result.

To avoid a large backlog of queued requests, you can set an upper limit on the size of the queue for queries that are queued. When the number of queued queries exceeds this limit, further queries are cancelled rather than being queued. You can also configure a timeout period per pool, after which queued queries are cancelled, to avoid indefinite waits. If a cluster reaches this state where queries are cancelled due to too many concurrent requests or long waits for query execution to begin, that is a signal for an administrator to take action, either by provisioning more resources, scheduling work on the cluster to smooth out the load, or by doing Impala performance tuning to enable higher throughput.

Most aspects of admission control work transparently with client interfaces such as JDBC and ODBC:

• If a SQL statement is put into a queue rather than running immediately, the API call blocks until the statement is dequeued and begins execution. At that point, the client program can request to fetch results, which might also block until results become available.
• If a SQL statement is cancelled because it has been queued for too long or because it exceeded the memory limit during execution, the error is returned to the client program with a descriptive error message.

In Impala 2.0 and higher, you can submit a SQL SET statement from the client application to change the REQUEST_POOL query option. This option lets you submit queries to different resource pools, as described in REQUEST_POOL Query Option.

At any time, the set of queued queries could include queries submitted through multiple different Impala daemon hosts. All the queries submitted through a particular host will be executed in order, so a CREATE TABLE followed by an INSERT on the same table would succeed. Queries submitted through different hosts are not guaranteed to be executed in the order they were received. Therefore, if you are using load-balancing or other round-robin scheduling where different statements are submitted through different hosts, set up all table structures ahead of time so that the statements controlled by the queuing system are primarily queries, where order is not significant. Or, if a sequence of statements needs to happen in strict order (such as an INSERT followed by a SELECT), submit all those statements through a single session, while connected to the same Impala daemon host.

Admission control has the following limitations or special behavior when used with JDBC or ODBC applications:

• The other resource-related query options, RESERVATION_REQUEST_TIMEOUT and V_CPU_CORES, are no longer used. Those query options only applied to using Impala with Llama, which is no longer supported.

When queries complete quickly and are tuned for optimal memory usage, there is less chance of performance or capacity problems during times of heavy load. Before setting up admission control, tune your Impala queries to ensure that the query plans are efficient and the memory estimates are accurate. Understanding the nature of your workload, and which queries are the most resource-intensive, helps you to plan how to divide the queries into different pools and decide what limits to define for each pool.

For large tables, especially those involved in join queries, keep their statistics up to date after loading substantial amounts of new data or adding new partitions. Use the COMPUTE STATS statement for unpartitioned tables, and COMPUTE INCREMENTAL STATS for partitioned tables.

When you use dynamic resource pools with a Max Memory setting enabled, you typically override the memory estimates that Impala makes based on the statistics from the COMPUTE STATS statement. You either set the MEM_LIMIT query option within a particular session to set an upper memory limit for queries within that session, or a default MEM_LIMIT setting for all queries processed by the impalad instance, or a default MEM_LIMIT setting for all queries assigned to a particular dynamic resource pool. By designating a consistent memory limit for a set of similar queries that use the same resource pool, you avoid unnecessary query queuing or out-of-memory conditions that can arise during high-concurrency workloads when memory estimates for some queries are inaccurate.

Follow other steps from Tuning Impala for Performance to tune your queries.

The configuration options for admission control range from the simple (a single resource pool with a single set of options) to the complex (multiple resource pools with different options, each pool handling queries for a different set of users and groups). Cloudera recommends configuring the settings through the Cloudera Manager user interface.

To see how admission control works for particular queries, examine the profile output for the query. This information is available through the PROFILE statement in impala-shell immediately after running a query in the shell, on the queries page of the Impala debug web UI, or in the Impala log file (basic information at log level 1, more detailed information at log level 2). The profile output contains details about the admission decision, such as whether the query was queued or not and which resource pool it was assigned to. It also includes the estimated and actual memory usage for the query, so you can fine-tune the configuration for the memory limits of the resource pools.

Where practical, use Cloudera Manager to configure the admission control parameters. The Cloudera Manager GUI is much simpler than editing the configuration files directly.

Remember that the limits imposed by admission control are "soft" limits. The decentralized nature of this mechanism means that each Impala node makes its own decisions about whether to allow queries to run immediately or to queue them. These decisions rely on information passed back and forth between nodes by the statestore service. If a sudden surge in requests causes more queries than anticipated to run concurrently, then throughput could decrease due to queries spilling to disk or contending for resources; or queries could be cancelled if they exceed the MEM_LIMIT setting while running.

In impala-shell, you can also specify which resource pool to direct queries to by setting the REQUEST_POOL query option.

If you set up different resource pools for different users and groups, consider reusing any classifications you developed for use with Sentry security. See Enabling Sentry Authorization for Impala for details.

For details about all the Fair Scheduler configuration settings, see Fair Scheduler Configuration, in particular the tags such as <queue> and <aclSubmitApps> to map users and groups to particular resource pools (queues).