blob: df99ed53e2cb0fcb37fea4eeb9e1512022b6c080 [file] [log] [blame]
<?xml version="1.0" encoding="UTF-8"?>
Licensed to the Apache Software Foundation (ASF) under one
or more contributor license agreements. See the NOTICE file
distributed with this work for additional information
regarding copyright ownership. The ASF licenses this file
to you under the Apache License, Version 2.0 (the
"License"); you may not use this file except in compliance
with the License. You may obtain a copy of the License at
Unless required by applicable law or agreed to in writing,
software distributed under the License is distributed on an
KIND, either express or implied. See the License for the
specific language governing permissions and limitations
under the License.
<!DOCTYPE concept PUBLIC "-//OASIS//DTD DITA Concept//EN" "concept.dtd">
<concept rev="1.3.0" id="admission_control">
<title>Admission Control and Query Queuing</title>
<data name="Category" value="Impala"/>
<data name="Category" value="Querying"/>
<data name="Category" value="Admission Control"/>
<data name="Category" value="Resource Management"/>
<p id="admission_control_intro"> Admission control is an Impala feature that
imposes limits on concurrent SQL queries, to avoid resource usage spikes
and out-of-memory conditions on busy clusters. 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. </p>
<p rev="2.5.0"> In <keyword keyref="impala25_full"/> 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. </p>
<p> In addition to the threshold values for currently executing queries, you
can place limits on the maximum number of queries that are queued
(waiting) and a limit on the amount of time they might wait before
returning with an error. These queue settings let you ensure that queries
do not wait indefinitely so that you can detect and correct
<q>starvation</q> scenarios. </p>
<p> Queries, DML statements, and some DDL statements, including
<codeph>CREATE TABLE AS SELECT</codeph> and <codeph>COMPUTE
STATS</codeph> are affected by admission control. </p>
<p> On a busy 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. </p>
<p> 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. </p>
<p outputclass="toc inpage"/>
<concept id="admission_concurrency">
<title>Concurrent Queries and Admission Control</title>
<p> 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. The settings can be specified separately
for each dynamic resource pool. </p>
<dt> Max Running Queries </dt>
<dd><p>Maximum number of concurrently running queries in this pool.
The default value is unlimited for Impala 2.5 or higher.
(optional)</p> The maximum number of queries that can run
concurrently in this pool. The default value is unlimited. Any
queries for this pool that exceed <uicontrol>Max Running
Queries</uicontrol> are added to the admission control queue until
other queries finish. You can use <uicontrol>Max Running
Queries</uicontrol> in the early stages of resource management,
when you do not have extensive data about query memory usage, to
determine if the cluster performs better overall if throttling is
applied to Impala queries. <p> For a workload with many small
queries, you typically specify a high value for this setting, or
leave the default setting of <q>unlimited</q>. For a workload with
expensive queries, where some number of concurrent queries
saturate the memory, I/O, CPU, or network capacity of the cluster,
set the value low enough that the cluster resources are not
overcommitted for Impala. </p><p>Once you have enabled
memory-based admission control using other pool settings, you can
still use <uicontrol>Max Running Queries</uicontrol> as a
safeguard. If queries exceed either the total estimated memory or
the maximum number of concurrent queries, they are added to the
queue. </p><p>If <uicontrol>Max Running Queries
Multiple</uicontrol> is set, the <uicontrol>Max Running
Queries</uicontrol> setting is ignored.</p>
<dt>Max Running Queries Multiple</dt>
<dd>This floating point number is multiplied by the current total
number of executors at runtime to give the maximum number of
concurrently running queries allowed in the pool. The effect of this
setting scales with the number of executors in the resource
pool.<p>This calculation is rounded up to the nearest integer, so
the result will always be at least one. </p><p>If set to zero or a
negative number, the setting is ignored.</p></dd>
<dt> Max Queued Queries </dt>
<dd> Maximum number of queries that can be queued in this pool. The
default value is 200 for Impala 2.1 or higher and 50 for previous
versions of Impala. (optional)<p>If <uicontrol>Max Queued Queries
Multiple</uicontrol> is set, the <uicontrol>Max Queued
Queries</uicontrol> setting is ignored.</p></dd>
<dt>Max Queued Queries Multiple</dt>
<dd>This floating point number is multiplied by the current total
number of executors at runtime to give the maximum number of queries
that can be queued in the pool. The effect of this setting scales
with the number of executors in the resource pool.<p>This
calculation is rounded up to the nearest integer, so the result
will always be at least one. </p><p>If set to zero or a negative
number, the setting is ignored.</p></dd>
<dt> Queue Timeout </dt>
<dd> The amount of time, in milliseconds, that a query waits in the
admission control queue for this pool before being canceled. The
default value is 60,000 milliseconds. <p>In the following cases,
<uicontrol>Queue Timeout</uicontrol> is not significant, and you
can specify a high value to avoid canceling queries
unexpectedly:<ul id="ul_kzr_rbg_gw">
<li>In a low-concurrency workload where few or no queries are
<li>In an environment without a strict SLA, where it does not
matter if queries occasionally take longer than usual because
they are held in admission control</li>
</ul>You might also need to increase the value to use Impala with
some business intelligence tools that have their own timeout
intervals for queries. </p><p>In a high-concurrency workload,
especially for queries with a tight SLA, long wait times in
admission control can cause a serious problem. For example, if a
query needs to run in 10 seconds, and you have tuned it so that it
runs in 8 seconds, it violates its SLA if it waits in the
admission control queue longer than 2 seconds. In a case like
this, set a low timeout value and monitor how many queries are
cancelled because of timeouts. This technique helps you to
discover capacity, tuning, and scaling problems early, and helps
avoid wasting resources by running expensive queries that have
already missed their SLA. </p><p> If you identify some queries
that can have a high timeout value, and others that benefit from a
low timeout value, you can create separate pools with different
values for this setting. </p>
<p> You can combine these settings with the memory-based approach
described in <xref href="impala_admission.xml#admission_memory"/>. 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. </p>
<concept id="admission_memory">
<title>Memory Limits and Admission Control</title>
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.
<p>Use the following settings to manage memory-based admission
<dt>Max Memory</dt>
The maximum amount of aggregate memory available across the
cluster to all queries executing in this pool. This should be a
portion of the aggregate configured memory for Impala daemons,
which will be shown in the settings dialog next to this option for
convenience. Setting this to a non-zero value enables memory based
admission control.
Impala determines the expected maximum memory used by all
queries in the pool and holds back any further queries that would
result in Max Memory being exceeded.
You set Max Memory in <codeph>fair-scheduler.xml</codeph> file
with the <codeph>maxResources</codeph> tag. For example:
<codeph>&lt;maxResources>2500 mb&lt;/maxResources></codeph>
If you specify Max Memory, you should specify the amount of
memory to allocate to each query in this pool. You can do this in
two ways:
<li>By setting Maximum Query Memory Limit and Minimum Query Memory
Limit. This is preferred in <keyword keyref="impala31_full"/>
and greater and gives Impala flexibility to set aside more
memory to queries that are expected to be memory-hungry.</li>
<li>By setting Default Query Memory Limit to the exact amount of
memory that Impala should set aside for queries in that
Note that in the following cases, Impala will rely entirely on
memory estimates to determine how much memory to set aside for
each query. This is not recommended because it can result in
queries not running or being starved for memory if the estimates
are inaccurate. And it can affect other queries running on the
same node.
<li>Max Memory, Maximum Query Memory Limit, and Minimum Query
Memory Limit are not set, and the <codeph>MEM_LIMIT</codeph>
query option is not set for the query.</li>
<li>Default Query Memory Limit is set to 0, and the
<codeph>MEM_LIMIT</codeph> query option is not set for the
<p>If <uicontrol>Max Memory Multiple</uicontrol> is set, the
<uicontrol>Max Memory</uicontrol> setting is ignored.</p>
<dt>Max Memory Multiple</dt>
<dd> This number of bytes is multiplied by the current total number of
executors at runtime to give the maximum memory available across the
cluster for the pool. The effect of this setting scales with the
number of executors in the resource pool.<p>If set to zero or a
negative number, the setting is ignored.</p></dd>
<dt>Minimum Query Memory Limit and Maximum Query Memory Limit</dt>
<p>These two options determine the minimum and maximum per-host
memory limit that will be chosen by Impala Admission control for
queries in this resource pool. If set, Impala Admission Control
will choose a memory limit between the minimum and maximum values
based on the per-host memory estimate for the query. The memory
limit chosen determines the amount of memory that Impala Admission
control will set aside for this query on each host that the query
is running on. The aggregate memory across all of the hosts that
the query is running on is counted against the pool’s Max
<p>Minimum Query Memory Limit must be less than or equal to Maximum
Query Memory Limit and Max Memory.</p>
<p>A user can override Impala’s choice of memory limit by setting
the <codeph>MEM_LIMIT</codeph> query option. If the Clamp
MEM_LIMIT Query Option setting is set to <codeph>TRUE</codeph> and
the user sets <codeph>MEM_LIMIT</codeph> to a value that is
outside of the range specified by these two options, then the
effective memory limit will be either the minimum or maximum,
depending on whether <codeph>MEM_LIMIT</codeph> is lower than or
higher the range.</p>
<p>For example, assume a resource pool with the following parameters
set: <ul>
<li>Minimum Query Memory Limit = 2GB</li>
<li>Maximum Query Memory Limit = 10GB</li>
</ul>If a user tries to submit a query with the
<codeph>MEM_LIMIT</codeph> query option set to 14 GB, the
following would happen:<ul>
<li>If Clamp MEM_LIMIT Query Option = true, admission controller
would override <codeph>MEM_LIMIT</codeph> with 10 GB and
attempt admission using that value.</li>
<li>If Clamp MEM_LIMIT Query Option = false, the admission
controller will retain the <codeph>MEM_LIMIT</codeph> of 14 GB
set by the user and will attempt admission using the
<dt>Default Query Memory Limit</dt>
<dd>The default memory limit applied to queries executing in this pool
when no explicit <codeph>MEM_LIMIT</codeph> query option is set. The
memory limit chosen determines the amount of memory that Impala
Admission control will set aside for this query on each host that
the query is running on. The aggregate memory across all of the
hosts that the query is running on is counted against the pool’s Max
Memory. This option is deprecated in <keyword keyref="impala31_full"
/> and higher and is replaced by Maximum Query Memory Limit and
Minimum Query Memory Limit. Do not set this if either Maximum Query
Memory Limit or Minimum Query Memory Limit is set.</dd>
<dt> Clamp MEM_LIMIT Query Option</dt>
<dd>If this field is not selected, the <codeph>MEM_LIMIT</codeph>
query option will not be bounded by the <b>Maximum Query Memory
Limit</b> and the <b>Minimum Query Memory Limit</b> values
specified for this resource pool. By default, this field is selected
in Impala 3.1 and higher. The field is disabled if both <b>Minimum
Query Memory Limit</b> and <b>Maximum Query Memory Limit</b> are
not set.</dd>
You can combine the memory-based settings with the upper limit on concurrent queries described in
<xref href="impala_admission.xml#admission_concurrency"/>. 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.
<concept id="set_per_query_memory_limits">
<title>Setting Per-query Memory Limits</title>
<p>Use per-query memory limits to prevent queries from consuming excessive
memory resources that impact other queries. We recommends that you set
the query memory limits whenever possible.</p>
<p>If you set the <b>Max Memory</b> for a resource pool, Impala attempts
to throttle queries if there is not enough memory to run them within the
specified resources.</p>
<p>Only use admission control with maximum memory resources if you can
ensure there are query memory limits. Set the pool <b>Maximum Query
Memory Limit</b> to be certain. You can override this setting with the
<codeph>MEM_LIMIT</codeph> query option, if necessary.</p>
<p>Typically, you set query memory limits using the <codeph>set
MEM_LIMIT=Xg;</codeph> query option. When you find the right value for
your business case, memory-based admission control works well. The
potential downside is that queries that attempt to use more memory might
perform poorly or even be cancelled.</p>
<concept id="admission_yarn">
<title>How Impala Admission Control Relates to Other Resource Management Tools</title>
<data name="Category" value="Concepts"/>
The admission control feature is similar in some ways to 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 <q>soft</q> 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 clusters where resources are shared between
Impala and other Hadoop components. This configuration is recommended when using Impala in a
<term>multitenant</term> 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 <codeph>fair-scheduler.xml</codeph> configuration file
behind the scenes, this file does not depend on which scheduler is used for YARN. You still use this file
even when YARN is using the capacity scheduler.
<concept id="admission_architecture">
<title>How Impala Schedules and Enforces Limits on Concurrent Queries</title>
<data name="Category" value="Concepts"/>
<data name="Category" value="Scheduling"/>
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 <xref href="impala_performance.xml#performance">Impala performance
tuning</xref> to enable higher throughput.
<concept id="admission_jdbc_odbc">
<title>How Admission Control works with Impala Clients (JDBC, ODBC, HiveServer2)</title>
<data name="Category" value="JDBC"/>
<data name="Category" value="ODBC"/>
<data name="Category" value="HiveServer2"/>
<data name="Category" value="Concepts"/>
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.
<p rev=""> In Impala 2.0 and higher, you can submit a SQL
<codeph>SET</codeph> statement from the client application to change
the <codeph>REQUEST_POOL</codeph> query option. This option lets you
submit queries to different resource pools, as described in <xref
href="impala_request_pool.xml#request_pool"/>. </p>
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
<codeph>CREATE TABLE</codeph> followed by an <codeph>INSERT</codeph> 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 <codeph>INSERT</codeph> followed by a
<codeph>SELECT</codeph>), 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
The other resource-related query options,
<codeph>RESERVATION_REQUEST_TIMEOUT</codeph> and <codeph>V_CPU_CORES</codeph>, are no longer used. Those query options only
applied to using Impala with Llama, which is no longer supported.
<concept id="admission_schema_config">
<title>SQL and Schema Considerations for Admission Control</title>
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 <codeph>COMPUTE STATS</codeph> statement for unpartitioned tables, and
<codeph>COMPUTE INCREMENTAL STATS</codeph> for partitioned tables.
When you use dynamic resource pools with a <uicontrol>Max Memory</uicontrol> setting enabled,
you typically override the memory estimates that Impala makes based on the statistics from the
<codeph>COMPUTE STATS</codeph> statement.
You either set the <codeph>MEM_LIMIT</codeph> query option within a particular session to
set an upper memory limit for queries within that session, or a default <codeph>MEM_LIMIT</codeph>
setting for all queries processed by the <cmdname>impalad</cmdname> instance, or
a default <codeph>MEM_LIMIT</codeph> 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 <xref href="impala_performance.xml#performance"/> to tune your queries.
<concept id="admission_guidelines">
<title>Guidelines for Using Admission Control</title>
<data name="Category" value="Planning"/>
<data name="Category" value="Guidelines"/>
<data name="Category" value="Best Practices"/>
<p> The limits imposed by admission control are de-centrally managed
<q>soft</q> limits. Each Impala coordinator 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 the throughput
could decrease due to queries spilling to disk or contending for
resources. Or queries could be cancelled if they exceed the
<codeph>MEM_LIMIT</codeph> setting while running. </p>
<p> In <cmdname>impala-shell</cmdname>, you can also specify which
resource pool to direct queries to by setting the
<codeph>REQUEST_POOL</codeph> query option. </p>
<p> To see how admission control works for particular queries, examine the
profile output or the summary output for the query. <ul>
<li>Profile<p>The information is available through the
<codeph>PROFILE</codeph> statement in
<cmdname>impala-shell</cmdname> immediately after running a
query in the shell, on the <uicontrol>queries</uicontrol> 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). </p><p>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.
<li>Summary<p>Starting in <keyword keyref="impala31"/>, the
information is available in <cmdname>impala-shell</cmdname> when
the <codeph>LIVE_PROGRESS</codeph> or
<codeph>LIVE_SUMMARY</codeph> query option is set to
<codeph>TRUE</codeph>.</p><p>You can also start an
<codeph>impala-shell</codeph> session with the
<codeph>--live_progress</codeph> or
<codeph>--live_summary</codeph> flags to monitor all queries in
that <codeph>impala-shell</codeph> session.</p><p>The summary
output includes the queuing status consisting of whether the query
was queued and what was the latest queuing reason.</p></li>
<p> For details about all the Fair Scheduler configuration settings, see
<xref keyref="FairScheduler">Fair Scheduler Configuration</xref>, in
particular the tags such as <codeph>&lt;queue&gt;</codeph> and
<codeph>&lt;aclSubmitApps&gt;</codeph> to map users and groups to
particular resource pools (queues). </p>