Google Git
Sign in
apache / impala / 79aae231443a305ce8503dbc7b4335e8ae3f3946 / . / docs / topics / impala_compute_stats.xml
blob: 95343f47fb426bc6742cb8a840a9e0b9940a49e1 [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
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing,
software distributed under the License is distributed on an
"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
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="1.2.2" id="compute_stats">
<title>COMPUTE STATS Statement</title>
<titlealts audience="PDF"><navtitle>COMPUTE STATS</navtitle></titlealts>
<prolog>
<metadata>
<data name="Category" value="Impala"/>
<data name="Category" value="Performance"/>
<data name="Category" value="Scalability"/>
<data name="Category" value="ETL"/>
<data name="Category" value="Ingest"/>
<data name="Category" value="SQL"/>
<data name="Category" value="Tables"/>
<data name="Category" value="Developers"/>
<data name="Category" value="Data Analysts"/>
</metadata>
</prolog>
<conbody>
<p>
<indexterm audience="hidden">COMPUTE STATS statement</indexterm> The
COMPUTE STATS statement gathers information about volume and distribution
of data in a table and all associated columns and partitions. The
information is stored in the metastore database, and used by Impala to
help optimize queries. For example, if Impala can determine that a table
is large or small, or has many or few distinct values it can organize and
parallelize the work appropriately for a join query or insert operation.
For details about the kinds of information gathered by this statement, see
<xref href="impala_perf_stats.xml#perf_stats"/>.
</p>
<p conref="../shared/impala_common.xml#common/syntax_blurb"/>
<codeblock rev="2.1.0"><ph rev="2.12.0 IMPALA-5310">COMPUTE STATS [<varname>db_name</varname>.]<varname>table_name</varname> [ ( <varname>column_list</varname> ) ] [TABLESAMPLE SYSTEM(<varname>percentage</varname>) [REPEATABLE(<varname>seed</varname>)]]</ph>
<varname>column_list</varname> ::= <varname>column_name</varname> [ , <varname>column_name</varname>, ... ]
COMPUTE INCREMENTAL STATS [<varname>db_name</varname>.]<varname>table_name</varname> [PARTITION (<varname>partition_spec</varname>)]
<varname>partition_spec</varname> ::= <varname>simple_partition_spec</varname> | <ph rev="IMPALA-1654"><varname>complex_partition_spec</varname></ph>
<varname>simple_partition_spec</varname> ::= <varname>partition_col</varname>=<varname>constant_value</varname>
<ph rev="IMPALA-1654"><varname>complex_partition_spec</varname> ::= <varname>comparison_expression_on_partition_col</varname></ph>
</codeblock>
<p conref="../shared/impala_common.xml#common/incremental_partition_spec"/>
<p conref="../shared/impala_common.xml#common/usage_notes_blurb"/>
<p>
Originally, Impala relied on users to run the Hive <codeph>ANALYZE
TABLE</codeph> statement, but that method of gathering statistics proved
unreliable and difficult to use. The Impala <codeph>COMPUTE STATS</codeph>
statement was built to improve the reliability and user-friendliness of
this operation. <codeph>COMPUTE STATS</codeph> does not require any setup
steps or special configuration. You only run a single Impala
<codeph>COMPUTE STATS</codeph> statement to gather both table and column
statistics, rather than separate Hive <codeph>ANALYZE TABLE</codeph>
statements for each kind of statistics.
</p>
<p rev="impala-3562">
For non-incremental <codeph>COMPUTE STATS</codeph>
statement, the columns for which statistics are computed can be specified
with an optional comma-separate list of columns.
</p>
<p rev="impala-3562">
If no column list is given, the <codeph>COMPUTE STATS</codeph> statement
computes column-level statistics for all columns of the table. This adds
potentially unneeded work for columns whose stats are not needed by
queries. It can be especially costly for very wide tables and unneeded
large string fields.
</p>
<p rev="impala-3562">
<codeph>COMPUTE STATS</codeph> returns an error when a specified column
cannot be analyzed, such as when the column does not exist, the column is
of an unsupported type for COMPUTE STATS, e.g. colums of complex types,
or the column is a partitioning column.
</p>
<p rev="impala-3562">
If an empty column list is given, no column is analyzed by <codeph>COMPUTE
STATS</codeph>.
</p>
<p rev="2.12.0 IMPALA-5310">
In <keyword keyref="impala212_full"/> and
higher, an optional <codeph>TABLESAMPLE</codeph> clause immediately after
a table reference specifies that the <codeph>COMPUTE STATS</codeph>
operation only processes a specified percentage of the table data. For
tables that are so large that a full <codeph>COMPUTE STATS</codeph>
operation is impractical, you can use <codeph>COMPUTE STATS</codeph> with
a <codeph>TABLESAMPLE</codeph> clause to extrapolate statistics from a
sample of the table data. See <keyword keyref="perf_stats"/>about the
experimental stats extrapolation and sampling features.
</p>
<p rev="2.1.0">
The <codeph>COMPUTE INCREMENTAL STATS</codeph> variation is a shortcut for partitioned tables that works on a
subset of partitions rather than the entire table. The incremental nature makes it suitable for large tables
with many partitions, where a full <codeph>COMPUTE STATS</codeph> operation takes too long to be practical
each time a partition is added or dropped. See <xref href="impala_perf_stats.xml#perf_stats_incremental"/>
for full usage details.
</p>
<note type="important">
<p conref="../shared/impala_common.xml#common/cs_or_cis"/>
<p conref="../shared/impala_common.xml#common/incremental_stats_after_full"/>
<p conref="../shared/impala_common.xml#common/incremental_stats_caveats"/>
</note>
<p>
<codeph>COMPUTE INCREMENTAL STATS</codeph> only applies to partitioned tables. If you use the
<codeph>INCREMENTAL</codeph> clause for an unpartitioned table, Impala automatically uses the original
<codeph>COMPUTE STATS</codeph> statement. Such tables display <codeph>false</codeph> under the
<codeph>Incremental stats</codeph> column of the <codeph>SHOW TABLE STATS</codeph> output.
</p>
<note>
<p>
Because many of the most performance-critical and resource-intensive
operations rely on table and column statistics to construct accurate and
efficient plans, <codeph>COMPUTE STATS</codeph> is an important step at
the end of your ETL process. Run <codeph>COMPUTE STATS</codeph> on all
tables as your first step during performance tuning for slow queries, or
troubleshooting for out-of-memory conditions:
<ul>
<li>
Accurate statistics help Impala construct an efficient query plan
for join queries, improving performance and reducing memory usage.
</li>
<li>
Accurate statistics help Impala distribute the work effectively
for insert operations into Parquet tables, improving performance and
reducing memory usage.
</li>
<li rev="1.3.0">
Accurate statistics help Impala estimate the memory
required for each query, which is important when you use resource
management features, such as admission control and the YARN resource
management framework. The statistics help Impala to achieve high
concurrency, full utilization of available memory, and avoid
contention with workloads from other Hadoop components.
</li>
<li rev="IMPALA-4572">
In <keyword keyref="impala28_full"/> and
higher, when you run the <codeph>COMPUTE STATS</codeph> or
<codeph>COMPUTE INCREMENTAL STATS</codeph> statement against a
Parquet table, Impala automatically applies the query option setting
<codeph>MT_DOP=4</codeph> to increase the amount of intra-node
parallelism during this CPU-intensive operation. See <xref
keyref="mt_dop"/> for details about what this query option does
and how to use it with CPU-intensive <codeph>SELECT</codeph>
statements.
</li>
</ul>
</p>
</note>
<p rev="IMPALA-1654">
<b>Computing stats for groups of partitions:</b>
</p>
<p rev="IMPALA-1654">
In <keyword keyref="impala28_full"/> and higher, you can run <codeph>COMPUTE INCREMENTAL STATS</codeph>
on multiple partitions, instead of the entire table or one partition at a time. You include
comparison operators other than <codeph>=</codeph> in the <codeph>PARTITION</codeph> clause,
and the <codeph>COMPUTE INCREMENTAL STATS</codeph> statement applies to all partitions that
match the comparison expression.
</p>
<p rev="IMPALA-1654">
For example, the <codeph>INT_PARTITIONS</codeph> table contains 4 partitions.
The following <codeph>COMPUTE INCREMENTAL STATS</codeph> statements affect some but not all
partitions, as indicated by the <codeph>Updated <varname>n</varname> partition(s)</codeph>
messages. The partitions that are affected depend on values in the partition key column <codeph>X</codeph>
that match the comparison expression in the <codeph>PARTITION</codeph> clause.
</p>
<codeblock rev="IMPALA-1654"><![CDATA[
show partitions int_partitions;
+-------+-------+--------+------+--------------+-------------------+---------+...
| x | #Rows | #Files | Size | Bytes Cached | Cache Replication | Format |...
+-------+-------+--------+------+--------------+-------------------+---------+...
| 99 | -1 | 0 | 0B | NOT CACHED | NOT CACHED | PARQUET |...
| 120 | -1 | 0 | 0B | NOT CACHED | NOT CACHED | TEXT |...
| 150 | -1 | 0 | 0B | NOT CACHED | NOT CACHED | TEXT |...
| 200 | -1 | 0 | 0B | NOT CACHED | NOT CACHED | TEXT |...
| Total | -1 | 0 | 0B | 0B | | |...
+-------+-------+--------+------+--------------+-------------------+---------+...
compute incremental stats int_partitions partition (x < 100);
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 1 partition(s) and 1 column(s). |
+-----------------------------------------+
compute incremental stats int_partitions partition (x in (100, 150, 200));
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 2 partition(s) and 1 column(s). |
+-----------------------------------------+
compute incremental stats int_partitions partition (x between 100 and 175);
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 2 partition(s) and 1 column(s). |
+-----------------------------------------+
compute incremental stats int_partitions partition (x in (100, 150, 200) or x < 100);
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 3 partition(s) and 1 column(s). |
+-----------------------------------------+
compute incremental stats int_partitions partition (x != 150);
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 3 partition(s) and 1 column(s). |
+-----------------------------------------+
]]>
</codeblock>
<p conref="../shared/impala_common.xml#common/complex_types_blurb"/>
<p rev="2.3.0">
Currently, the statistics created by the <codeph>COMPUTE STATS</codeph> statement do not include
information about complex type columns. The column stats metrics for complex columns are always shown
as -1. For queries involving complex type columns, Impala uses
heuristics to estimate the data distribution within such columns.
</p>
<p conref="../shared/impala_common.xml#common/hbase_blurb"/>
<p>
<codeph>COMPUTE STATS</codeph> works for HBase tables also. The statistics gathered for HBase tables are
somewhat different than for HDFS-backed tables, but that metadata is still used for optimization when HBase
tables are involved in join queries.
</p>
<p conref="../shared/impala_common.xml#common/s3_blurb"/>
<p rev="2.2.0">
<codeph>COMPUTE STATS</codeph> also works for tables where data resides in the Amazon Simple Storage Service (S3).
See <xref href="impala_s3.xml#s3"/> for details.
</p>
<p conref="../shared/impala_common.xml#common/performance_blurb"/>
<p>
The statistics collected by <codeph>COMPUTE STATS</codeph> are used to optimize join queries
<codeph>INSERT</codeph> operations into Parquet tables, and other resource-intensive kinds of SQL statements.
See <xref href="impala_perf_stats.xml#perf_stats"/> for details.
</p>
<p>
For large tables, the <codeph>COMPUTE STATS</codeph> statement itself might take a long time and you
might need to tune its performance. The <codeph>COMPUTE STATS</codeph> statement does not work with the
<codeph>EXPLAIN</codeph> statement, or the <codeph>SUMMARY</codeph> command in <cmdname>impala-shell</cmdname>.
You can use the <codeph>PROFILE</codeph> statement in <cmdname>impala-shell</cmdname> to examine timing information
for the statement as a whole. If a basic <codeph>COMPUTE STATS</codeph> statement takes a long time for a
partitioned table, consider switching to the <codeph>COMPUTE INCREMENTAL STATS</codeph> syntax so that only
newly added partitions are analyzed each time.
</p>
<p conref="../shared/impala_common.xml#common/example_blurb"/>
<p>
This example shows two tables, <codeph>T1</codeph> and <codeph>T2</codeph>, with a small number distinct
values linked by a parent-child relationship between <codeph>T1.ID</codeph> and <codeph>T2.PARENT</codeph>.
<codeph>T1</codeph> is tiny, while <codeph>T2</codeph> has approximately 100K rows. Initially, the statistics
includes physical measurements such as the number of files, the total size, and size measurements for
fixed-length columns such as with the <codeph>INT</codeph> type. Unknown values are represented by -1. After
running <codeph>COMPUTE STATS</codeph> for each table, much more information is available through the
<codeph>SHOW STATS</codeph> statements. If you were running a join query involving both of these tables, you
would need statistics for both tables to get the most effective optimization for the query.
</p>
<!-- Note: chopped off any excess characters at position 87 and after,
to avoid weird wrapping in PDF.
Applies to any subsequent examples with output from SHOW ... STATS too. -->
<codeblock>[localhost:21000] &gt; show table stats t1;
Query: show table stats t1
+-------+--------+------+--------+
| #Rows | #Files | Size | Format |
+-------+--------+------+--------+
| -1 | 1 | 33B | TEXT |
+-------+--------+------+--------+
Returned 1 row(s) in 0.02s
[localhost:21000] &gt; show table stats t2;
Query: show table stats t2
+-------+--------+----------+--------+
| #Rows | #Files | Size | Format |
+-------+--------+----------+--------+
| -1 | 28 | 960.00KB | TEXT |
+-------+--------+----------+--------+
Returned 1 row(s) in 0.01s
[localhost:21000] &gt; show column stats t1;
Query: show column stats t1
+--------+--------+------------------+--------+----------+----------+
| Column | Type | #Distinct Values | #Nulls | Max Size | Avg Size |
+--------+--------+------------------+--------+----------+----------+
| id | INT | -1 | -1 | 4 | 4 |
| s | STRING | -1 | -1 | -1 | -1 |
+--------+--------+------------------+--------+----------+----------+
Returned 2 row(s) in 1.71s
[localhost:21000] &gt; show column stats t2;
Query: show column stats t2
+--------+--------+------------------+--------+----------+----------+
| Column | Type | #Distinct Values | #Nulls | Max Size | Avg Size |
+--------+--------+------------------+--------+----------+----------+
| parent | INT | -1 | -1 | 4 | 4 |
| s | STRING | -1 | -1 | -1 | -1 |
+--------+--------+------------------+--------+----------+----------+
Returned 2 row(s) in 0.01s
[localhost:21000] &gt; compute stats t1;
Query: compute stats t1
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 1 partition(s) and 2 column(s). |
+-----------------------------------------+
Returned 1 row(s) in 5.30s
[localhost:21000] &gt; show table stats t1;
Query: show table stats t1
+-------+--------+------+--------+
| #Rows | #Files | Size | Format |
+-------+--------+------+--------+
| 3 | 1 | 33B | TEXT |
+-------+--------+------+--------+
Returned 1 row(s) in 0.01s
[localhost:21000] &gt; show column stats t1;
Query: show column stats t1
+--------+--------+------------------+--------+----------+----------+
| Column | Type | #Distinct Values | #Nulls | Max Size | Avg Size |
+--------+--------+------------------+--------+----------+----------+
| id | INT | 3 | -1 | 4 | 4 |
| s | STRING | 3 | -1 | -1 | -1 |
+--------+--------+------------------+--------+----------+----------+
Returned 2 row(s) in 0.02s
[localhost:21000] &gt; compute stats t2;
Query: compute stats t2
+-----------------------------------------+
| summary |
+-----------------------------------------+
| Updated 1 partition(s) and 2 column(s). |
+-----------------------------------------+
Returned 1 row(s) in 5.70s
[localhost:21000] &gt; show table stats t2;
Query: show table stats t2
+-------+--------+----------+--------+
| #Rows | #Files | Size | Format |
+-------+--------+----------+--------+
| 98304 | 1 | 960.00KB | TEXT |
+-------+--------+----------+--------+
Returned 1 row(s) in 0.03s
[localhost:21000] &gt; show column stats t2;
Query: show column stats t2
+--------+--------+------------------+--------+----------+----------+
| Column | Type | #Distinct Values | #Nulls | Max Size | Avg Size |
+--------+--------+------------------+--------+----------+----------+
| parent | INT | 3 | -1 | 4 | 4 |
| s | STRING | 6 | -1 | 14 | 9.3 |
+--------+--------+------------------+--------+----------+----------+
Returned 2 row(s) in 0.01s</codeblock>
<p rev="2.1.0">
The following example shows how to use the <codeph>INCREMENTAL</codeph> clause, available in Impala 2.1.0 and
higher. The <codeph>COMPUTE INCREMENTAL STATS</codeph> syntax lets you collect statistics for newly added or
changed partitions, without rescanning the entire table.
</p>
<codeblock conref="../shared/impala_common.xml#common/compute_stats_walkthrough"/>
<p conref="../shared/impala_common.xml#common/file_format_blurb"/>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with tables created with any of the file formats supported
by Impala. See <xref href="impala_file_formats.xml#file_formats"/> for details about working with the
different file formats. The following considerations apply to <codeph>COMPUTE STATS</codeph> depending on the
file format of the table.
</p>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with text tables with no restrictions. These tables can be
created through either Impala or Hive.
</p>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with Parquet tables. These tables can be created through
either Impala or Hive.
</p>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with Avro tables without restriction in <keyword keyref="impala22_full"/>
and higher. In earlier releases, <codeph>COMPUTE STATS</codeph> worked only for Avro tables created through Hive,
and required the <codeph>CREATE TABLE</codeph> statement to use SQL-style column names and types rather than an
Avro-style schema specification.
</p>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with RCFile tables with no restrictions. These tables can
be created through either Impala or Hive.
</p>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with SequenceFile tables with no restrictions. These
tables can be created through either Impala or Hive.
</p>
<p>
The <codeph>COMPUTE STATS</codeph> statement works with partitioned tables, whether all the partitions use
the same file format, or some partitions are defined through <codeph>ALTER TABLE</codeph> to use different
file formats.
</p>
<p conref="../shared/impala_common.xml#common/ddl_blurb"/>
<p conref="../shared/impala_common.xml#common/cancel_blurb_maybe"/>
<p conref="../shared/impala_common.xml#common/restrictions_blurb"/>
<note conref="../shared/impala_common.xml#common/compute_stats_nulls"/>
<p conref="../shared/impala_common.xml#common/internals_blurb"/>
<p>
Behind the scenes, the <codeph>COMPUTE STATS</codeph> statement
executes two statements: one to count the rows of each partition
in the table (or the entire table if unpartitioned) through the
<codeph>COUNT(*)</codeph> function,
and another to count the approximate number of distinct values
in each column through the <codeph>NDV()</codeph> function.
You might see these queries in your monitoring and diagnostic displays.
The same factors that affect the performance, scalability, and
execution of other queries (such as parallel execution, memory usage,
admission control, and timeouts) also apply to the queries run by the
<codeph>COMPUTE STATS</codeph> statement.
</p>
<p conref="../shared/impala_common.xml#common/permissions_blurb"/>
<p rev="">
The user ID that the <cmdname>impalad</cmdname> daemon runs under,
typically the <codeph>impala</codeph> user, must have read
permission for all affected files in the source directory:
all files in the case of an unpartitioned table or
a partitioned table in the case of <codeph>COMPUTE STATS</codeph>;
or all the files in partitions without incremental stats in
the case of <codeph>COMPUTE INCREMENTAL STATS</codeph>.
It must also have read and execute permissions for all
relevant directories holding the data files.
(Essentially, <codeph>COMPUTE STATS</codeph> requires the
same permissions as the underlying <codeph>SELECT</codeph> queries it runs
against the table.)
</p>
<p rev="kudu" conref="../shared/impala_common.xml#common/kudu_blurb"/>
<p rev="IMPALA-2830">
The <codeph>COMPUTE STATS</codeph> statement applies to Kudu tables.
Impala does not compute the number of rows for each partition for
Kudu tables. Therefore, you do not need to re-run the operation when
you see -1 in the <codeph># Rows</codeph> column of the output from
<codeph>SHOW TABLE STATS</codeph>. That column always shows -1 for
all Kudu tables.
</p>
<p conref="../shared/impala_common.xml#common/related_info"/>
<p>
<xref href="impala_drop_stats.xml#drop_stats"/>, <xref href="impala_show.xml#show_table_stats"/>,
<xref href="impala_show.xml#show_column_stats"/>, <xref href="impala_perf_stats.xml#perf_stats"/>
</p>
</conbody>
</concept>