| <?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 id="parquet"> |
| |
| <title>Using the Parquet File Format with Impala Tables</title> |
| |
| <titlealts audience="PDF"> |
| |
| <navtitle>Parquet Data Files</navtitle> |
| |
| </titlealts> |
| |
| <prolog> |
| <metadata> |
| <data name="Category" value="Impala"/> |
| <data name="Category" value="File Formats"/> |
| <data name="Category" value="Parquet"/> |
| <data name="Category" value="Developers"/> |
| <data name="Category" value="Data Analysts"/> |
| <data name="Category" value="Tables"/> |
| <data name="Category" value="Schemas"/> |
| </metadata> |
| </prolog> |
| |
| <conbody> |
| |
| <p> |
| Impala allows you to create, manage, and query Parquet tables. Parquet is a |
| column-oriented binary file format intended to be highly efficient for the types of |
| large-scale queries that Impala is best at. Parquet is especially good for queries |
| scanning particular columns within a table, for example, to query <q>wide</q> tables with |
| many columns, or to perform aggregation operations such as <codeph>SUM()</codeph> and |
| <codeph>AVG()</codeph> that need to process most or all of the values from a column. Each |
| Parquet data file written by Impala contains the values for a set of rows (referred to as |
| the <q>row group</q>). Within a data file, the values from each column are organized so |
| that they are all adjacent, enabling good compression for the values from that column. |
| Queries against a Parquet table can retrieve and analyze these values from any column |
| quickly and with minimal I/O. |
| </p> |
| |
| <p> |
| See <xref href="impala_file_formats.xml#file_formats"/> for the summary of Parquet format |
| support. |
| </p> |
| |
| <p outputclass="toc inpage"/> |
| |
| </conbody> |
| |
| <concept id="parquet_ddl"> |
| |
| <title>Creating Parquet Tables in Impala</title> |
| |
| <conbody> |
| |
| <p> |
| To create a table named <codeph>PARQUET_TABLE</codeph> that uses the Parquet format, you |
| would use a command like the following, substituting your own table name, column names, |
| and data types: |
| </p> |
| |
| <codeblock>[impala-host:21000] > create table <varname>parquet_table_name</varname> (x INT, y STRING) STORED AS PARQUET;</codeblock> |
| |
| <p> |
| Or, to clone the column names and data types of an existing table: |
| </p> |
| |
| <codeblock>[impala-host:21000] > create table <varname>parquet_table_name</varname> LIKE <varname>other_table_name</varname> STORED AS PARQUET;</codeblock> |
| |
| <p rev="1.4.0"> |
| In Impala 1.4.0 and higher, you can derive column definitions from a raw Parquet data |
| file, even without an existing Impala table. For example, you can create an external |
| table pointing to an HDFS directory, and base the column definitions on one of the files |
| in that directory: |
| </p> |
| |
| <codeblock rev="1.4.0">CREATE EXTERNAL TABLE ingest_existing_files LIKE PARQUET '/user/etl/destination/datafile1.dat' |
| STORED AS PARQUET |
| LOCATION '/user/etl/destination'; |
| </codeblock> |
| |
| <p> |
| Or, you can refer to an existing data file and create a new empty table with suitable |
| column definitions. Then you can use <codeph>INSERT</codeph> to create new data files or |
| <codeph>LOAD DATA</codeph> to transfer existing data files into the new table. |
| </p> |
| |
| <codeblock rev="1.4.0">CREATE TABLE columns_from_data_file LIKE PARQUET '/user/etl/destination/datafile1.dat' |
| STORED AS PARQUET; |
| </codeblock> |
| |
| <p> |
| The default properties of the newly created table are the same as for any other |
| <codeph>CREATE TABLE</codeph> statement. For example, the default file format is text; |
| if you want the new table to use the Parquet file format, include the <codeph>STORED AS |
| PARQUET</codeph> file also. |
| </p> |
| |
| <p> |
| In this example, the new table is partitioned by year, month, and day. These partition |
| key columns are not part of the data file, so you specify them in the <codeph>CREATE |
| TABLE</codeph> statement: |
| </p> |
| |
| <codeblock rev="1.4.0">CREATE TABLE columns_from_data_file LIKE PARQUET '/user/etl/destination/datafile1.dat' |
| PARTITION (year INT, month TINYINT, day TINYINT) |
| STORED AS PARQUET; |
| </codeblock> |
| |
| <p rev="1.4.0"> |
| See <xref href="impala_create_table.xml#create_table"/> for more details about the |
| <codeph>CREATE TABLE LIKE PARQUET</codeph> syntax. |
| </p> |
| |
| <p> |
| Once you have created a table, to insert data into that table, use a command similar to |
| the following, again with your own table names: |
| </p> |
| |
| <codeblock>[impala-host:21000] > insert overwrite table <varname>parquet_table_name</varname> select * from <varname>other_table_name</varname>;</codeblock> |
| |
| <p> |
| If the Parquet table has a different number of columns or different column names than |
| the other table, specify the names of columns from the other table rather than |
| <codeph>*</codeph> in the <codeph>SELECT</codeph> statement. |
| </p> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_etl"> |
| |
| <title>Loading Data into Parquet Tables</title> |
| |
| <prolog> |
| <metadata> |
| <data name="Category" value="ETL"/> |
| </metadata> |
| </prolog> |
| |
| <conbody> |
| |
| <p> |
| Choose from the following techniques for loading data into Parquet tables, depending on |
| whether the original data is already in an Impala table, or exists as raw data files |
| outside Impala. |
| </p> |
| |
| <p> |
| If you already have data in an Impala or Hive table, perhaps in a different file format |
| or partitioning scheme, you can transfer the data to a Parquet table using the Impala |
| <codeph>INSERT...SELECT</codeph> syntax. You can convert, filter, repartition, and do |
| other things to the data as part of this same <codeph>INSERT</codeph> statement. See |
| <xref |
| href="#parquet_compression"/> for some examples showing how to insert |
| data into Parquet tables. |
| </p> |
| |
| <p> |
| When inserting into partitioned tables, especially using the Parquet file format, you |
| can include a hint in the <codeph>INSERT</codeph> statement to fine-tune the overall |
| performance of the operation and its resource usage. See <keyword keyref="hints"/> for |
| using hints in the <codeph>INSERT</codeph> statements. |
| </p> |
| |
| <p conref="../shared/impala_common.xml#common/insert_parquet_blocksize"/> |
| |
| <p> |
| Avoid the <codeph>INSERT...VALUES</codeph> syntax for Parquet tables, because |
| <codeph>INSERT...VALUES</codeph> produces a separate tiny data file for each |
| <codeph>INSERT...VALUES</codeph> statement, and the strength of Parquet is in its |
| handling of data (compressing, parallelizing, and so on) in |
| <ph rev="parquet_block_size">large</ph> chunks. |
| </p> |
| |
| <p> |
| If you have one or more Parquet data files produced outside of Impala, you can quickly |
| make the data queryable through Impala by one of the following methods: |
| </p> |
| |
| <ul> |
| <li> |
| The <codeph>LOAD DATA</codeph> statement moves a single data file or a directory full |
| of data files into the data directory for an Impala table. It does no validation or |
| conversion of the data. The original data files must be somewhere in HDFS, not the |
| local filesystem. |
| </li> |
| |
| <li> |
| The <codeph>CREATE TABLE</codeph> statement with the <codeph>LOCATION</codeph> clause |
| creates a table where the data continues to reside outside the Impala data directory. |
| The original data files must be somewhere in HDFS, not the local filesystem. For extra |
| safety, if the data is intended to be long-lived and reused by other applications, you |
| can use the <codeph>CREATE EXTERNAL TABLE</codeph> syntax so that the data files are |
| not deleted by an Impala <codeph>DROP TABLE</codeph> statement. |
| </li> |
| |
| <li> |
| If the Parquet table already exists, you can copy Parquet data files directly into it, |
| then use the <codeph>REFRESH</codeph> statement to make Impala recognize the newly |
| added data. Remember to preserve the block size of the Parquet data files by using the |
| <codeph>hadoop distcp -pb</codeph> command rather than a <codeph>-put</codeph> or |
| <codeph>-cp</codeph> operation on the Parquet files. See |
| <xref href="#parquet_compression_multiple"/> for an example of this kind of operation. |
| </li> |
| </ul> |
| |
| <note |
| conref="../shared/impala_common.xml#common/restrictions_nonimpala_parquet"/> |
| |
| <p> |
| Recent versions of Sqoop can produce Parquet output files using the |
| <codeph>--as-parquetfile</codeph> option. |
| </p> |
| |
| <p conref="../shared/impala_common.xml#common/sqoop_timestamp_caveat" |
| audience="hidden"/> |
| |
| <p> |
| If the data exists outside Impala and is in some other format, combine both of the |
| preceding techniques. First, use a <codeph>LOAD DATA</codeph> or <codeph>CREATE EXTERNAL |
| TABLE ... LOCATION</codeph> statement to bring the data into an Impala table that uses |
| the appropriate file format. Then, use an <codeph>INSERT...SELECT</codeph> statement to |
| copy the data to the Parquet table, converting to Parquet format as part of the process. |
| </p> |
| |
| <p> |
| Loading data into Parquet tables is a memory-intensive operation, because the incoming |
| data is buffered until it reaches <ph |
| rev="parquet_block_size">one data |
| block</ph> in size, then that chunk of data is organized and compressed in memory before |
| being written out. The memory consumption can be larger when inserting data into |
| partitioned Parquet tables, because a separate data file is written for each combination |
| of partition key column values, potentially requiring several |
| <ph rev="parquet_block_size">large</ph> chunks to be manipulated in memory at once. |
| </p> |
| |
| <p> |
| When inserting into a partitioned Parquet table, Impala redistributes the data among the |
| nodes to reduce memory consumption. You might still need to temporarily increase the |
| memory dedicated to Impala during the insert operation, or break up the load operation |
| into several <codeph>INSERT</codeph> statements, or both. |
| </p> |
| |
| <note> |
| All the preceding techniques assume that the data you are loading matches the structure |
| of the destination table, including column order, column names, and partition layout. To |
| transform or reorganize the data, start by loading the data into a Parquet table that |
| matches the underlying structure of the data, then use one of the table-copying |
| techniques such as <codeph>CREATE TABLE AS SELECT</codeph> or <codeph>INSERT ... |
| SELECT</codeph> to reorder or rename columns, divide the data among multiple partitions, |
| and so on. For example to take a single comprehensive Parquet data file and load it into |
| a partitioned table, you would use an <codeph>INSERT ... SELECT</codeph> statement with |
| dynamic partitioning to let Impala create separate data files with the appropriate |
| partition values; for an example, see <xref |
| href="impala_insert.xml#insert"/>. |
| </note> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_performance"> |
| |
| <title>Query Performance for Impala Parquet Tables</title> |
| |
| <prolog> |
| <metadata> |
| <data name="Category" value="Performance"/> |
| </metadata> |
| </prolog> |
| |
| <conbody> |
| |
| <p> |
| Query performance for Parquet tables depends on the number of columns needed to process |
| the <codeph>SELECT</codeph> list and <codeph>WHERE</codeph> clauses of the query, the |
| way data is divided into <ph rev="parquet_block_size">large data files with block size |
| equal to file size</ph>, the reduction in I/O by reading the data for each column in |
| compressed format, which data files can be skipped (for partitioned tables), and the CPU |
| overhead of decompressing the data for each column. |
| </p> |
| |
| <p> |
| For example, the following is an efficient query for a Parquet table: |
| <codeblock>select avg(income) from census_data where state = 'CA';</codeblock> |
| The query processes only 2 columns out of a large number of total columns. If the table |
| is partitioned by the <codeph>STATE</codeph> column, it is even more efficient because |
| the query only has to read and decode 1 column from each data file, and it can read only |
| the data files in the partition directory for the state <codeph>'CA'</codeph>, skipping |
| the data files for all the other states, which will be physically located in other |
| directories. |
| </p> |
| |
| <p> |
| The following is a relatively inefficient query for a Parquet table: |
| <codeblock>select * from census_data;</codeblock> |
| Impala would have to read the entire contents of each |
| <ph rev="parquet_block_size">large</ph> data file, and decompress the contents of each |
| column for each row group, negating the I/O optimizations of the column-oriented format. |
| This query might still be faster for a Parquet table than a table with some other file |
| format, but it does not take advantage of the unique strengths of Parquet data files. |
| </p> |
| |
| <p> |
| Impala can optimize queries on Parquet tables, especially join queries, better when |
| statistics are available for all the tables. Issue the <codeph>COMPUTE STATS</codeph> |
| statement for each table after substantial amounts of data are loaded into or appended |
| to it. See <xref href="impala_compute_stats.xml#compute_stats"/> for details. |
| </p> |
| |
| <p rev="2.5.0"> |
| The runtime filtering feature, available in <keyword keyref="impala25_full"/> and |
| higher, works best with Parquet tables. The per-row filtering aspect only applies to |
| Parquet tables. See <xref href="impala_runtime_filtering.xml#runtime_filtering"/> for |
| details. |
| </p> |
| |
| <p conref="../shared/impala_common.xml#common/s3_block_splitting"/> |
| <p>Starting in Impala 3.4.0, use the query option |
| <codeph>PARQUET_OBJECT_STORE_SPLIT_SIZE</codeph> to control the |
| Parquet split size for non-block stores (e.g. S3, ADLS, etc.). The |
| default value is 256 MB.</p> |
| |
| <p rev="IMPALA-3909"> |
| In <keyword keyref="impala29"/> and higher, Parquet files written by Impala include |
| embedded metadata specifying the minimum and maximum values for each column, within each |
| row group and each data page within the row group. Impala-written Parquet files |
| typically contain a single row group; a row group can contain many data pages. Impala |
| uses this information (currently, only the metadata for each row group) when reading |
| each Parquet data file during a query, to quickly determine whether each row group |
| within the file potentially includes any rows that match the conditions in the |
| <codeph>WHERE</codeph> clause. For example, if the column <codeph>X</codeph> within a |
| particular Parquet file has a minimum value of 1 and a maximum value of 100, then a |
| query including the clause <codeph>WHERE x > 200</codeph> can quickly determine that |
| it is safe to skip that particular file, instead of scanning all the associated column |
| values. This optimization technique is especially effective for tables that use the |
| <codeph>SORT BY</codeph> clause for the columns most frequently checked in |
| <codeph>WHERE</codeph> clauses, because any <codeph>INSERT</codeph> operation on such |
| tables produces Parquet data files with relatively narrow ranges of column values within |
| each file. |
| </p> |
| <p>To disable Impala from writing the Parquet page index when creating |
| Parquet files, set the <codeph>PARQUET_WRITE_PAGE_INDEX</codeph> query |
| option to <codeph>FALSE</codeph>.</p> |
| |
| </conbody> |
| |
| <concept id="parquet_partitioning"> |
| |
| <title>Partitioning for Parquet Tables</title> |
| |
| <conbody> |
| |
| <p> |
| As explained in <xref href="impala_partitioning.xml#partitioning"/>, partitioning is |
| an important performance technique for Impala generally. This section explains some of |
| the performance considerations for partitioned Parquet tables. |
| </p> |
| |
| <p> |
| The Parquet file format is ideal for tables containing many columns, where most |
| queries only refer to a small subset of the columns. As explained in |
| <xref href="#parquet_data_files"/>, the physical layout of Parquet data files lets |
| Impala read only a small fraction of the data for many queries. The performance |
| benefits of this approach are amplified when you use Parquet tables in combination |
| with partitioning. Impala can skip the data files for certain partitions entirely, |
| based on the comparisons in the <codeph>WHERE</codeph> clause that refer to the |
| partition key columns. For example, queries on partitioned tables often analyze data |
| for time intervals based on columns such as <codeph>YEAR</codeph>, |
| <codeph>MONTH</codeph>, and/or <codeph>DAY</codeph>, or for geographic regions. |
| Remember that Parquet data files use a <ph rev="parquet_block_size">large</ph> block |
| size, so when deciding how finely to partition the data, try to find a granularity |
| where each partition contains <ph rev="parquet_block_size">256 MB</ph> or more of |
| data, rather than creating a large number of smaller files split among many |
| partitions. |
| </p> |
| |
| <p> |
| Inserting into a partitioned Parquet table can be a resource-intensive operation, |
| because each Impala node could potentially be writing a separate data file to HDFS for |
| each combination of different values for the partition key columns. The large number |
| of simultaneous open files could exceed the HDFS <q>transceivers</q> limit. To avoid |
| exceeding this limit, consider the following techniques: |
| </p> |
| |
| <ul> |
| <li> |
| Load different subsets of data using separate <codeph>INSERT</codeph> statements |
| with specific values for the <codeph>PARTITION</codeph> clause, such as |
| <codeph>PARTITION (year=2010)</codeph>. |
| </li> |
| |
| <li> |
| Increase the <q>transceivers</q> value for HDFS, sometimes spelled <q>xcievers</q> |
| (sic). The property value in the <filepath>hdfs-site.xml</filepath> configuration |
| file is <codeph>dfs.datanode.max.transfer.threads</codeph>. For example, if you were |
| loading 12 years of data partitioned by year, month, and day, even a value of 4096 |
| might not be high enough. This |
| <xref |
| keyref="hbase-hadoop-xceivers">blog post</xref> explores the |
| considerations for setting this value higher or lower, using HBase examples for |
| illustration. |
| </li> |
| |
| <li> |
| Use the <codeph>COMPUTE STATS</codeph> statement to collect |
| <xref href="impala_perf_stats.xml#perf_column_stats">column statistics</xref> on the |
| source table from which data is being copied, so that the Impala query can estimate |
| the number of different values in the partition key columns and distribute the work |
| accordingly. |
| </li> |
| </ul> |
| |
| </conbody> |
| |
| </concept> |
| |
| </concept> |
| |
| <concept id="parquet_compression"> |
| |
| <title>Compressions for Parquet Data Files</title> |
| |
| <prolog> |
| <metadata> |
| <data name="Category" value="Snappy"/> |
| <data name="Category" value="Gzip"/> |
| <data name="Category" value="Compression"/> |
| </metadata> |
| </prolog> |
| |
| <conbody> |
| |
| <p> |
| When Impala writes Parquet data files using the <codeph>INSERT</codeph> statement, the |
| underlying compression is controlled by the <codeph>COMPRESSION_CODEC</codeph> query |
| option. (Prior to Impala 2.0, the query option name was |
| <codeph>PARQUET_COMPRESSION_CODEC</codeph>.) The allowed values for this query option |
| are <codeph>snappy</codeph> (the default), <codeph>gzip</codeph>, <codeph>zstd</codeph>, |
| <codeph>lz4</codeph>, and <codeph>none</codeph>. The option value is not case-sensitive. |
| If the option is set to an unrecognized value, all kinds of queries will fail due to |
| the invalid option setting, not just queries involving Parquet tables. |
| </p> |
| |
| </conbody> |
| |
| <concept id="parquet_snappy"> |
| |
| <title>Example of Parquet Table with Snappy Compression</title> |
| |
| <conbody> |
| |
| <p> |
| By default, the underlying data files for a Parquet table are compressed with Snappy. |
| The combination of fast compression and decompression makes it a good choice for many |
| data sets. To ensure Snappy compression is used, for example after experimenting with |
| other compression codecs, set the <codeph>COMPRESSION_CODEC</codeph> query option to |
| <codeph>snappy</codeph> before inserting the data: |
| </p> |
| |
| <codeblock>[localhost:21000] > create database parquet_compression; |
| [localhost:21000] > use parquet_compression; |
| [localhost:21000] > create table parquet_snappy like raw_text_data; |
| [localhost:21000] > set COMPRESSION_CODEC=snappy; |
| [localhost:21000] > insert into parquet_snappy select * from raw_text_data; |
| Inserted 1000000000 rows in 181.98s |
| </codeblock> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_gzip"> |
| |
| <title>Example of Parquet Table with GZip Compression</title> |
| |
| <conbody> |
| |
| <p> |
| If you need more intensive compression (at the expense of more CPU cycles for |
| uncompressing during queries), set the <codeph>COMPRESSION_CODEC</codeph> query option |
| to <codeph>gzip</codeph> before inserting the data: |
| </p> |
| |
| <codeblock>[localhost:21000] > create table parquet_gzip like raw_text_data; |
| [localhost:21000] > set COMPRESSION_CODEC=gzip; |
| [localhost:21000] > insert into parquet_gzip select * from raw_text_data; |
| Inserted 1000000000 rows in 1418.24s |
| </codeblock> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_none"> |
| |
| <title>Example of Uncompressed Parquet Table</title> |
| |
| <conbody> |
| |
| <p> |
| If your data compresses very poorly, or you want to avoid the CPU overhead of |
| compression and decompression entirely, set the <codeph>COMPRESSION_CODEC</codeph> |
| query option to <codeph>none</codeph> before inserting the data: |
| </p> |
| |
| <codeblock>[localhost:21000] > create table parquet_none like raw_text_data; |
| [localhost:21000] > set COMPRESSION_CODEC=none; |
| [localhost:21000] > insert into parquet_none select * from raw_text_data; |
| Inserted 1000000000 rows in 146.90s |
| </codeblock> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_compression_examples"> |
| |
| <title>Examples of Sizes and Speeds for Compressed Parquet Tables</title> |
| |
| <conbody> |
| |
| <p> |
| Here are some examples showing differences in data sizes and query speeds for 1 |
| billion rows of synthetic data, compressed with each kind of codec. As always, run |
| similar tests with realistic data sets of your own. The actual compression ratios, and |
| relative insert and query speeds, will vary depending on the characteristics of the |
| actual data. |
| </p> |
| |
| <p> |
| In this case, switching from Snappy to GZip compression shrinks the data by an |
| additional 40% or so, while switching from Snappy compression to no compression |
| expands the data also by about 40%: |
| </p> |
| |
| <codeblock>$ hdfs dfs -du -h /user/hive/warehouse/parquet_compression.db |
| 23.1 G /user/hive/warehouse/parquet_compression.db/parquet_snappy |
| 13.5 G /user/hive/warehouse/parquet_compression.db/parquet_gzip |
| 32.8 G /user/hive/warehouse/parquet_compression.db/parquet_none |
| </codeblock> |
| |
| <p> |
| Because Parquet data files are typically <ph rev="parquet_block_size">large</ph>, each |
| directory will have a different number of data files and the row groups will be |
| arranged differently. |
| </p> |
| |
| <p> |
| At the same time, the less agressive the compression, the faster the data can be |
| decompressed. In this case using a table with a billion rows, a query that evaluates |
| all the values for a particular column runs faster with no compression than with |
| Snappy compression, and faster with Snappy compression than with Gzip compression. |
| Query performance depends on several other factors, so as always, run your own |
| benchmarks with your own data to determine the ideal tradeoff between data size, CPU |
| efficiency, and speed of insert and query operations. |
| </p> |
| |
| <codeblock>[localhost:21000] > desc parquet_snappy; |
| Query finished, fetching results ... |
| +-----------+---------+---------+ |
| | name | type | comment | |
| +-----------+---------+---------+ |
| | id | int | | |
| | val | int | | |
| | zfill | string | | |
| | name | string | | |
| | assertion | boolean | | |
| +-----------+---------+---------+ |
| Returned 5 row(s) in 0.14s |
| [localhost:21000] > select avg(val) from parquet_snappy; |
| Query finished, fetching results ... |
| +-----------------+ |
| | _c0 | |
| +-----------------+ |
| | 250000.93577915 | |
| +-----------------+ |
| Returned 1 row(s) in 4.29s |
| [localhost:21000] > select avg(val) from parquet_gzip; |
| Query finished, fetching results ... |
| +-----------------+ |
| | _c0 | |
| +-----------------+ |
| | 250000.93577915 | |
| +-----------------+ |
| Returned 1 row(s) in 6.97s |
| [localhost:21000] > select avg(val) from parquet_none; |
| Query finished, fetching results ... |
| +-----------------+ |
| | _c0 | |
| +-----------------+ |
| | 250000.93577915 | |
| +-----------------+ |
| Returned 1 row(s) in 3.67s |
| </codeblock> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_compression_multiple"> |
| |
| <title>Example of Copying Parquet Data Files</title> |
| |
| <conbody> |
| |
| <p> |
| Here is a final example, to illustrate how the data files using the various |
| compression codecs are all compatible with each other for read operations. The |
| metadata about the compression format is written into each data file, and can be |
| decoded during queries regardless of the <codeph>COMPRESSION_CODEC</codeph> setting in |
| effect at the time. In this example, we copy data files from the |
| <codeph>PARQUET_SNAPPY</codeph>, <codeph>PARQUET_GZIP</codeph>, and |
| <codeph>PARQUET_NONE</codeph> tables used in the previous examples, each containing 1 |
| billion rows, all to the data directory of a new table |
| <codeph>PARQUET_EVERYTHING</codeph>. A couple of sample queries demonstrate that the |
| new table now contains 3 billion rows featuring a variety of compression codecs for |
| the data files. |
| </p> |
| |
| <p> |
| First, we create the table in Impala so that there is a destination directory in HDFS |
| to put the data files: |
| </p> |
| |
| <codeblock>[localhost:21000] > create table parquet_everything like parquet_snappy; |
| Query: create table parquet_everything like parquet_snappy |
| </codeblock> |
| |
| <p> |
| Then in the shell, we copy the relevant data files into the data directory for this |
| new table. Rather than using <codeph>hdfs dfs -cp</codeph> as with typical files, we |
| use <codeph>hadoop distcp -pb</codeph> to ensure that the special |
| <ph rev="parquet_block_size"> block size</ph> of the Parquet data files is preserved. |
| </p> |
| |
| <codeblock>$ hadoop distcp -pb /user/hive/warehouse/parquet_compression.db/parquet_snappy \ |
| /user/hive/warehouse/parquet_compression.db/parquet_everything |
| ...<varname>MapReduce output</varname>... |
| $ hadoop distcp -pb /user/hive/warehouse/parquet_compression.db/parquet_gzip \ |
| /user/hive/warehouse/parquet_compression.db/parquet_everything |
| ...<varname>MapReduce output</varname>... |
| $ hadoop distcp -pb /user/hive/warehouse/parquet_compression.db/parquet_none \ |
| /user/hive/warehouse/parquet_compression.db/parquet_everything |
| ...<varname>MapReduce output</varname>... |
| </codeblock> |
| |
| <p> |
| Back in the <cmdname>impala-shell</cmdname> interpreter, we use the |
| <codeph>REFRESH</codeph> statement to alert the Impala server to the new data files |
| for this table, then we can run queries demonstrating that the data files represent 3 |
| billion rows, and the values for one of the numeric columns match what was in the |
| original smaller tables: |
| </p> |
| |
| <codeblock>[localhost:21000] > refresh parquet_everything; |
| Query finished, fetching results ... |
| |
| Returned 0 row(s) in 0.32s |
| [localhost:21000] > select count(*) from parquet_everything; |
| Query finished, fetching results ... |
| +------------+ |
| | _c0 | |
| +------------+ |
| | 3000000000 | |
| +------------+ |
| Returned 1 row(s) in 8.18s |
| [localhost:21000] > select avg(val) from parquet_everything; |
| Query finished, fetching results ... |
| +-----------------+ |
| | _c0 | |
| +-----------------+ |
| | 250000.93577915 | |
| +-----------------+ |
| Returned 1 row(s) in 13.35s |
| </codeblock> |
| |
| </conbody> |
| |
| </concept> |
| |
| </concept> |
| |
| <concept rev="2.3.0" id="parquet_complex_types"> |
| |
| <title>Parquet Tables for Impala Complex Types</title> |
| |
| <conbody> |
| |
| <p conref="../shared/impala_common.xml#common/complex_types_short_intro"/> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_interop"> |
| |
| <title>Exchanging Parquet Data Files with Other Hadoop Components</title> |
| |
| <prolog> |
| <metadata> |
| <data name="Category" value="Hadoop"/> |
| </metadata> |
| </prolog> |
| |
| <conbody> |
| |
| <p> |
| You can read and write Parquet data files from other Hadoop components. See |
| <xref keyref="cdh_ig_parquet"/> for details. |
| </p> |
| |
| <!-- These couple of paragraphs reused in the release notes 'incompatible changes' section. --> |
| |
| <!-- But conbodydiv tag too restrictive, can't have just paragraphs and codeblocks inside. --> |
| |
| <!-- So I will physically copy the info for the time being. --> |
| |
| <!-- <conbodydiv id="upgrade_parquet_metadata"> --> |
| |
| <p> |
| Previously, it was not possible to create Parquet data through Impala and reuse that |
| table within Hive. Now that Parquet support is available for Hive, reusing existing |
| Impala Parquet data files in Hive requires updating the table metadata. Use the |
| following command if you are already running Impala 1.1.1 or higher: |
| </p> |
| |
| <codeblock>ALTER TABLE <varname>table_name</varname> SET FILEFORMAT PARQUET; |
| </codeblock> |
| |
| <p> |
| If you are running a level of Impala that is older than 1.1.1, do the metadata update |
| through Hive: |
| </p> |
| |
| <codeblock>ALTER TABLE <varname>table_name</varname> SET SERDE 'parquet.hive.serde.ParquetHiveSerDe'; |
| ALTER TABLE <varname>table_name</varname> SET FILEFORMAT |
| INPUTFORMAT "parquet.hive.DeprecatedParquetInputFormat" |
| OUTPUTFORMAT "parquet.hive.DeprecatedParquetOutputFormat"; |
| </codeblock> |
| |
| <p> |
| Impala 1.1.1 and higher can reuse Parquet data files created by Hive, without any action |
| required. |
| </p> |
| |
| <!-- </conbodydiv> --> |
| |
| <p rev="2.2.0"> |
| Impala supports the scalar data types that you can encode in a Parquet data file, but |
| not composite or nested types such as maps or arrays. In |
| <keyword keyref="impala22_full"/> and higher, Impala can query Parquet data files that |
| include composite or nested types, as long as the query only refers to columns with |
| scalar types. |
| <!-- TK: could include an example here, but would require setup in Hive or Pig or something. --> |
| </p> |
| |
| <p> |
| If you copy Parquet data files between nodes, or even between different directories on |
| the same node, make sure to preserve the block size by using the command <codeph>hadoop |
| distcp -pb</codeph>. To verify that the block size was preserved, issue the command |
| <codeph>hdfs fsck -blocks <varname>HDFS_path_of_impala_table_dir</varname></codeph> and |
| check that the average block size is at or near <ph rev="parquet_block_size">256 MB (or |
| whatever other size is defined by the <codeph>PARQUET_FILE_SIZE</codeph> query |
| option).</ph>. (The <codeph>hadoop distcp</codeph> operation typically leaves some |
| directories behind, with names matching <filepath>_distcp_logs_*</filepath>, that you |
| can delete from the destination directory afterward.) |
| <!-- The Apache wiki page keeps disappearing, even though Google still points to it as of Nov. 11/2014. --> |
| <!-- Now there is a 'distcp2' guide: http://hadoop.apache.org/docs/r1.2.1/distcp2.html but I haven't tried that so let's play it safe for now and hide the link. --> |
| <!-- See the <xref href="http://hadoop.apache.org/docs/r0.19.0/distcp.html" scope="external" format="html">Hadoop DistCP Guide</xref> for details. --> |
| Issue the command <cmdname>hadoop distcp</cmdname> for details about |
| <cmdname>distcp</cmdname> command syntax. |
| </p> |
| |
| <!-- Sample commands/output for when the 'distcp' business is expanded into a tutorial later. |
| <codeblock>$ hdfs fsck -blocks /user/impala/warehouse/parquet_compression.db/parquet_everything |
| Connecting to namenode via http://a1730.example.com:50070 |
| FSCK started by jrussell (auth:SIMPLE) from /10.20.198.130 for path /user/impala/warehouse/parquet_compression.db/parquet_everything at Fri Aug 23 11:35:37 PDT 2013 |
| ............................................................................Status: HEALTHY |
| Total size: 74504481213 B |
| Total dirs: 1 |
| Total files: 76 |
| Total blocks (validated): 76 (avg. block size 980322121 B) |
| Minimally replicated blocks: 76 (100.0 %) |
| Over-replicated blocks: 0 (0.0 %) |
| Under-replicated blocks: 0 (0.0 %) |
| Mis-replicated blocks: 0 (0.0 %) |
| Default replication factor: 3 |
| Average block replication: 3.0 |
| Corrupt blocks: 0 |
| Missing replicas: 0 (0.0 %) |
| Number of data-nodes: 4 |
| Number of racks: 1 |
| FSCK ended at Fri Aug 23 11:35:37 PDT 2013 in 8 milliseconds |
| |
| |
| The filesystem under path '/user/impala/warehouse/parquet_compression.db/parquet_everything' is HEALTHY |
| </codeblock> |
| --> |
| |
| <p conref="../shared/impala_common.xml#common/impala_parquet_encodings_caveat"/> |
| |
| <p conref="../shared/impala_common.xml#common/parquet_tools_blurb"/> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_data_files"> |
| |
| <title>How Parquet Data Files Are Organized</title> |
| |
| <prolog> |
| <metadata> |
| <data name="Category" value="Concepts"/> |
| </metadata> |
| </prolog> |
| |
| <conbody> |
| |
| <p> |
| Although Parquet is a column-oriented file format, do not expect to find one data file |
| for each column. Parquet keeps all the data for a row within the same data file, to |
| ensure that the columns for a row are always available on the same node for processing. |
| What Parquet does is to set a large HDFS block size and a matching maximum data file |
| size, to ensure that I/O and network transfer requests apply to large batches of data. |
| </p> |
| |
| <p> |
| Within that data file, the data for a set of rows is rearranged so that all the values |
| from the first column are organized in one contiguous block, then all the values from |
| the second column, and so on. Putting the values from the same column next to each other |
| lets Impala use effective compression techniques on the values in that column. |
| </p> |
| |
| <note> |
| <p> |
| Impala <codeph>INSERT</codeph> statements write Parquet data files using an HDFS block |
| size <ph rev="parquet_block_size">that matches the data file size</ph>, to ensure that |
| each data file is represented by a single HDFS block, and the entire file can be |
| processed on a single node without requiring any remote reads. |
| </p> |
| |
| <p> |
| If you create Parquet data files outside of Impala, such as through a MapReduce or Pig |
| job, ensure that the HDFS block size is greater than or equal to the file size, so |
| that the <q>one file per block</q> relationship is maintained. Set the |
| <codeph>dfs.block.size</codeph> or the <codeph>dfs.blocksize</codeph> property large |
| enough that each file fits within a single HDFS block, even if that size is larger |
| than the normal HDFS block size. |
| </p> |
| |
| <p> |
| If the block size is reset to a lower value during a file copy, you will see lower |
| performance for queries involving those files, and the <codeph>PROFILE</codeph> |
| statement will reveal that some I/O is being done suboptimally, through remote reads. |
| See <xref href="impala_parquet.xml#parquet_compression_multiple"/> for an example |
| showing how to preserve the block size when copying Parquet data files. |
| </p> |
| </note> |
| |
| <p> |
| When Impala retrieves or tests the data for a particular column, it opens all the data |
| files, but only reads the portion of each file containing the values for that column. |
| The column values are stored consecutively, minimizing the I/O required to process the |
| values within a single column. If other columns are named in the <codeph>SELECT</codeph> |
| list or <codeph>WHERE</codeph> clauses, the data for all columns in the same row is |
| available within that same data file. |
| </p> |
| |
| <p> |
| If an <codeph>INSERT</codeph> statement brings in less than |
| <ph rev="parquet_block_size">one Parquet block's worth</ph> of data, the resulting data |
| file is smaller than ideal. Thus, if you do split up an ETL job to use multiple |
| <codeph>INSERT</codeph> statements, try to keep the volume of data for each |
| <codeph>INSERT</codeph> statement to approximately <ph rev="parquet_block_size">256 MB, |
| or a multiple of 256 MB</ph>. |
| </p> |
| |
| </conbody> |
| |
| <concept id="parquet_encoding"> |
| |
| <title>RLE and Dictionary Encoding for Parquet Data Files</title> |
| |
| <conbody> |
| |
| <p> |
| Parquet uses some automatic compression techniques, such as run-length encoding (RLE) |
| and dictionary encoding, based on analysis of the actual data values. Once the data |
| values are encoded in a compact form, the encoded data can optionally be further |
| compressed using a compression algorithm. Parquet data files created by Impala can use |
| Snappy, GZip, or no compression; the Parquet spec also allows LZO compression, but |
| currently Impala does not support LZO-compressed Parquet files. |
| </p> |
| |
| <p> |
| RLE and dictionary encoding are compression techniques that Impala applies |
| automatically to groups of Parquet data values, in addition to any Snappy or GZip |
| compression applied to the entire data files. These automatic optimizations can save |
| you time and planning that are normally needed for a traditional data warehouse. For |
| example, dictionary encoding reduces the need to create numeric IDs as abbreviations |
| for longer string values. |
| </p> |
| |
| <p> |
| Run-length encoding condenses sequences of repeated data values. For example, if many |
| consecutive rows all contain the same value for a country code, those repeating values |
| can be represented by the value followed by a count of how many times it appears |
| consecutively. |
| </p> |
| |
| <p> |
| Dictionary encoding takes the different values present in a column, and represents |
| each one in compact 2-byte form rather than the original value, which could be several |
| bytes. (Additional compression is applied to the compacted values, for extra space |
| savings.) This type of encoding applies when the number of different values for a |
| column is less than 2**16 (16,384). It does not apply to columns of data type |
| <codeph>BOOLEAN</codeph>, which are already very short. <codeph>TIMESTAMP</codeph> |
| columns sometimes have a unique value for each row, in which case they can quickly |
| exceed the 2**16 limit on distinct values. The 2**16 limit on different values within |
| a column is reset for each data file, so if several different data files each |
| contained 10,000 different city names, the city name column in each data file could |
| still be condensed using dictionary encoding. |
| </p> |
| |
| </conbody> |
| |
| </concept> |
| |
| </concept> |
| |
| <concept rev="1.4.0" id="parquet_compacting"> |
| |
| <title>Compacting Data Files for Parquet Tables</title> |
| |
| <conbody> |
| |
| <p> |
| If you reuse existing table structures or ETL processes for Parquet tables, you might |
| encounter a <q>many small files</q> situation, which is suboptimal for query efficiency. |
| For example, statements like these might produce inefficiently organized data files: |
| </p> |
| |
| <codeblock>-- In an N-node cluster, each node produces a data file |
| -- for the INSERT operation. If you have less than |
| -- N GB of data to copy, some files are likely to be |
| -- much smaller than the <ph rev="parquet_block_size">default Parquet</ph> block size. |
| insert into parquet_table select * from text_table; |
| |
| -- Even if this operation involves an overall large amount of data, |
| -- when split up by year/month/day, each partition might only |
| -- receive a small amount of data. Then the data files for |
| -- the partition might be divided between the N nodes in the cluster. |
| -- A multi-gigabyte copy operation might produce files of only |
| -- a few MB each. |
| insert into partitioned_parquet_table partition (year, month, day) |
| select year, month, day, url, referer, user_agent, http_code, response_time |
| from web_stats; |
| </codeblock> |
| |
| <p> |
| Here are techniques to help you produce large data files in Parquet |
| <codeph>INSERT</codeph> operations, and to compact existing too-small data files: |
| </p> |
| |
| <ul> |
| <li> |
| <p> |
| When inserting into a partitioned Parquet table, use statically partitioned |
| <codeph>INSERT</codeph> statements where the partition key values are specified as |
| constant values. Ideally, use a separate <codeph>INSERT</codeph> statement for each |
| partition. |
| </p> |
| </li> |
| |
| <li> |
| <p conref="../shared/impala_common.xml#common/num_nodes_tip"/> |
| </li> |
| |
| <li> |
| <p> |
| Be prepared to reduce the number of partition key columns from what you are used to |
| with traditional analytic database systems. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| Do not expect Impala-written Parquet files to fill up the entire Parquet block size. |
| Impala estimates on the conservative side when figuring out how much data to write |
| to each Parquet file. Typically, the of uncompressed data in memory is substantially |
| reduced on disk by the compression and encoding techniques in the Parquet file |
| format. |
| <!-- |
| Impala reserves <ph rev="parquet_block_size">1 GB</ph> of memory to buffer the data before writing, |
| but the actual data file might be smaller, in the hundreds of megabytes. |
| --> |
| The final data file size varies depending on the compressibility of the data. |
| Therefore, it is not an indication of a problem if <ph rev="parquet_block_size">256 |
| MB</ph> of text data is turned into 2 Parquet data files, each less than |
| <ph rev="parquet_block_size">256 MB</ph>. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| If you accidentally end up with a table with many small data files, consider using |
| one or more of the preceding techniques and copying all the data into a new Parquet |
| table, either through <codeph>CREATE TABLE AS SELECT</codeph> or <codeph>INSERT ... |
| SELECT</codeph> statements. |
| </p> |
| |
| <p> |
| To avoid rewriting queries to change table names, you can adopt a convention of |
| always running important queries against a view. Changing the view definition |
| immediately switches any subsequent queries to use the new underlying tables: |
| </p> |
| <codeblock>create view production_table as select * from table_with_many_small_files; |
| -- CTAS or INSERT...SELECT all the data into a more efficient layout... |
| alter view production_table as select * from table_with_few_big_files; |
| select * from production_table where c1 = 100 and c2 < 50 and ...; |
| </codeblock> |
| </li> |
| </ul> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept rev="1.4.0" id="parquet_schema_evolution"> |
| |
| <title>Schema Evolution for Parquet Tables</title> |
| |
| <conbody> |
| |
| <p> |
| Schema evolution refers to using the statement <codeph>ALTER TABLE ... REPLACE |
| COLUMNS</codeph> to change the names, data type, or number of columns in a table. You |
| can perform schema evolution for Parquet tables as follows: |
| </p> |
| |
| <ul> |
| <li> |
| <p> |
| The Impala <codeph>ALTER TABLE</codeph> statement never changes any data files in |
| the tables. From the Impala side, schema evolution involves interpreting the same |
| data files in terms of a new table definition. Some types of schema changes make |
| sense and are represented correctly. Other types of changes cannot be represented in |
| a sensible way, and produce special result values or conversion errors during |
| queries. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| The <codeph>INSERT</codeph> statement always creates data using the latest table |
| definition. You might end up with data files with different numbers of columns or |
| internal data representations if you do a sequence of <codeph>INSERT</codeph> and |
| <codeph>ALTER TABLE ... REPLACE COLUMNS</codeph> statements. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| If you use <codeph>ALTER TABLE ... REPLACE COLUMNS</codeph> to define additional |
| columns at the end, when the original data files are used in a query, these final |
| columns are considered to be all <codeph>NULL</codeph> values. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| If you use <codeph>ALTER TABLE ... REPLACE COLUMNS</codeph> to define fewer columns |
| than before, when the original data files are used in a query, the unused columns |
| still present in the data file are ignored. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| Parquet represents the <codeph>TINYINT</codeph>, <codeph>SMALLINT</codeph>, and |
| <codeph>INT</codeph> types the same internally, all stored in 32-bit integers. |
| </p> |
| <ul> |
| <li> |
| That means it is easy to promote a <codeph>TINYINT</codeph> column to |
| <codeph>SMALLINT</codeph> or <codeph>INT</codeph>, or a <codeph>SMALLINT</codeph> |
| column to <codeph>INT</codeph>. The numbers are represented exactly the same in |
| the data file, and the columns being promoted would not contain any out-of-range |
| values. |
| </li> |
| |
| <li> |
| <p> |
| If you change any of these column types to a smaller type, any values that are |
| out-of-range for the new type are returned incorrectly, typically as negative |
| numbers. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| You cannot change a <codeph>TINYINT</codeph>, <codeph>SMALLINT</codeph>, or |
| <codeph>INT</codeph> column to <codeph>BIGINT</codeph>, or the other way around. |
| Although the <codeph>ALTER TABLE</codeph> succeeds, any attempt to query those |
| columns results in conversion errors. |
| </p> |
| </li> |
| |
| <li> |
| <p> |
| Any other type conversion for columns produces a conversion error during |
| queries. For example, <codeph>INT</codeph> to <codeph>STRING</codeph>, |
| <codeph>FLOAT</codeph> to <codeph>DOUBLE</codeph>, <codeph>TIMESTAMP</codeph> to |
| <codeph>STRING</codeph>, <codeph>DECIMAL(9,0)</codeph> to |
| <codeph>DECIMAL(5,2)</codeph>, and so on. |
| </p> |
| </li> |
| </ul> |
| </li> |
| </ul> |
| |
| <p rev="2.6.0 IMPALA-2835"> |
| You might find that you have Parquet files where the columns do not line up in the same |
| order as in your Impala table. For example, you might have a Parquet file that was part |
| of a table with columns <codeph>C1,C2,C3,C4</codeph>, and now you want to reuse the same |
| Parquet file in a table with columns <codeph>C4,C2</codeph>. By default, Impala expects |
| the columns in the data file to appear in the same order as the columns defined for the |
| table, making it impractical to do some kinds of file reuse or schema evolution. In |
| <keyword keyref="impala26_full"/> and higher, the query option |
| <codeph>PARQUET_FALLBACK_SCHEMA_RESOLUTION=name</codeph> lets Impala resolve columns by |
| name, and therefore handle out-of-order or extra columns in the data file. For example: |
| <codeblock conref="../shared/impala_common.xml#common/parquet_fallback_schema_resolution_example"/> |
| See |
| <xref href="impala_parquet_fallback_schema_resolution.xml#parquet_fallback_schema_resolution"/> |
| for more details. |
| </p> |
| |
| </conbody> |
| |
| </concept> |
| |
| <concept id="parquet_data_types"> |
| |
| <title>Data Type Considerations for Parquet Tables</title> |
| |
| <conbody> |
| |
| <p> |
| The Parquet format defines a set of data types whose names differ from the names of the |
| corresponding Impala data types. If you are preparing Parquet files using other Hadoop |
| components such as Pig or MapReduce, you might need to work with the type names defined |
| by Parquet. The following tables list the Parquet-defined types and the equivalent types |
| in Impala. |
| </p> |
| |
| <p> |
| <b>Primitive types</b> |
| </p> |
| |
| <simpletable frame="all" id="simpletable_am3_rxn_wgb"> |
| |
| <sthead> |
| |
| <stentry>Parquet type</stentry> |
| |
| <stentry>Impala type</stentry> |
| |
| </sthead> |
| |
| <strow> |
| |
| <stentry>BINARY</stentry> |
| |
| <stentry>STRING</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>BOOLEAN</stentry> |
| |
| <stentry>BOOLEAN</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>DOUBLE</stentry> |
| |
| <stentry>DOUBLE</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>FLOAT</stentry> |
| |
| <stentry>FLOAT</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>INT32</stentry> |
| |
| <stentry>INT</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>INT64</stentry> |
| |
| <stentry>BIGINT</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>INT96</stentry> |
| |
| <stentry>TIMESTAMP</stentry> |
| |
| </strow> |
| |
| </simpletable> |
| |
| <p> |
| <b>Logical types</b> |
| </p> |
| |
| <p> |
| Parquet uses type annotations to extend the types that it can store, by specifying how |
| the primitive types should be interpreted. |
| </p> |
| |
| <simpletable frame="all" id="simpletable_az3_byn_wgb"> |
| |
| <sthead> |
| |
| <stentry>Parquet primitive type and annotation</stentry> |
| |
| <stentry>Impala type</stentry> |
| |
| </sthead> |
| |
| <strow> |
| |
| <stentry>BINARY annotated with the UTF8 OriginalType</stentry> |
| |
| <stentry>STRING</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>BINARY annotated with the STRING LogicalType</stentry> |
| |
| <stentry>STRING</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>BINARY annotated with the ENUM OriginalType</stentry> |
| |
| <stentry>STRING</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>BINARY annotated with the DECIMAL OriginalType</stentry> |
| |
| <stentry>DECIMAL</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>INT64 annotated with the TIMESTAMP_MILLIS |
| OriginalType</stentry> |
| |
| <stentry>TIMESTAMP (in <keyword keyref="impala32"/> or |
| higher)<p> |
| or |
| </p>BIGINT (for backward compatibility)</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>INT64 annotated with the TIMESTAMP_MICROS |
| OriginalType</stentry> |
| |
| <stentry>TIMESTAMP (in <keyword keyref="impala32"/> or |
| higher)<p> |
| or |
| </p>BIGINT (for backward compatibility)</stentry> |
| |
| </strow> |
| |
| <strow> |
| |
| <stentry>INT64 annotated with the TIMESTAMP LogicalType</stentry> |
| |
| <stentry>TIMESTAMP (in <keyword keyref="impala32"/> or |
| higher)<p> |
| or |
| </p>BIGINT (for backward compatibility)</stentry> |
| |
| </strow> |
| |
| </simpletable> |
| |
| <p rev="2.3.0"> |
| <b>Complex types:</b> |
| </p> |
| |
| <p rev="2.3.0"> |
| For the complex types (<codeph>ARRAY</codeph>, <codeph>MAP</codeph>, and |
| <codeph>STRUCT</codeph>) available in <keyword keyref="impala23_full"/> and higher, |
| Impala only supports queries against those types in Parquet tables. |
| </p> |
| |
| </conbody> |
| |
| </concept> |
| |
| </concept> |