Cloning tables (LIKE clause):

To create an empty table with the same columns, comments, and other attributes as another table, use the following variation. The CREATE TABLE ... LIKE form allows a restricted set of clauses, currently only the LOCATION, COMMENT, and STORED AS clauses.

When you clone the structure of an existing table using the CREATE TABLE ... LIKE syntax, the new table keeps the same file format as the original one, so you only need to specify the STORED AS clause if you want to use a different file format, or when specifying a view as the original table. (Creating a table like a view produces a text table by default.)

Although normally Impala cannot create an HBase table directly, Impala can clone the structure of an existing HBase table with the CREATE TABLE ... LIKE syntax, preserving the file format and metadata from the original table.

There are some exceptions to the ability to use CREATE TABLE ... LIKE with an Avro table. For example, you cannot use this technique for an Avro table that is specified with an Avro schema but no columns. When in doubt, check if a CREATE TABLE ... LIKE operation works in Hive; if not, it typically will not work in Impala either.

If the original table is partitioned, the new table inherits the same partition key columns. Because the new table is initially empty, it does not inherit the actual partitions that exist in the original one. To create partitions in the new table, insert data or issue ALTER TABLE ... ADD PARTITION statements.

Because CREATE TABLE ... LIKE only manipulates table metadata, not the physical data of the table, issue INSERT INTO TABLE statements afterward to copy any data from the original table into the new one, optionally converting the data to a new file format. (For some file formats, Impala can do a CREATE TABLE ... LIKE to create the table, but Impala cannot insert data in that file format; in these cases, you must load the data in Hive. See for details.)

CREATE TABLE AS SELECT:

The CREATE TABLE AS SELECT syntax is a shorthand notation to create a table based on column definitions from another table, and copy data from the source table to the destination table without issuing any separate INSERT statement. This idiom is so popular that it has its own acronym, CTAS.

The following examples show how to copy data from a source table T1 to a variety of destinations tables, applying various transformations to the table properties, table layout, or the data itself as part of the operation:

See for details about query syntax for the SELECT portion of a CREATE TABLE AS SELECT statement.

The newly created table inherits the column names that you select from the original table, which you can override by specifying column aliases in the query. Any column or table comments from the original table are not carried over to the new table.

Prior to you could use a partitioned table as the source and copy data from it, but could not specify any partitioning clauses for the new table. In and higher, you can now use the PARTITIONED BY clause with a CREATE TABLE AS SELECT statement. The following example demonstrates how you can copy data from an unpartitioned table in a CREATE TABLE AS SELECT operation, creating a new partitioned table in the process. The main syntax consideration is the column order in the PARTITIONED BY clause and the select list: the partition key columns must be listed last in the select list, in the same order as in the PARTITIONED BY clause. Therefore, in this case, the column order in the destination table is different from the source table. You also only specify the column names in the PARTITIONED BY clause, not the data types or column comments.

The most convenient layout for partitioned tables is with all the partition key columns at the end. The CTAS PARTITIONED BY syntax requires that column order in the select list, resulting in that same column order in the destination table.

Attempting to use a select list with the partition key columns not at the end results in an error due to a column name mismatch:

For example, the following statements show how you can clone all the data in a table, or a subset of the columns and/or rows, or reorder columns, rename them, or construct them out of expressions:

As part of a CTAS operation, you can convert the data to any file format that Impala can write (currently, TEXTFILE and PARQUET). You cannot specify the lower-level properties of a text table, such as the delimiter.

CREATE TABLE LIKE PARQUET:

The variation CREATE TABLE ... LIKE PARQUET 'hdfs_path_of_parquet_file' lets you skip the column definitions of the CREATE TABLE statement. The column names and data types are automatically configured based on the organization of the specified Parquet data file, which must already reside in HDFS. You can use a data file located outside the Impala database directories, or a file from an existing Impala Parquet table; either way, Impala only uses the column definitions from the file and does not use the HDFS location for the LOCATION attribute of the new table. (Although you can also specify the enclosing directory with the LOCATION attribute, to both use the same schema as the data file and point the Impala table at the associated directory for querying.)

The following considerations apply when you use the CREATE TABLE LIKE PARQUET technique:

• Any column comments from the original table are not preserved in the new table. Each column in the new table has a comment stating the low-level Parquet field type used to deduce the appropriate SQL column type.
• If you use a data file from a partitioned Impala table, any partition key columns from the original table are left out of the new table, because they are represented in HDFS directory names rather than stored in the data file. To preserve the partition information, repeat the same PARTITION clause as in the original CREATE TABLE statement.
• The file format of the new table defaults to text, as with other kinds of CREATE TABLE statements. To make the new table also use Parquet format, include the clause STORED AS PARQUET in the CREATE TABLE LIKE PARQUET statement.
• If the Parquet data file comes from an existing Impala table, currently, any TINYINT or SMALLINT columns are turned into INT columns in the new table. Internally, Parquet stores such values as 32-bit integers.
• When the destination table uses the Parquet file format, the CREATE TABLE AS SELECT and INSERT ... SELECT statements always create at least one data file, even if the SELECT part of the statement does not match any rows. You can use such an empty Parquet data file as a template for subsequent CREATE TABLE LIKE PARQUET statements.

For more details about creating Parquet tables, and examples of the CREATE TABLE LIKE PARQUET syntax, see .

Visibility and Metadata (TBLPROPERTIES and WITH SERDEPROPERTIES clauses):

You can associate arbitrary items of metadata with a table by specifying the TBLPROPERTIES clause. This clause takes a comma-separated list of key-value pairs and stores those items in the metastore database. You can also change the table properties later with an ALTER TABLE statement. You can observe the table properties for different delimiter and escape characters using the DESCRIBE FORMATTED command, and change those settings for an existing table with ALTER TABLE ... SET TBLPROPERTIES.

You can also associate SerDes properties with the table by specifying key-value pairs through the WITH SERDEPROPERTIES clause. This metadata is not used by Impala, which has its own built-in serializer and deserializer for the file formats it supports. Particular property values might be needed for Hive compatibility with certain variations of file formats, particularly Avro.

Some DDL operations that interact with other Hadoop components require specifying particular values in the SERDEPROPERTIES or TBLPROPERTIES fields, such as creating an Avro table or an HBase table. (You typically create HBase tables in Hive, because they require additional clauses not currently available in Impala.)

To see the column definitions and column comments for an existing table, for example before issuing a CREATE TABLE ... LIKE or a CREATE TABLE ... AS SELECT statement, issue the statement DESCRIBE table_name. To see even more detail, such as the location of data files and the values for clauses such as ROW FORMAT and STORED AS, issue the statement DESCRIBE FORMATTED table_name. DESCRIBE FORMATTED is also needed to see any overall table comment (as opposed to individual column comments).

After creating a table, your impala-shell session or another impala-shell connected to the same node can immediately query that table. There might be a brief interval (one statestore heartbeat) before the table can be queried through a different Impala node. To make the CREATE TABLE statement return only when the table is recognized by all Impala nodes in the cluster, enable the SYNC_DDL query option.

HDFS caching (CACHED IN clause):

If you specify the CACHED IN clause, any existing or future data files in the table directory or the partition subdirectories are designated to be loaded into memory with the HDFS caching mechanism. See for details about using the HDFS caching feature.

Column order:

If you intend to use the table to hold data files produced by some external source, specify the columns in the same order as they appear in the data files.

If you intend to insert or copy data into the table through Impala, or if you have control over the way externally produced data files are arranged, use your judgment to specify columns in the most convenient order:

• If certain columns are often NULL, specify those columns last. You might produce data files that omit these trailing columns entirely. Impala automatically fills in the NULL values if so.

• If an unpartitioned table will be used as the source for an INSERT ... SELECT operation into a partitioned table, specify last in the unpartitioned table any columns that correspond to partition key columns in the partitioned table, and in the same order as the partition key columns are declared in the partitioned table. This technique lets you use INSERT ... SELECT * when copying data to the partitioned table, rather than specifying each column name individually.

• If you specify columns in an order that you later discover is suboptimal, you can sometimes work around the problem without recreating the table. You can create a view that selects columns from the original table in a permuted order, then do a SELECT * from the view. When inserting data into a table, you can specify a permuted order for the inserted columns to match the order in the destination table.

Impala queries can make use of metadata about the table and columns, such as the number of rows in a table or the number of different values in a column. Prior to Impala 1.2.2, to create this metadata, you issued the ANALYZE TABLE statement in Hive to gather this information, after creating the table and loading representative data into it. In Impala 1.2.2 and higher, the COMPUTE STATS statement produces these statistics within Impala, without needing to use Hive at all.

The Impala CREATE TABLE statement cannot create an HBase table, because it currently does not support the STORED BY clause needed for HBase tables. Create such tables in Hive, then query them through Impala. For information on using Impala with HBase tables, see .