IBM Cloud Docs
SQL reference

SQL reference

Introduction

With IBM Cloud® Data Engine, you can analyze and transform open data with SQL. It supports the various types of SELECT statements from the ANSI SQL standard.

The SELECT statement (or query statement) is used to read object data from IBM Cloud® Object Storage (COS), process the data, and store it back on Cloud Object Storage eventually.

You can use Data Engine as a data transformation service, as it always writes the results of a query to a specified location in either Object Storage or Db2 tables. Data Engine provides extended SQL syntax inside a special INTO clause to control how the result data is stored physically. This extended SQL syntax includes control over data location, format, layout, and partitioning.

A query statement can be submitted through Data Engine's web UI or programmatically, either by using the service's REST API, or by using the Python or Node.JS SDK. You can also use IBM Watson® Studio and the Python SDK to use Data Engine interactively with Jupyter Notebooks. In addition, you can submit SQL queries that use IBM Cloud® Functions.

In addition to the ad hoc usage of data in IBM Cloud® Object Storage, you can also register and manage your data in a catalog as tables, consisting of columns and partitions.

Several benefits to cataloging your data exist:

  • It simplifies SQL SELECT statements because the SQL author does not need not know and specify exactly where and how the data is stored.
  • The SQL execution can skip the inference of schema and partitioning because this information is available in the metastore. Thus, cataloging improves your query performance, especially for text-based data formats, such as CSV and JSON, where the schema inference requires a full scan of the data before the actual query execution.

Select

See the following examples for an outline of the general syntax of an SQL query statement that uses the query clause and the namedQuery clause.

query

namedQuery

intoClause

The query statement supports common table expressions. A common table expression permits defining a result table with a table name that can be specified as a table name in any FROM clause of the fullselect that follows.

Common table expressions are defined by using the reserved keyword WITH followed by one or more named queries. Each common table expression that is specified can also be referenced by name in the FROM clause of subsequent common table expressions.

Creating a common table expression avoids the overhead of creating and dropping an intermediate result object on Cloud Object Storage that is needed only for a certain query.

Moreover, a common table expression is beneficial when the same result table must be shared in a fullselect.

Examples

The common table expression examples use values clauses to define tables inline. For more information about the values clause, see valuesClause.

-- find the department with the highest total pay
WITH dtotal AS (
    SELECT
        col1 AS deptno,
        SUM(col3+col4) AS totalpay
    FROM  VALUES -- deptno, empid, salary, bonus
        (1, 1, 1000, 0), (1, 2, 2000, 500), (1, 3, 3000, 0),
        (2, 4, 5000, 200), (2, 5, 6000, 0), (2, 6, 4000, 0),
        (3, 7, 2000, 500), (3, 8, 2000, 500), (3, 9, 8000, 0)
    GROUP BY col1
)
SELECT deptno
FROM dtotal
WHERE totalpay = (SELECT MAX(totalpay) FROM dtotal)

The result of the example queries is shown in the following table.

Query result for the example find
DEPTNO
2
-- list products with a price above the average price
WITH products AS (
    SELECT
        col1 AS productid,
        col2 AS price
    FROM VALUES -- productid, price
        (1, 10), (2, 20), (3, 30), (4, 40),
        (5, 5), (6, 10), (7, 15), (8, 40),
        (9, 100), (10, 200), (11, 300), (12, 400)
), avg_product_price AS (
    SELECT
        AVG(price) as avg_price
    FROM products
)
SELECT
    productid,
	price,
	(SELECT * FROM avg_product_price) AS avg_price,
	price-(SELECT * FROM avg_product_price) AS price_dist
FROM products WHERE price > (SELECT * FROM avg_product_price)
Query result for the example list
PRODUCTID PRICE AVG_PRICE PRICE_DIST
9 100 97.5 2.5
10 200 97.5 102.5
11 300 97.5 202.5
12 400 97.5 302.5

cosResultClause

You can use the Cloud Object Storage result clause to apply detailed control over the location, format, and layout of the SQL query result set being stored on Cloud Object Storage.

The default is JOBPREFIX JOBID, which means that jobid= is always appended to the target prefix. You can optionally specify JOBID NONE, which skips the appending of jobid=. The results are then written exactly to the requested path. However, if that path contains existing objects (for example, from a previous query execution with the same target path), all existing objects get removed when the new result objects are written.

A query cannot run if JOBPREFIX NONE is specified and the target overlaps with at least one of the input URIs. The reason is that data cannot be overwritten by a query that is reading that same data. For example, SELECT * FROM cos://us-geo/mybucket/myprefix/mysubprefix INTO cos://us-geo/mybucket/myprefix JOBPREFIX NONE returns an error when you try to submit it.

In the COS result clause, you can explicitly specify the storage location and type of a query result on Cloud Object Storage. The storage location is specified by a COSURI.

Valid query result object formats are: AVRO, CSV, JSON, ORC, and PARQUET. Object type names are not case-sensitive.

Being able to explicitly specify the location and the type of the result object enables a user to take the following actions:

  • Read raw data stored in one format, for example, CSV.
  • Cleanse the data by using SQL constructs.
  • Store the cleansed data in a format that offers benefits regarding query performance and storage consumption on Cloud Object Storage, for example, a column-based object format, such as Parquet.

Moreover, a user can explicitly define the way a query result is stored physically on Cloud Object Storage by using the result partitioned clause.

As shown in the syntax diagrams, three main use cases exist to define the physical layout of a query's result on Cloud Object Storage:

  • Partition by columns, which is the so-called Hive-style partitioning.
  • Partition into buckets or objects (both terms can be used synonymously), which generates the query result into objects, with or without specifying columns.
  • Partition by number of rows.

A partition is an object on Cloud Object Storage that is potentially a part of an aggregated object. The presence of multiple partitions allows for parallel input/output (I/O) during query execution. If no result partitioned clause is specified, the query result is stored in a single partition on Cloud Object Storage.

partitionedClause

You can use the result partitioned clause to control the layout of the SQL query result set being stored. The default behavior is to store the result into one single partition, which is a single object in Cloud Object Storage.

sortClause

This clause can be used to sort in many ways. When specified in combination with PARTITIONED BY, it sorts the rows within each partition by the sort order that is specified in the SORT BY clause. When specified in combination with PARTITIONED INTO, the same is done, which is often referred to as clustering the rows by the specified columns into the fixed number of partitions specified by PARTITIONED INTO. When specified without the PARTITIONED clause, it is equivalent to an ORDER BY clause specified at the top level of the SQL SELECT statement. If PARTITIONED INTO is specified, the ORDER BY clause is ignored.

Partition by columns

When you use the PARTITIONED BY (column-list) clause without specifying INTO x BUCKETS/OBJECTS, you can store the query result by using Hive-style partitioning, which is to create partitions that contain only rows that have certain values for one or more columns. Choose this physical layout if the stored object is further analyzed by using SQL queries that specify predicates on the partition columns.

For example, a result object that contains worldwide order data has a column country to represent the country that the order is initiated from. Partitioning the result object by the column PARTITIONED BY (country), would create a result object with a partition for each country present in the query result.

When the result object is stored this way on Cloud Object Storage, each SQL query that contains a predicate, such as country = 'USA' or country in ('MALTA', 'ITALY', 'VATICAN CITY'), benefits from this physical layout. The reason is that during SQL query execution partitions must be read only if they contain data for the countries of interest. This layout tremendously cuts down the I/O traffic of the SQL query.

See the following extra remarks on Hive-style partitioning.

  • Hive-style partitions have an eye-catching naming scheme because the column names that are used for partitioning are part of the partition object prefix, for example, /order/COUNTRY=USA/part-m-00000.snappy.parquet.
  • Hive-style partitions do not contain any values for partition columns since their values are stored in the object prefix of the partition. Thus, if you copy a HIVE-style partition and rename the object prefix by removing the partition column values, you lose data.
  • Hive-style partitions have a tendency for data skewing. For example, the partition that represents order data from Malta is likely much smaller than the partition that represents order data from the US. You can partition the query result into separate objects if you want to have equally sized partitions.

Partition by columns into objects

By partitioning a query result into objects, you can specify the exact number of equally sized result partitions. With this partitioning, you can experimentally fine-tune the number of objects to meet certain criteria for partition size. To specify the number of partitions, use the PARTITIONED INTO x BUCKETS/OBJECTS clause.

For example, knowing the size of the query result, it is possible to calculate the number of objects to end up with partitions that have a certain size. For example, 128 MB, which is the Hadoop default file size, or any other size that meets application requirements.

The INTO x BUCKETS/OBJECTS clause can be combined with the BY (column-list) clause to create some partitions that support data affinity regarding specified partition columns.

Continue with the preceding example that specifies PARTITION BY (customerid) INTO 10 OBJECTS stores the query result into 10 objects, which ensures that all data for a customer is stored in the same partition. Although it is ensured that all data for a certain customer is stored in the same partition, it is not ensured that the data is physically sorted according to the specified column.

Partition by number of rows

By partitioning by number of rows you can specify the number of rows that go into a single partition. To specify the number of rows stored in each partition, use the EVERY x ROWS clause. In case the row length of the result object is not varying heavily, with the partition every rows clause you can also create almost equally sized result partitions.

The use of the PARTITIONED EVERY x ROWS clause on a sorted query result ensures that partitions are created to have some rows that are sorted according to the query's SORT BY clause. This physical layout can be useful to create partitions that are processed by an application in a pagination manner, for example, browsing order data sorted by order date and customer ID.

The use of PARTITIONED EVERY x ROWS clause causes data to be written single-threaded to Cloud Object Storage, which means that no parallel I/O is performed to write query results to Cloud Object Storage.

dbResultClause

You can use the Db result clause to specify that you want query results to be stored as a relational database table in IBM Cloud. Currently, IBM® Db2® on Cloud is the only supported target database.

Storing query results in a database creates a new table with the columns that are determined by the query result. When you write to Db2, the following type-mapping rules apply.

  • String types are mapped to VARCHAR(32000) columns.
  • Struct types are not mapped and must be flattened first. See the FLATTEN table transformation function.
  • Arrays, time series, and spatial data types are not mapped and must be converted with appropriate SQL functions.

The table name and optional schema are specified as part of the target URI. Important: If a table with the name that is indicated exists in the target database, that table is dropped before the query executes and all existing data is deleted.

Use the PARALLELISM x clause to specify that multiple parallel database connections are to be opened to write out the result. Depending on the size of your result and the network connectivity of your target database service, this clause can reduce the query processing time significantly.

accessSecrets

By default, either the credentials that are needed to access the target database are taken from the credentials object of a CRN_URI, or the IAM user who submits the statement is used to connect to the DB2_TABLE_URI. You can override this default by specifying either a combination of USER and PASSWORD or an APIKEY. However, the password or API key is not included in the SQL statement as plain text. Instead, you must store it as a custom key in a IBM® Key Protect for IBM Cloud® instance to which you have access. For a description how to store and manage the secrets in Key Protect, see Setting up custom secrets in Key Protect.

More topics - query

For more information about clauses that are used in a query, see the following topics:

Fullselect clause

fullselect

A fullselect is the core component of a query. It is the only mandatory general component for a valid query statement. The other components outside of fullselect are optional. Its syntax is defined by the following syntax diagram.

The result set defined by a single fullselect can be combined with the result set of one or more other fullselects by using set operators.

The following set operators are supported and each set operator derives a result set from two other result sets R1 and R2:

  • INTERSECT: The result consists of all rows in both R1 and R2.
  • UNION: The result consists of all rows in R1 and all rows in R2.
  • EXCEPT: The result consists of all rows that do not have a corresponding row in R2.
  • MINUS: The minus operator is a synonym for the except operator and is supported for compatibility with other SQL implementations.

These set operators can be further refined by using the following modifiers:

  • DISTINCT: This modifier ensures that the overall result set does not contain any duplicates. It is the default modifier that applies if no modifier is present.
  • ALL: All rows of a fullselect's result set are combined by using a set operator. Thus, the overall result set can contain duplicates.

The following combinations of set operators and set modifiers are not supported:

  • INTERSECT ALL
  • EXCEPT ALL
  • MINUS ALL

The characteristics of a result set defined by a fullselect can be further defined by using the following clauses:

  • ORDER BY: Define an overall ordering of the result set based on the criteria that are defined by the list of sortItem clauses. The default order direction is ascending if not explicitly specified by a sortItem clause. The order by clause cannot be used with cluster by, distribute by, or sort by clause. When you use partitioned output, the ORDER BY clause gets ignored. Use the sortClause instead.
  • DISTRIBUTE BY: Distribute result set rows into new partitions based on the criteria that are defined by the list of expression clauses. Result set rows that have the same expression values are moved to the same partition. The distribute by clause cannot be used with order by or cluster by clause.
  • SORT BY: You can define an overall ordering on a partition base as defined by the list of expression clauses. The default order direction is ascending if not explicitly specified by an expression clause. The sort by clause is used along with the distribute by clause.
  • CLUSTER BY: Distribute result set rows into new partitions based on the criteria that are defined by the list of expression clauses. Moreover, each partition is sorted in ascending order based on the criteria that are defined by the set of expression clauses. Thus, this clause represents a shortcut for distribute by clause that is combined with sort by in ascending order. You cannot use the cluster by attribute with the order by, distribute by, sort by clause.
  • LIMIT: Restrict the number of rows that are returned from the result set of the fullselect. The number of rows can be defined by an expression or by using the keyword ALL that causes all rows to be returned.

DISTRIBUTE BY, SORT BY, and CLUSTER BY have an effect only during your SQL query execution and do not influence the query result that is written back to Cloud Object Storage. Use these clauses only in execution of subqueries to optimize the outer query execution that works on the intermediate result sets produced by the sub queries. To define the persistent target of the overall query that is written back to Cloud Object Storage, you need to use the dedicated resultClause instead.

Examples - values clause

The set operator examples use values clauses to define result sets for the set operations. For more information about the values clause, see valuesClause.

-- set union eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 UNION VALUES 1, 2, 3
-- set union eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 UNION DISTINCT VALUES 1, 2, 3

The result of the example queries is shown in the following table.

Query result for the example set
COL1
1
3
2
-- set union with duplicate rows
SELECT * FROM VALUES 1, 2 ,3 UNION ALL VALUES 1, 2, 3

The result of the example query is shown in the following table.

Query result for the example set
COL1
1
2
3
1
2
3
-- intersecting set eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 INTERSECT VALUES 2, 2, 3, 3, 3
-- intersecting set eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 INTERSECT DISTINCT VALUES 2, 2, 3, 3, 3

The result of the example queries is shown in the following table.

Query result for the example intersecting
COL1
3
2
-- Difference quantity eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 EXCEPT VALUES 2, 2, 3, 3, 3
-- difference quantity eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 EXCEPT DISTINCT VALUES 2, 2, 3, 3, 3
-- difference quantity eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 MINUS VALUES 2, 2, 3, 3, 3
-- difference quantity eliminating duplicate rows
SELECT * FROM VALUES 1, 2 ,3 MINUS DISTINCT VALUES 2, 2, 3, 3, 3

The result of the example queries is shown in the following table.

Query result for the example difference
COL1
1

More topics - fullselect

For more information about the clauses that are used in a fullselect, see the following topics:

Simpleselect Clause

A simpleselect is a component of a fullselect. Its syntax is defined by the following syntax diagram.

simpleselect

With a simpleselect, you can specify the following characteristics of a result set:

  • The list of result columns from relations or lateral views that are part of the final result set. The result column list can be further redefined by using the following modifier keywords:
    • DISTINCT: Eliminates all but one of each set of duplicate rows of the final result set.
    • ALL: Retains all rows of the final result set, and does not eliminate redundant duplicates. It is the default.
  • The FROM clause defines the list of relations or lateral views that are combined to derive a result set.
  • The WHERE clause defines the way how relations and lateral views are filtered and joined to derive a result set.
  • In its simplest form, the GROUP BY clause defines how rows that qualify for the final result set are grouped based on grouping expressions. Each group is represented by a single row in the final result set.
  • The HAVING clause is used with the group by clause to filter out groups from the final result set.

resultColumn

A result column can be any expression that can optionally be associated with an identifier, which is a new name. By providing custom identifiers, you can control the column names that are used in the result data set written to Cloud Object Storage. See the following examples for such expressions:

  • A column name from one of the relations.
  • An expression that uses column names from one or multiple relations.
  • Arithmetic expressions that perform calculations.

Group by clause

groupByClause

groupingSet

More complex group by clauses use so called grouping sets to provide more insights into the set of rows grouped by a grouping expression.

Grouping sets are best explained by using examples. The following set examples use values clauses to define result sets for group by operations. For more information about the values clause, see valuesClause.

In the following example SQL query, sales data is grouped per year and quarter.

-- grouping sales data per year and quarter
SELECT
    sales.col1 AS year,
    sales.col2 AS quarter,
    SUM(sales.col3) AS amount
FROM VALUES
    (2017, 1 ,100),
    (2017, 1 ,50),
    (2017, 2 ,200),
    (2017, 2 ,300),
    (2018, 1 ,300),
    (2018, 1 ,100),
    (2018, 2 ,400) AS sales
GROUP BY col1, col2
ORDER BY year, quarter

The result of the example query is shown in the following table.

Query result for the example grouping
YEAR QUARTER AMOUNT
2017 1 150
2017 2 500
2018 1 400
2018 2 400

A ROLLUP grouping is an extension to the group by clause that produces a result set containing subtotal rows in addition to the regular grouped rows. A GROUP BY COL1, COL2 WITH ROLLUP generates the following grouping sets: (COL1, COL2), (COL1), (). The N grouping expressions of the ROLLUP convert to N+1 grouping sets.

Referring to the preceding example, adding a WITH ROLLUP modifier to the group by clause computes rollup sales data on quarter by year basis, a yearly basis, and a grand total as shown by the following example.

-- rollup sales data on quarter by year basis, a yearly basis, and a grand total
SELECT
    sales.col1 AS year,
    sales.col2 AS quarter,
    SUM(sales.col3) AS amount
FROM VALUES
    (2017, 1 ,100),
    (2017, 1 ,50),
    (2017, 2 ,200),
    (2017, 2 ,300),
    (2018, 1 ,300),
    (2018, 1 ,100),
    (2018, 2 ,400) AS sales
GROUP BY col1, col2
WITH ROLLUP
ORDER BY year, quarter

The result of the example query is shown in the following table.

Query result for the example rollup
YEAR QUARTER AMOUNT
null null 1450
2017 null 650
2017 1 150
2017 2 500
2018 null 800
2018 1 400
2018 2 400

A CUBE grouping is an extension to the group by clause that produces a result set that contains all the rows of a ROLLUP aggregation and in addition, grouping sets that do not represent a subtotal or grand total. A GROUPY BY COL1, COL2 WITH CUBE generates the following grouping sets: (COL1, COL2), (COL1), (COL2), (). The N elements of a CUBE convert to 2**N (2 to the power N) grouping sets.

Referring to the preceding example, adding a WITH CUBE modifier to the group by clause computes rollup sales data on a quarter by year basis, a yearly basis, a quarterly year-independent basis, and a grand total as shown by the following example.

-- rollup sales data on a quarter by year basis, a yearly basis,
-- a quarterly year-independent basis and a grand total
SELECT
    sales.col1 AS year,
    sales.col2 AS quarter,
    SUM(sales.col3) AS amount
FROM VALUES
    (2017, 1 ,100),
    (2017, 1 ,50),
    (2017, 2 ,200),
    (2017, 2 ,300),
    (2018, 1 ,300),
    (2018, 1 ,100),
    (2018, 2 ,400) AS sales
GROUP BY col1, col2
WITH CUBE
ORDER BY year, quarter

The result of the example query is shown in the following table.

Query result for the example rollup
YEAR QUARTER AMOUNT
null null 1450
null 1 550
null 2 900
2017 null 650
2017 1 150
2017 2 500
2018 null 800
2018 1 400
2018 2 400

With a GROUPING SETS grouping, an extension to the group by clause, you can explicitly specify the grouping sets of interest. In other words, the ROLLUP and the CUBE groupings are shortcuts for common grouping-set use cases.

Referring to the preceding example, by adding a GROUPING SETS modifier to the group by clause you can compute rollup sales data on a quarter-by-year basis and a yearly basis only as shown in the following example.

-- rollup sales data on a quarter by year basis and a yearly basis only
SELECT
    sales.col1 AS year,
    sales.col2 AS quarter,
    SUM(sales.col3) AS amount
FROM VALUES
    (2017, 1 ,100),
    (2017, 1 ,50),
    (2017, 2 ,200),
    (2017, 2 ,300),
    (2018, 1 ,300),
    (2018, 1 ,100),
    (2018, 2 ,400) AS sales
GROUP BY col1, col2
GROUPING SETS ( (COL1), (COL1, COL2) )
ORDER BY year, quarter

The result of the example query is shown in the following table.

Query result for the example rollup
YEAR QUARTER AMOUNT
2017 null 650
2017 1 150
2017 2 500
2018 null 800
2018 1 400
2018 2 400

More topics - simpleselect

For more information about the clauses that are used in a simpleselect, see the following topics:

Sort item clause

Sort items are a component of a fullselect or a window specification.

sortItem

The semantics of the sort item components are as follows.

  • expression: The expression represents a sort key. The value of the sort key is used to order the rows of the result.
  • ASC: Uses the values of the sort key in ascending order. ASC is the default.
  • DESC: Uses the values of the sort key in descending order.
  • NULLS:
    • FIRST: Specifies that NULL values appear first in the order.
    • LAST: Specifies that NULL values appear last in the order.

More topics - expression clause

For more information about the clauses that are used in an expression clause, see the following topic:

Relations

A relation is a component of a simpleselect or a relation primary (which itself is a component of a relation). It is usually referred to as a table.

A relation is an entity that represents input data that is either stored physically on Cloud Object Storage or taken from virtual intermediate results. Such an intermediate result can be, for example, the result of a subselect, a values clause, a common table expression, or a table-valued function. Relations can be queried or joined with other relations.

Specifying the physical data stored on Cloud Object Storage as a relation is done by using the externalTableSpec syntax. An example of a valid table URI is cos://us-geo/sql/orders.parquet, which references one of the sample tables that are provided by Data Engine out of the box.

Multiple relations can be composed by using join operators. The syntax for joining relations is defined by the following syntax diagrams.

relation

joinClause

naturalJoinClause

Relations can be joined by using several types of joins that are described in detail in section joinType.

Apart from the join type, the following two different types of joins exist:

  • Joins that specify a join condition by using a booleanExpression or a USING clause.
  • NATURAL joins that make an implicit guess on which columns to use for joining relations. Use natural joins carefully.

relationPrimary

externalTableSpec

An external table specification represents an URI for an object that is stored on Cloud Object Storage combined with a specification of the object type. Valid values for object type identifier are AVRO, CSV, JSON, ORC, PARQUET, and TEXT.

If the file format is CSV, with the optional FIELDS TERMINATED BY clause you can specify a field separator other than the default , (comma). The following example shows a query for parsing a CSV with | (vertical bar) as the delimiter:

SELECT *
   FROM cos://us-geo/sql/BlackFriday.csv
       STORED AS CSV FIELDS TERMINATED BY '|'
   LIMIT 3`

All single Unicode characters are allowed as delimiters.

By default, it is assumed that CSV input objects have a header line that specifies the names of the input columns. If the objects don't have a header line, you must specify the option NOHEADER in the STORED AS CSV clause. In this case, the names _C0, _C1, ... are used for the input columns. For more information, see COS URI.

By default, if the format of the input data is JSON, each line must contain a separate, self-contained, and valid JSON object, also called newline-delimited JSON. However, if you specify the option MULTILINE, Data Engine can process JSON input data even if individual data records span multiple lines, such as when the data was formatted to make it easier to read. Specify this option only if you really need it because it limits input parallelization and can significantly reduce performance when you process large volumes of JSON data. If you need to frequently query large amounts of multiline JSON data, use Data Engine to transform the data into single -line JSON, or into a more performance optimized format, such as Parquet, before querying the transformed data.

If the file format is Parquet, with the optional MERGE SCHEMA clause you can handle Parquet schema evolution by specifying to scan all qualifying Parquet objects for their schema, and to merge the final schema across all objects. By default, for Parquet input only the first Parquet object that is found is used to infer the schema, which guarantees minimal overhead for compiling the SQL. Thus, use this option if your Parquet input data does not have a homogeneous schema.

If the object type identifier TEXT is used, the input is read line by line without inferring any schema. The whole line is returned in one column, called value. To query specific fields from the input data, you can use the options regexp_extract or get_json_object.

timeSeriesProperties

The TIME_SERIES_FORMAT option triggers a read transformation mechanism that uses a set of timeSeriesProperties to dynamically generate one or more native time series columns (defined by the IN clause) from the specified value and key columns of the input data.

The parameters timetick and value are the only parameters that are required to be specified.

Following you see the descriptions of each parameter and how they affect the time series:

  • timetick: The column that contains the timestamp or timetick. Ultimately, the resulting time series is sorted by this column. If two rows contain the same timetick, it is uncertain which timetick comes first in the time series.

  • value: The column that contains the value.

  • key: Optionally specify a key column that you can use to group each time series by. If a key is indicated, you can assume that n time series are created, where n is the set of all keys in the key column. If no key column is specified, a single time series is created from the indicated data set.

  • starttime: Optionally specify a starttime string (any properly formatted DateTime) for which to set the time series TRS. If starttime is not indicated, and granularity is indicated, the starttime defaults to 1 January 1970 12 AM (midnight) GMT. However, if no granularity is indicated, a TRS is not associated with the created time series.

  • granularity: Optionally specify a granularity string (a properly formatted ISO-8601 duration format) for which to set the time series reference system TRS. If granularity is not indicated, and starttime is indicated, the default granularity is 1 millisecond. However, if no starttime is indicated, a TRS is not associated with the created time series.

The following examples show you how to use TIME_SERIES_FORMAT parameters for dynamic time series generation during the read process.

Create a time series per location, set the time series TRS with default start time and 1 ms granularity, and store with it with the name "ts".

SELECT
	location,
	ts
FROM cos://us-geo/sql/temperature_humidity.csv
USING TIME_SERIES_FORMAT(key="location", timetick="timestamp", value="humidity", granularity="PT0.001S") in ts

Create a time series per location, set the time series TRS with start time "2011-12-03T10:15:30" and default granularity (1 ms), and store it with the name "ts".

SELECT
	location,
	ts
FROM cos://us-geo/sql/temperature_humidity.csv
USING TIME_SERIES_FORMAT(key="location", timetick="timestamp", value="humidity", starttime="2011-12-03T10:15:30") in ts

Create a time series per location with no TRS, store it with the name "ts". If no granularity or start time is provided, a TRS is not associated with the time series, and therefore with_trs runs into exception.

SELECT
	location,
	ts
FROM cos://us-geo/sql/temperature_humidity.csv
USING TIME_SERIES_FORMAT(key="location", timetick="timestamp", value="humidity") in ts

Create a single time series, store it with the default name "time_series". Without specifying a key, it is not possible to create multiple time series.

SELECT
	location,
	time_series
FROM cos://us-geo/sql/temperature_humidity.csv
USING TIME_SERIES_FORMAT(timetick="timestamp", value="humidity")

tableTransformer

A table transformer is a function that is applied to the input data set before it is sent to the actual SQL query compilation and execution.

You can wrap your external table definition optionally with the FLATTEN table transformation function. It preprocesses your input table before query compilation to a fully flat column schema. This table transformation function can be useful when you have hierarchical input data as it is often found in JSON documents. By using FLATTEN, you do not need to dereference all nested columns explicitly in your SQL statement.

For example, you can run a simple SELECT * FROM FLATTEN(cos://us-geo/sql/iotmessages STORED AS JSON) on a flattened JSON input and use CSV output to easily browse a sample of your JSON input data.

The FLATTEN table transformation function creates a flat list of columns by concatenating all nested column names with _. You can optionally also combine FLATTEN with CLEANCOLS.

You can wrap your external table definition optionally with the CLEANCOLS table transformation function. It preprocesses your input table before query compilation by renaming all columns that have characters that are NOT supported by certain target formats, such as Parquet. These characters are ,, ;, ,,,, =, (, ), {, and }. They are replaced by the corresponding URL-encoded representation, for example, %20 for space ( ). With this function, you can write results, for example, into Parquet, without the need to provide column by column alias names in your SQL when your input data has columns with these characters. A typical situation is the existence of space ( ) in input columns.

For example, you can use SELECT * FROM CLEANCOLS(cos://us-geo/sql/iotmessages STORED AS JSON) INTO cos://us-geo/mybucket/myprefix STORED AS PARQUET to produce a result set that can be stored as is into Parquet target format.

If you wrap your external table definition with the DESCRIBE table transformer, the table does not show its actual content but the schema that is inferred from the objects in IBM Cloud® Object Storage instead. With this function, you can explore the schema before you author your actual SQL statements against it.

When you use the DESCRIBE table transformer in your SQL statement, the default output format is JSON instead of CSV.

You can also wrap DESCRIBE around the other table transformers to explore the transformed table schema. However, you cannot wrap other table transformers around the DESCRIBE transformer.

tableValuedFunction

A table-valued function returns a relation, which is a set of rows. An example of a table-valued function is range(). For more information, see SQL functions.

More topics - relation clause

For more information about the clauses that are used in relation clauses, see the following topics:

Values clause

A values clause is a component of a fullselect or represents a primary relation. Its syntax is defined by the following syntax diagram.

valuesClause

With a values clause, you can define a result set by specifying actual values for each column of a row by using expressions.

Each expression in the list of expressions represents a row of the result set that is defined.

In a single-column result set, each expression represents the value of this column in a row.

In a multi-column result set, each expression represents a list of n expressions that are enclosed by parentheses, where n is the number of columns in the result set.

To join a values clause with other types of result sets, specify an identifier that acts as an alias for the values clause.

Examples - values clause

-- single column result set with 3 rows
SELECT * FROM VALUES 1, 2 , 3

The result of the example query is shown in the following table.

Query result for the example single
COL1
1
2
3
-- single column result set with 3 rows specifying parentheses for row expressions
SELECT * FROM VALUES (1), (2) , (3)

The result of the example query is shown in the following table.

Query result for the example single
COL1
1
2
3
--- joining two multi column result sets by using their identifier
SELECT
    emp.col1 AS id,
    emp.col2 AS name,
    mission.col2 AS missions
FROM
    VALUES
        (1, 'Spock'),
        (2,'Kirk') ,
        (3, 'McCoy'),
        (4,'Scotty') AS emp,
    VALUES
        (1,3000),
        (2,2000),
        (3,3000),
        (4,4000) AS mission
WHERE emp.col1 = mission.col1

The result of the example query is shown in the following table.

Query result for the example joining
ID NAME MISSIONS
1 Spock 3000
2 Kirk 2000
3 McCoy 3000
4 Scotty 4000

Values statement

A values statement is a statement on its own. It can be used instead of a fullselect if your statement references only a single value and does not contain any join with other relations or values clauses.

Examples - values statement

-- values statement with single column result set with 3 rows
VALUES 1, 2, 3

The result of the example query is shown in the following table.

Query result for the example values
COL1
1
2
3
--- values statement with multi column result set
VALUES
        (1, 'Spock'),
        (2,'Kirk') ,
        (3, 'McCoy'),
        (4,'Scotty')

The result of the example query is shown in the following table.

Query result for the example values
COL1 COL2
1 Spock
2 Kirk
3 McCoy
4 Scotty

More topics - values clause

For more information about the clauses that are used in a values clause, see the following topics:

Lateral views

A lateral view is a component of a simpleselect. With lateral views, you can build virtual tables at query execution time by using table-generating functions. Examples of table-generating functions are explode(), posexplode(), and posexplode_outer(). The explode()-style functions take an array or map as input and return a row for each element in the array. For more information, see SQL functions.

lateralView

The syntax of a lateral view clause is described by the following syntax diagram.

The semantics of the entities in order of appearance in the syntax diagrams is as follows:

  • OUTER: Specifying this keyword ensures that the lateral view contains at least one row with null values in case the table-generating function does not return any rows.
  • qualifiedName: Name of a table-generating function.
  • expression: An expression that resolves to an array.
  • identifier: Lateral view name, which is the name of the new virtual table.
  • identifier: Lateral view column names.

Examples - lateralView

Lateral views are useful when you deal with repeating groups within a table that is not normalized tables. The examples show how to deal with tables that represent a 1-n relation and an n-m relation.

-- deal with a 1-n relation

SELECT
    master_child.col1 AS master_id,
    child_table.child_id
FROM VALUES
    ( 1 , ARRAY(1,2,3)),
    (2, ARRAY(4,5,6)) AS master_child
LATERAL VIEW EXPLODE(master_child.col2) child_table AS child_id

The result of the example query is shown in the following table.

Query result for example: deal with a 1-n relation.
MASTER_ID CHILD_ID
1 1
1 2
1 3
2 4
2 5
2 6
-- deal with an n-m relation

SELECT
    master_table.master_id,
    child_table.child_id
FROM VALUES
    ( ARRAY(10,20) , ARRAY(1,2,3)),
    ( ARRAY(30,40), ARRAY(4,5,6)) AS master_child
LATERAL VIEW EXPLODE(master_child.col1) master_table AS master_id
LATERAL VIEW EXPLODE(master_child.col2) child_table AS child_id

The result of the example query is shown in the following table.

Query result for example: deal with an n-m relation.
MASTER_ID CHILD_ID
10 1
10 2
10 3
20 1
20 2
20 3
30 4
30 5
30 6
40 4
40 5
40 6

More topics - lateral view

For more information about the clauses that are used in a lateral view clause, see the following topics:

Note:

  • A simpleselect can contain multiple lateral view clauses that are evaluated in order of appearance.
  • Within a lateral view clause, you can refer to columns defined in any (virtual) table on the left of the current lateral view clause.

Join types

SELECT statements can retrieve and join column values from two or more tables into a single row. The retrieval is based on a specified condition, typically of matching column values.

The main characteristic of a join is, typically, matching column values in rows of each table that participates in the join. The result of a join associates rows from one table with rows from another table. Depending on the type of join operation, some rows might be formed that contain column values in one table that do not match column values in another table.

A joined table specifies an intermediate result table that is the result of either an INNER join, an OUTER join, a CROSS join, or an ANTI join. The table is derived by applying one of the join operators to its operands.

joinType

Inner join

An INNER join combines each row of the left table with each row of the right table, keeping only the rows in which the join condition is true.

-- inner join query
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2,
    right_table.col1 AS r_col1,
    right_table.col2 AS r_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
    INNER JOIN
    VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
    ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example.
L_COL1 L_COL2 R_COL1 R_COL2
0 10 0 10
2 12 2 12
4 14 4 14

Outer join

An OUTER join includes the rows that are produced by the inner join, plus the missing rows, depending on the type of outer join.

A LEFT OUTER or LEFT join includes the rows from the left table that were missing from the inner join.

-- left outer join query
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2,
    right_table.col1 AS r_col1,
    right_table.col2 AS r_col2
FROM VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
LEFT OUTER JOIN
VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example: left outer join query.
L_COL1 L_COL2 R_COL1 R_COL2
0 10 0 10
1 11 null null
2 12 2 12
3 13 null null
4 14 4 14
5 14 null null

A RIGHT OUTER or RIGHT join includes the rows from the right table that were missing from the inner join.

-- right outer join query
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2,
    right_table.col1 AS r_col1,
    right_table.col2 AS r_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
    RIGHT OUTER JOIN
    VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
    ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example: right outer join query.
L_COL1 L_COL2 R_COL1 R_COL2
0 10 0 10
2 12 2 12
4 14 4 14
null null 6 16

A FULL OUTER or FULL join includes the rows from both tables that were missing from the inner join.

-- full outer join query
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2,
    right_table.col1 AS r_col1,
    right_table.col2 AS r_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3, 13), (4, 14), (5, 14) AS left_table
    FULL OUTER JOIN
    VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
    ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example: full outer join query.
L_COL1 L_COL2 R_COL1 R_COL2
1 11 null null
null null 6 16
3 13 null null
5 14 null null
4 14 4 14
2 12 2 12
0 10 0 10

Cross join

A CROSS join creates a Cartesian product of the tables that are involved in the join operation, if a CROSS join that specifies a join condition behaves like an inner join.

-- cross join that specifies a join condition

SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2,
    right_table.col1 AS r_col1,
    right_table.col2 AS r_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
    CROSS JOIN VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
    ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example: cross join that specifies a join condition.
L_COL1 L_COL2 R_COL1 R_COL2
0 10 0 10
2 12 2 12
4 14 4 14
-- cross join that specifies no join condition
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2,
    right_table.col1 AS r_col1,
    right_table.col2 AS r_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
    CROSS JOIN VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table

The result of the example query is shown in the following table.

Query result for the example cross
L_COL1 L_COL2 R_COL1 R_COL2
0 10 0 10
0 10 2 12
0 10 4 14
0 10 6 16
1 11 0 10
1 11 2 12
1 11 4 14
1 11 6 16
2 12 0 10
2 12 2 12
2 12 4 14
2 12 6 16
3 13 0 10
3 13 2 12
3 13 4 14
3 13 6 16
4 14 0 10
4 14 2 12
4 14 4 14
4 14 6 16
5 14 0 10
5 14 2 12
5 14 4 14
5 14 6 16

Anti join

A LEFT ANTI or ANTI join returns only rows from the left table that do not have a matching row in the right table. Columns from the right table cannot be included in the column list of the select statement.

-- left anti join query
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
    ANTI JOIN
    VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
    ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example: left anti join query.
L_COL1 L_COL2
1 11
3 13
5 14

Left semi join

A LEFT SEMI join acts like an inner join but does not include the columns of the right table.

-- left semi join query
SELECT
    left_table.col1 AS l_col1,
    left_table.col2 AS l_col2
FROM
    VALUES (0, 10), (1, 11), (2, 12), (3,13), (4, 14), (5, 14) AS left_table
    LEFT SEMI JOIN
    VALUES (0, 10), (2, 12), (4, 14), (6, 16) AS right_table
    ON left_table.col1 = right_table.col1

The result of the example query is shown in the following table.

Query result for example: left semi join query.
L_COL1 L_COL2
0 10
2 12
4 14

Sampling table data

Any table that is object stored on Cloud Object Storage, used in a from clause, can be associated with a table sample clause. The table sample clause defines how to retrieve a subset of rows from the underlying table (object stored on Cloud Object Storage). Thus, you can write queries for samples of the data, for example, for interactive data exploration and data mining.

The general syntax of a table sample clause is described by the following syntax diagram.

sample

bucketSampleClause

Three sampling types are supported:

  • With TABLESAMPLE <number> PERCENT, you can sample a certain percentage of rows from the underlying table.
  • With TABLESAMPLE <expression> ROWS, you can sample some rows from the underlying table.
  • With TABLESAMPLE BUCKET x OUT OF y, you can bucketize the underlying data into y buckets and returns rows from bucket x. Buckets are numbered from 1 to y.

Examples - sample

The following examples demonstrate how to sample a subset of data from a Parquet object on Cloud Object Storage. The object that is referenced is accessible from the web UI as part of the provided sample queries.

-- retrieve 10 percent of employee data
SELECT * FROM cos://us-geo/sql/employees.parquet STORED AS PARQUET TABLESAMPLE (10 PERCENT)
-- retrieve 10 rows from the employee data object
SELECT * FROM cos://us-geo/sql/employees.parquet STORED AS PARQUET TABLESAMPLE (10 ROWS)
-- bucketize the employee data in 10 buckets and retrieve data from 2 buckets
SELECT * FROM cos://us-geo/sql/employees.parquet STORED AS PARQUET TABLESAMPLE (BUCKET 2 OUT OF 10)

More topics - table sample clause

For more information about the clauses that are used in a table sample clause, see the following topics:

SQL functions

The syntax for SQL function invocation is described by the following syntax diagram.

functionOrAggregate

Most function invocations look like function(argument1, ..., argumentN) but functions like TRIM(), POSITION(), FIRST(), LAST(), STRUCT(), EXTRACT(), and SUBSTRING() support a different invocation style.

Refer to section SQL functions for details about supported functions.

More topics - function or aggregate clause

For more information about the clauses that are used in a function or aggregate clause, see the following topics:

Window functions

Classic SQL aggregation functions like SUM(), MAX(), or MIN() process a group of rows to derive a single value. Window functions take this one step further by allowing to process a group of rows and derive a single value for each row in the group. Note the difference to scalar functions that return a single value for each row. Scalar functions derive a single value from a single row and not a group of rows.

With window functions, it is possible to calculate things like moving averages or cumulative sums.

Working with window functions involves two steps:

  1. Choose a window function to answer the question of interest.
  2. Define a window the chosen window function is applied to.

The three types of window functions are:

  • Ranking functions, for example, rank(), ntile(), or rowNumber()
  • Analytic functions, for example, cume_dist(), first_value(), or last_value()
  • Aggregation functions, for example, sum(), max(), or min()

For more information, see SQL functions.

A window can be defined in two ways:

  • With the WINDOW keyword, you can define an identifier for a window specification in a fullselect or simpleselect. This named window specification can then be referenced by the OVER keyword.
  • Unnamed window specifications can be defined inline following the keyword OVER in a fullselect or simpleselect.

The syntax of a window specification is defined by the following syntax diagrams.

namedWindows

namedWindow

windowSpec

windowClusterBy

windowPartitionBy

windowOrderBy

The window specification consists of the following clauses:

  • The PARTITION BY clause defines which rows belong to the same window partition. DISTRIBUTE BY can be used as a synonym for PARTITION BY.
  • The ORDER BY clause defines the ordering of rows within a window partition. SORT BY can be used as a synonym for ORDER BY.
  • The two ways to define a window size are:
    • RANGE: The window size is defined as a value range.
    • ROW: The window size is defined as the number of rows before and/or after the current row.
  • The following keywords can be used to define range boundaries:
    • CURRENT ROW: Specifies to use the current row as a bound.
    • UNBOUNDED: Specifies to use negative or positive infinity as a lower or upper bound.
    • Depending on the context, PRECEDING specifies:
      • RANGE: The lower bound of the value range.
      • ROWS: The number of rows before the current row.
    • Depending on the context, FOLLOWING specifies:
      • RANGE: The upper bound of the value range.
      • ROWS: The number of rows after the current row.

windowFrame

frameBound

Examples - window function

The following window function examples use values clauses to define result sets for group by operations. For more information about the values clause, see valuesClause.

Ranking function example

This example uses a table that contains information about employee social media activity (posts):

  • Column 1: employee ID
  • Column 2: department ID
  • Column 3: number of social media posts

The example shows how to retrieve the ID of the two most active employees in each department.

-- derive posts ranking by using a named window specification
SELECT *
FROM (
    SELECT
        posts.col1 AS emp_id,
        posts.col2 AS dept_id,
        posts.col3 AS posts,
        DENSE_RANK() OVER post_ranking AS rank
    FROM VALUES
        (1, 1 ,100),
        (2, 1 ,50),
        (8, 1 ,250),
        (3, 2 ,200),
        (4, 2 ,300),
        (9, 2 ,1000),
        (5, 3 ,300),
        (6, 3 ,100),
        (7, 3 ,400) AS posts
    WINDOW post_ranking AS (
        PARTITION BY posts.col2
        ORDER BY posts.col3 DESC
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW)
)
WHERE rank <= 2

The result of the example query is shown in the following table.

Query result for example: derive posts ranking by using a named window specification.
EMP_ID DEPT_ID POSTS RANK
8 1 250 1
1 1 100 2
7 3 400 1
5 3 300 2
9 2 1000 1
4 2 300 2
-- derive posts ranking by using an inline window specification
SELECT * FROM (
    SELECT
        posts.col1 AS emp_id,
        posts.col2 AS dept_id,
        posts.col3 AS posts,
        DENSE_RANK() OVER (
                        PARTITION BY posts.col2
                        ORDER BY posts.col3 DESC
                        ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
                        ) AS rank
        FROM VALUES
            (1, 1 ,100),
            (2, 1 ,50),
            (8, 1 ,250),
            (3, 2 ,200),
            (4, 2 ,300),
            (9, 2 ,1000),
            (5, 3 ,300),
            (6, 3 ,100),
            (7, 3 ,400) AS posts
) WHERE rank <= 2

The result of the example query is shown in the following table.

Query result for example: derive posts ranking by using an inline window specification.
EMP_ID DEPT_ID POSTS RANK
8 1 250 1
1 1 100 2
7 3 400 1
5 3 300 2
9 2 1000 1
4 2 300 2

Analytic function example

This example uses a table that contains transaction information. The layout is as follows:

  • Column 1: transaction ID
  • Column 2: account number
  • Column 3: transaction amount

The example shows how to create a cumulative distribution of transaction amounts by using the analytic function CUME_DIST(). The CUME_DIST() function returns the percentage of rows that have a value less than or equal to the current row’s value.

-- cumulative distribution of transaction amounts
SELECT
    txn_amount,
    MAX(balance_dist)
FROM (
        SELECT
            txn.col2 AS account,
            txn.col3 AS txn_amount,
            CUME_DIST() OVER current_balance AS balance_dist
        FROM VALUES
            (1, 42, 1000),
            (2, 4711, 2000),
            (3, 42, -200),
            (4, 42, -200),
            (5, 4711, 1000),
            (6, 4711, -300),
            (7, 4711, -300),
            (8, 42, 1000),
            (9, 4711, -400) AS txn
        WINDOW current_balance AS (ORDER BY txn.col3)
        ORDER BY txn.col3
)
GROUP BY txn_amount

The result of the example query is shown in the following table.

Query result for example: cumulative distribution of transaction amounts.
TXN_AMOUNT MAX(BALANCE_DIST)
-400 0.1111111111111111
-300 0.3333333333333333
-200 0.5555555555555556
1000 0.8888888888888888
2000 1.0

Aggregation Function Example

This example uses a table that contains transaction information. The layout is as follows:

  • Column 1: transaction ID
  • Column 2: account number
  • Column 3: transaction amount

The example shows how to retrieve the total balance of each account at the time of each transaction.

--- total balance of each account at the time of each transaction
SELECT
    txn.col1 AS txn_id,
    txn.col2 AS account,
    txn.col3 AS txn_amount,
    SUM(txn.col3) OVER current_balance AS balance
FROM VALUES
    (1, 42, 100),
    (2, 4711, 300),
    (3, 42, 50),
    (4, 42, -250),
    (5, 4711, 1000),
    (6, 4711, -300),
    (7, 4711, 100),
    (8, 42, 200),
    (9, 4711, -400) AS txn
WINDOW current_balance AS (
        PARTITION BY txn.col2
        ORDER BY txn.col1
        ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
) ORDER BY account, txn_id

The result of the example query is shown in the following table.

Query result for example: total balance of each account at the time of each transaction.
TXN_ID ACCOUNT TXN_AMOUNT BALANCE
1 42 100 100
3 42 50 150
4 42 -250 -100
8 42 200 100
2 4711 300 300
5 4711 1000 1300
6 4711 -300 1000
7 4711 100 1100
9 4711 -400 700

More topics - window function

For more information about the clauses that are used by a window function, see the following topics:

SQL expressions

Expressions

In the context of an SQL query statement, an expression is always a Boolean expression.

expression

More topics - Boolean expression

For more information, see booleanExpression.

Boolean expressions

The syntax of a Boolean expression is defined by the following syntax diagrams.

booleanExpression

A Boolean expression is one of the following:

  • A negation of a Boolean expression by using Boolean operator NOT.
  • A conjunction of Boolean expressions by using Boolean operator AND.
  • A disjunction of Boolean expressions by using Boolean operator OR.
  • An EXIST predicate.
  • A value expression optionally followed by a predicate.

Refer to booleanOperator for details about Boolean operators NOT, AND, and OR.

The EXISTS predicate tests for the existence of certain rows. The query can specify any number of columns, and the following applies:

  • The result is true only if the number of rows that is specified by the fullselect is not zero.
  • The result is false only if the number of rows that is specified is zero.
  • The result cannot be unknown.

Value expressions

valueExpression

A value expression is one of the following:

  • A primary expression.
  • The result of applying an unary operator to a value expression.
  • The result of performing an arithmetic operation with two value expressions.
  • The result of performing a comparison operation between two value expressions.

More topics - value expression

For more information about the clauses that are used by a value expression, see the following topics:

Primary expressions

primaryExpression

constant

interval

With an interval clause, you can define time duration constants that can be used in expressions to add or subtract time ranges from a timestamp value.

timeUnitSpec

The following time units are valid:

  • Singular form: SECOND, MINUTE, DAY, MONTH, YEAR
  • Plural form: SECONDS, MINUTES, DAYS, MONTHS, YEARS

Both singular and plural forms can be used interchangeably.

The following example demonstrates how to add and subtract several time units from the current timestamp.

-- add and subtract several time units from the current timestamp
SELECT
    CURRENT_TIMESTAMP
    - INTERVAL 2 YEARS
    + INTERVAL 1 MONTH
    - INTERVAL 3 DAYS
    + INTERVAL 10 HOURS
    + interval 30 MINUTES
    - INTERVAL 20 SECOND AS past_timestamp
FROM VALUES ("dummy")

The result of the example query is shown in the following table.

Query result for example: add and subtract several time units from the current timestamp.
PAST_TIMESTAMP
2016-11-13 20:23:43.621

Since interval clauses can get long, especially if days, hours, minutes, and seconds are involved, it is possible to use an abbreviated syntax by specifying a format STRING and by using the TO keyword.

The format STRING can be used to specify the following time intervals:

  • YEAR TO MONTH interval by using a format string that complies with signYEAR-MONTH with the following:
    • sign: optional + or - sign
    • YEAR: number of years
    • MONTH: number of months
  • DAY TO SECOND interval by using a format string that complies with signDAY HOUR:MINUTE:SECOND with the following:
    • sign: optional + or - sign
    • DAY: number of days
    • HOUR: number of hours
    • MINUTE: number of minutes
    • SECOND: number of seconds

When you specify a time interval by using the keyword TO, only the singular form of a time unit identifier is supported.

The following table shows equivalent interval clauses.

Equivalent INTERVAL clauses.
INTERVAL expression INTERVAL TO expression
INTERVAL 1 DAY - INTERVAL 2 HOURS - INTERVAL 3 MINUTES - INTERVAL 4 SECONDS INTERVAL '1 2:3:4.100' DAY TO SECOND
INTERVAL 1 YEAR - INTERVAL 2 MONTH INTERVAL '1-2' YEAR TO MONTH

The following example demonstrates equivalent interval expressions when you deal with YEAR and MONTH time units.

-- equivalent interval expressions with YEAR and MONTH time units
WITH ts AS (
    SELECT CURRENT_TIMESTAMP AS now FROM VALUES ('dummy')
    )
SELECT
    now - INTERVAL 1 YEAR - INTERVAL 2 MONTH AS LONG_VERSION,
    now - INTERVAL '1-2' YEAR TO MONTH  AS SHORT_VERSION
FROM ts

The result of the example query is shown in the following table.

Query result for example: equivalent interval expressions with YEAR and MONTH time units.
LONG_VERSION SHORT_VERSION
2017-08-16 10:38:07.304 2017-08-16 10:38:07.304

The following example demonstrates equivalent interval expressions, when you deal with DAY, HOUR, MINUTE, and SECOND time units.

-- equivalent interval expressions when you deal with DAY, HOUR, MINUTE, and SECOND time units
WITH ts AS (
    SELECT CURRENT_TIMESTAMP AS now FROM VALUES ('dummy')
    )
SELECT
    now - INTERVAL 1 DAY - INTERVAL 2 HOURS - INTERVAL 3 MINUTES - INTERVAL 4 SECONDS AS LONG_VERSION,
    now - INTERVAL '1 2:3:4.100' DAY TO SECOND AS SHORT_VERSION
FROM ts

The result of the example query is shown in the following table.

Query result for example: equivalent interval expressions when you deal with DAY, HOUR, MINUTE and SECOND time units.
LONG_VERSION SHORT_VERSION
2018-10-15 08:33:00.84 2018-10-15 08:33:00.84

columnReference

qualifiedName

A qualified name is a sequence of identifiers that are separated by .. For example, a column name can be qualified by the name of the relation the column is defined in.

Qualified names are available in the following context:

Predicates

predicate

The BETWEEN ... AND predicate compares a value with a range of values. If NOT is specified, the result is reversed.

The IN predicate compares a value or values with a collection of values. The range of values is either defined by a query or a list of expressions that are enclosed in parentheses. The query must identify a number of columns that are the same as the number of expressions that are specified to the left of the IN keyword. In addition, the number of elements in the list of expressions must be the same as the number of expressions that are specified to the left of the IN keyword. If NOT is specified, the result is reversed.

The LIKE predicate searches for strings that have a certain pattern. The pattern is specified by a string in which certain characters have a special meaning.

  • The underscore character _ represents any single character.
  • The percent sign % represents a string of zero or more characters.
  • Any other character represents itself. Thus trailing blanks in a pattern are part of the pattern. If NOT is specified, the result is reversed.

The RLIKE predicate searches for a regular expression pattern in a string. If the pattern expression is found, the result is true. If the pattern expression is not found, the result is false. If the value of any of the arguments is null, the result of the RLIKE predicate is unknown. If NOT is specified, the result is reversed.

The regular expression pattern must be a Java™ regular expression as defined by Java class java.util.regex.Pattern. Meta characters that start with a \ must be escaped for the regular expression to work, for example, use \\d instead of \d in a pattern string to represent a digit. For more information, such as supported meta characters and predefined character classes, see the latest Java documentation.

The IS NULL predicate tests for null values. The result of a NULL predicate cannot be unknown. If the value of the expression is null, the result is true. If the value is not null, the result is false. If NOT is specified, the result is reversed.

The IS DISTINCT FROM predicate compares two expressions and evaluates to TRUE if their values are not identical. The result of a DISTINCT predicate cannot be null. If NOT is specified, the result is reversed. The result of a DISTINCT predicate depends on whether either or both of its input expressions are null:

NULL Values and DISTINCT predicate value.
Input expressions IS DISTINCT FROM IS NOT DISTINCT FROM
Both inputs are nonnull. Evaluates to TRUE if the inputs are not identical and FALSE if they are. Equivalent to the <> operator. Evaluates to FALSE if the inputs are not identical and TRUE if they are. Equivalent to the = operator.
One input is null. Evaluates to TRUE. Evaluates to FALSE.
Both inputs are null. Evaluates to FALSE. Evaluates to TRUE.

The following DISTINCT predicates are logically equivalent to the corresponding search conditions:

DISTINCT predicate and search condition.
DISTINCT predicate Search condition
value 1 IS NOT DISTINCT FROM value2 (value1 IS NOT NULL AND value2 IS NOT NULL AND value1 = value 2) OR (value1 IS NULL AND value2 IS NULL)
value 1 IS DISTINCT FROM value2 NOT (value1 IS NOT DISTINCT FROM value2)

Examples - predicate

IS DISTINCT FROM examples
-- select all rows with distinct values in column A and B
SELECT * FROM (
    SELECT
        col1 AS a,
        col2  AS b
    FROM VALUES
            (1 , 2),
            (2, 2),
            (null, 2),
            (1, null),
            (2, null),
            (null, null)
) WHERE a IS DISTINCT FROM b

The result of the example query is shown in the following table.

Query result for example: select all rows with distinct values in column A and B.
A B
1 2
null 2
1 null
2 null
-- all rows that have no distinct values in column A and B
SELECT * FROM (
    SELECT
        col1 AS a,
        col2 AS b
    FROM VALUES
            (1, 2),
            (2, 2),
            (null, 2),
            (1, null),
            (2, null),
            (null, null)
) WHERE a IS NOT DISTINCT FROM b

The result of the example query is shown in the following table.

Query result for example: all rows that have no distinct values in column A and B.
A B
2 2
null null
BETWEEN ... AND ... examples
-- all employees with a salary between 4000 and 8000
SELECT
    emp.col1 AS emp_id,
    emp.col2 AS salary
FROM VALUES
    (0, 1000),
    (2, 2000),
    (3, 3000),
    (4, 4000),
    (5, 5000),
    (6, 6000),
    (7, null),
    (8, 8000),
    (9,9000) AS emp
WHERE emp.col2 BETWEEN 4000 AND 8000

The result of the example query is shown in the following table.

Query result for example: all employees with a salary in the range 4000 - 8000.
EMP_ID SALARY
4 4000
5 5000
6 6000
8 8000
-- all employees with a salary not between 4000 and 8000
SELECT
    emp.col1 AS emp_id,
    emp.col2 AS salary
FROM VALUES
    (0, 1000),
    (2, 2000),
    (3, 3000),
    (4, 4000),
    (5, 5000),
    (6, 6000),
    (7, null),
    (8, 8000),
    (9,9000) AS emp
WHERE emp.col2 NOT BETWEEN 4000 AND 8000

The result of the example query is shown in the following table.

Query result for the example all
EMP_ID SALARY
0 1000
2 2000
3 3000
9 9000
IN examples
-- all employees working in department D01 or D02
SELECT
    emp.col1 AS emp_id,
    emp.col2 AS emp_dept
FROM VALUES
    (0, 'D01'),
    (2, 'C01'),
    (3, 'C02'),
    (4, 'D01'),
    (5, 'D02'),
    (6, 'C01'),
    (7, 'D01'),
    (8, 'C03'),
    (9,'D01') AS emp
WHERE emp.col2 IN ('D01','D02')

The result of the example query is shown in the following table.

Query result for the example all
EMP_ID EMP_DEPT
0 D01
4 D01
5 D02
7 D01
9 D01
-- all employees that are managing a department
SELECT
    emp.col1 AS emp_id,
    emp.col2 AS emp_dept
FROM VALUES
    (0, 'D01'),
    (2, 'C01'),
    (3, 'C02'),
    (4, 'D01'),
    (5, 'D02'),
    (6, 'C01'),
    (7, 'D01'),
    (8, 'C03'),
    (9,'D01') AS emp
WHERE (emp.col1,emp.col2) IN (
    SELECT
        mgr.col1,
        mgr.col2
    FROM VALUES
        (2, 'C01'),
        (4, 'D01'),
        (5, 'D02') AS mgr
    )

The result of the example query is shown in the following table.

Query result for the example all
EMP_ID EMP_DEPT
2 C01
4 D01
5 D02
LIKE examples
-- all employees that work in a department that starts with letter C
SELECT
    emp.col1 AS emp_id,
    emp.col2 AS emp_dept
FROM VALUES
        (0, 'D01'),
        (2, 'C01'),
        (3, 'C02'),
        (4, 'D01'),
        (5, 'D02'),
        (6, 'C01'),
        (7, 'D01'),
        (8, 'C03'),
        (9, 'D01') AS emp
WHERE emp.col2 LIKE 'C%'

The result of the example query is shown in the following table.

Query result for the example all
EMP_ID EMP_DEPT
2 C01
3 C02
6 C01
8 C03
-- all department names that do not start with letter C
SELECT
    DISTINCT emp.col2 AS emp_dept
FROM VALUES
    (0, 'D01'),
    (2, 'C01'),
    (3, 'C02'),
    (4, 'D01'),
    (5, 'D02'),
    (6, 'C01'),
    (7, 'D01'),
    (8, 'C03'),
    (9,'D01') AS emp
WHERE emp.col2 NOT LIKE 'C%'

The result of the example query is shown in the following table.

Query result for example: all department names that do not start with letter C.
EMP_DEPT
D01
D02
RLIKE examples
-- all rows that contain in col2 a value ending with 'bc'
SELECT *
FROM VALUES
    (0, 'Abc'),
    (1, 'xyz abc'),
    (2, 'abcabcabc'),
    (3, 'abc xyzxyz abc'),
    (4, '123 456 789') AS data
WHERE data.col2 RLIKE 'bc$'

The result of the example query is shown in the following table.

Query result for the example all
COL1 COL2
0 Abc
1 xyz abc
2 abcabcabc
3 abc xyzxyz abc
-- all rows that contain in col2 a sequence of 3 'abc' string occurrences
SELECT *
FROM VALUES
    (0, 'Abc'),
    (1, 'xyz abc'),
    (2, 'abcabcabc'),
    (3, 'abc xyzxyz abc'),
    (4, '123 456 789') AS data
WHERE data.col2 RLIKE '(abc){3}'

The result of the example query is shown in the following table.

Query result for the example all
COL1 COL2
2 abcabcabc
-- all rows that contain in col2 a sequence of integer values (three digits) separated by blank or tab
SELECT *
FROM VALUES
        (0, 'Abc'),
        (1, 'xyz abc'),
        (2, 'abcabcabc'),
        (3, 'abc xyzxyz abc'),
        (4, '123\t456 789') AS data
WHERE data.col2 RLIKE '\\d{3}[ \\t]\\d{3}[ \\t]\\d{3}'

The result of the example query is shown in the following table.

Query result for example: all rows that contain in col2 a sequence of integer values (3 digits) separated by blank or tab.
COL1 COL2
4 123 456 789
IS NULL examples
--- all employees with missing salary information
SELECT
    emp.col1 AS emp_id,
    emp.col2 AS salary
FROM VALUES
    (0, 1000),
    (2, 2000),
    (3, 3000),
    (4, 4000),
    (5, 5000),
    (6, 6000),
    (7, NULL),
    (8, 8000),
    (9,9000) AS emp
WHERE emp.col2 IS NULL

The result of the example query is shown in the following table.

Query result for example: all employees with missing salary information.
EMP_ID SALARY
7 null

CAST expression

The syntax of a cast expression is described by the following syntax diagrams.

The cast specification returns the cast operand (the first operand) cast to the type specified by the data type.

If the specified data type is not supported, you receive an error.

castExpression

In case an expression cannot be cast to the data type specified in the cast expression, the expression result is null.

More topics - cast expression

For more information about the clauses that are used by a cast expression, see the following topics:

Case expressions

A case expression allows an expression to be selected based on the evaluation of one or more conditions.

The syntax of a case expression is described by the following syntax diagrams.

caseExpression

The upper path in the syntax diagram represents a searched when clause, which means that the WHEN keyword follows directly after the CASE keyword. The lower path is a simple when clause, which means that an expression follows the CASE keyword.

In general, the value of the case expression is the value of the result expression, following the first (leftmost) case that evaluates to true. If no case evaluates to true and the ELSE keyword is present, the result is the value of the ELSE case result expression. If no case evaluates to true and the ELSE keyword is not present, the result is NULL. When a case evaluates to unknown (because of NULLs), the case is not true, and hence is treated the same way as a case that evaluates to false.

When you use the simple when clause, the value of the expression before the first WHEN keyword is tested for equality with the value of the expression that follows the WHEN keyword. Therefore, the data type of the expression before the first WHEN keyword must be comparable to the data types of each expression that follows the WHEN keywords.

A result expression is an expression that follows the THEN or ELSE keywords.

whenClause

Examples - whenClause

-- simple case expression with no ELSE clause
SELECT
    dep.col1 AS dep_id,
    CASE dep.col2
        WHEN 'A' THEN 'Administration'
        WHEN 'B' THEN 'Human Resource'
        WHEN 'C' THEN 'Development'
    END AS dep_name
FROM VALUES (0, 'A'), (1, 'B'), (2, 'C'), (3, 'D'), (4, 'E') AS dep

The result of the example query is shown in the following table.

Query result for example: simple case expression with no ELSE clause.
DEP_ID DEP_NAME
0 Administration
1 Human Resource
2 Development
3 null
4 null
-- simple case expression with ELSE clause
SELECT
    dep.col1 AS dep_id,
    CASE dep.col2
        WHEN 'A' THEN 'Administration'
        WHEN 'B' THEN 'Human Resource'
        WHEN 'C' THEN 'Development'
        ELSE 'UNKOWN'
    END AS dep_name
FROM VALUES (0, 'A'), (1, 'B'), (2, 'C'), (3, 'D'), (4, 'E') AS dep

The result of the example query is shown in the following table.

Query result for example: simple case expression with ELSE clause.
DEP_ID DEP_NAME
0 Administration
1 Human Resource
2 Development
3 UNKOWN
4 UNKOWN

The two scalar functions that are specialized to handle a subset of the functionality that is provided by CASE are NULLIF() and COALESCE().

CASE, NULLIF(), and COALESCE().
Expression Equivalent expression
CASE WHEN e1=e2 THEN NULL ELSE e1 END NULLIF(e1,e2)
CASE WHEN e1 IS NOT NULL THEN e1 ELSE e2 END COALESCE(e1,e2)
CASE WHEN e1 IS NOT NULL THEN e1 ELSE COALESCE(e2,...,eN) END COALESCE(e1,e2,...,eN)

For more information, see SQL functions.

For more information about the clauses that are used by a case expression, see the following topic:

Time series expressions

The syntax of a time series expression is described by the following syntax diagrams.

timeSeriesExpression

The syntax shows time series functions that require expressions, such as TS_MAP(), TS_FILTER(), TS_SEGMENT_BY_ANCHOR(), TS_SEGMENT_BY_MARKER(), TS_SEGMENT_BY_DUAL_MARKER(), TS_FIND(), and TS_COUNT_ANCHOR().

For more information about each function, see Data processing functions.

Example - time series

WITH timeseries_input AS (SELECT location, TIME_SERIES_WITH_TRS(TS_TIMESTAMP(timestamp), humidity, TS_TRS_DEFAULT()) AS ts
                          FROM cos://us-geo/sql/temperature_humidity.csv STORED AS CSV
                          GROUP BY location),
    only_40_or_above_ts AS (
	    SELECT location,
	   		TS_FILTER(ts, TS_EXP_GT(TS_EXP_ID(), 40.0)) AS above_40_ts
	    FROM timeseries_input
	)
SELECT location, TS_EXPLODE(above_40_ts) AS (timestamp, humidity) FROM only_40_or_above_ts

A time series expression is referenced by the following clause:

booleanTimeSeriesExpression

The Boolean time series expression syntax shows the available Boolean expresssions, such as TS_EXP_GT(), which is also used in the previous example.

For more information about each function, see Artifact creation functions.

valueTimeSeriesExpression

Time series values for expressions can either be a string or a double datatype.

doubleTimeSeriesExpression

The functions shown in the double time series expressions, such as TS_EXP_ABS() and TS_EXP_LENGTH(), are able to consume again double time series expressions, number, or an identity time series expression.

For more information about each function, see Artifact creation functions.

stringTimeSeriesExpression

The string function TS_EXP_ID_TO_STRING() converts an ID to a string and the TS_EXP_CONCAT() function concatenates the result of two string expressions.

For more information about each function, see Artifact creation functions.

stringConditionalExpression

The three conditional expression functions for string values are TS_EXP_IF_THEN_ELSE(), TS_EXP_IF_THEN(), and TS_EXP_MATCH_CASE().

For more information about each function, see Artifact creation functions.

identityTimeSeriesExpression

The identity expression denotes current observation values in time series.

Operator

The following types of operators can be used:

Unary operators

Unary operators.
Operator Operand types Description
-A All number types Unary negative operator. The type of the result is the same as the type of A.
+A All number types Unary positive operator. The type of the result is the same as the type of A.
~A All number types Bitwise NOT operator. The type of the result is the same as the type of A.

Arithmetic operators

Arithmetic operators.
Operator Operand types Description
A + B All number types Returns the result of adding A and B. The type of the result is the same as the type of the operand that is highest in the type hierarchy. For example, if A is of type FLOAT and B is of type INT, the result is of type FLOAT.
A - B All number types Returns the result of subtracting B from A. The type of the result is the same as the type of the operand that is highest in the type hierarchy. For example, if A is of type FLOAT and B is of type INT, the result is of type FLOAT.
A * B All number types Returns the result of multiplying A and B. The type of the result is the same as the type of the operand that is highest in the type hierarchy. For example, if A is of type FLOAT and B is of type INT, the result is of type FLOAT. If the operation causes an overflow, cast at least one of the operators to a type that is higher in the type hierarchy.
A / B All number types Returns the result of dividing A by B. The type of the result is DOUBLE.
A % B All number types Returns the remainder after dividing A by B. For example, 13.7 % 3 returns 1.7. The type of the result is the same as the type of the operand that is highest in the type hierarchy. For example, if A is of type FLOAT and B is of type INT, the result is of type FLOAT.
A DIV B Integer types Returns the integer part of the result of dividing A by B. For example, 13.7 DIV 3 returns the integer 4.
A & B All number types Returns the result of bitwise AND of A and B. The type of the result is the same as the type of the operand that is highest in the type hierarchy.
`A B` All number types
A ^ B All number types Returns the result of bitwise XOR of A and B. The type of the result is the same as the type of the operand that is highest in the type hierarchy.

String operator

String operator.
Operator Operand types Description
`A B`

Comparison operators

Comparison operators.
Operator Operand types Description
A = B All primitive types Returns TRUE if A is equal to B, FALSE otherwise.
A == B All primitive types Synonym for the equal (=) operator.
A <> B All primitive types Returns NULL if A or B is NULL, TRUE if A is not equal to B, FALSE otherwise.
A != B All primitive types Synonym for the not equal (<>) operator.
A < B All primitive types Returns NULL if A or B is NULL, TRUE if A is less than B, FALSE otherwise.
A <= B All primitive types Returns NULL if A or B is NULL, TRUE if A is less than or equal to B, FALSE otherwise.
A !> B All primitive types Returns NULL if A or B is NULL, TRUE if A is not greater than B, FALSE otherwise.
A > B All primitive types Returns NULL if A or B is NULL, TRUE if A is greater than B, FALSE otherwise.
A >= B All primitive types Returns NULL if A or B is NULL, TRUE if A is greater than or equal to B, FALSE otherwise.
A !< B All primitive types Returns NULL if A or B is NULL, TRUE if A is not less than B, FALSE otherwise.
A <=> B All primitive types Returns the same result as the equal (=) operator if both A and B are not NULL, TRUE if both A and B are NULL, FALSE if A or B (but not both) is NULL.

Boolean operators

Boolean operators.
Operator Operand types Description
NOT A Boolean expressions TRUE if A is FALSE, FALSE otherwise.
A AND B Boolean expressions TRUE if A and B are both TRUE, FALSE otherwise.
A OR B Boolean expressions FALSE if A and B are both FALSE, TRUE otherwise.

Data types

Data types can be either primitive types like numeric or string types, or they can be composite types that are built from other primitive or composite types. Composite types can have the following structure:

  • Struct types describe types that are built from a fixed number of named fields. Each field can have its own type.
  • Array types describe a sequence of elements that can have an arbitrary length, but must all have the same type.
  • Map types describe a mapping from keys to values.

Composite types can be nested, as in the following example:

STRUCT<
    firstName: STRING,
    lastName: STRING,
    age: INTEGER,
    addresses: ARRAY<
        STRUCT<
            streetAddress: STRING,
            city: STRING,
            postalCode: STRING,
            country: STRING
        >
    >
>

Some data formats, particularly CSV, do not support composite types. When your query result contains data with a composite type, use an INTO clause to specify an appropriate target format, like JSON.

Primitive types overview

The following primitive data types are supported in Data Engine:

  • Numeric types
  • Strings
  • Date and timestamp
  • Boolean
  • Binary

Numeric types

Numeric data types are summarized in the following table.

Numeric data types.
Identifier Type Bytes Minimum value Maximum value
TINYINT signed integer 1 -128 127
SMALLINT signed integer 2 -32768 32767
INT or INTEGER signed integer 4 -2147483648 2147483647
INTEGER signed integer 4 -2147483648 2147483647
BIGINT or LONG signed integer 8 -9223372036854775808 9223372036854775807
FLOAT single precision floating point number 4 n/a n/a
DOUBLE double precision floating point number 8 n/a n/a
DECIMAL precision of 38 digits n/a -10e37+1 10e37-1

Numeric literals with fractional digits are parsed as DECIMAL by default. For example, 123.45 is treated as DECIMAL(5,2), and 1.234e-3 is treated as DECIMAL(6,6) with value 0.001234, and not as a floating point number. You can force specific data types with a CAST expression or by adding one of the following suffix types:

  • 123L forces BIGINT data type
  • 123BD forces DECIMAL data type
  • 123D forces DOUBLE data type

String types

Strings are represented as STRING data type. The type definitions VARCHAR(n) and CHAR(n) can be used as aliases for STRING. The syntax requires that you specify a maximum length for these types, but no length restriction is enforced.

Date and timestamp types

String values with appropriate formats can be converted to a timestamp or date, by using data types TIMESTAMP or DATE.

-- cast string values to timestamp and date types
SELECT
    CAST('2018-10-31 23:55:00' AS TIMESTAMP),
    CAST('2018-10-31 23:55:00' AS DATE),
    CAST('HELLO' AS TIMESTAMP)
FROM VALUES ('dummy')

The result of the example query is shown in the following table.

Query result for the example cast
CAST(2018-10-31 23:55:00 AS TIMESTAMP) CAST(2018-2-28 23:55:00 AS DATE) CAST(HELLO AS TIMESTAMP)
2018-10-31 23:55:00.0 2018-02-28 null

Boolean type

The BOOLEAN type represents a domain with two values, true or false. Any numeric value that represents zero, for example, 0, 0.0, or 0.0E10, can be cast to false. Numeric values that represent a nonzero value, for example, 1, 1.0, 1.0E10, or 21474.83648 can be cast to true. The string value '0' can be cast to false and '1' can be cast to true. Any other string value is cast to false.

Binary type

A BINARY type represents an array of byte values. Thus, string values can be cast to type BINARY.

Catalog management

The following commands allow users to store table metadata catalog in the Data Engine catalog. By defining the tables, columns, and partitions in the catalog, you can use short table names in the SQL SELECT statements. Each instance of Data Engine has its own catalog, and table definitions are not visible from other instances. For more information, see catalog management.

Create table

createTable

columnDefinition

Create a table definition in the catalog based on the objects in the specified Object Storage location. The LOCATION option is mandatory. If a table or view with the same name exists in the same Data Engine instance, you receive an error, unless the IF NOT EXISTS clause is specified.

The column and partition definitions are optional. If they are not provided, the table schema and partitioning is detected from the structure of the data at the indicated location. If you explicitly provide these definitions, ensure that they match the objects that are stored in Object Storage. See data types for details on the supported column types.

-- create a definition for the table customer
CREATE TABLE customers (
    customerID string,
    companyName string,
    contactName string,
    contactTitle string,
    address string,
    city string,
    region string,
    postalCode string,
    country string,
    phone string,
    fax string
)
USING CSV
LOCATION  cos://us-geo/sql/customers.csv

Before you can use a newly created partitioned table, you must call ALTER TABLE tablename RECOVER PARTITIONS. Otherwise, querying the table returns an empty result.

-- create a definition for the table customers_partitioned
CREATE TABLE customers_partitioned (
    customerID string,
    companyName string,
    contactName string,
    contactTitle string,
    address string,
    city string,
    region string,
    postalCode string,
    country string,
    phone string,
    fax string
)
USING CSV
PARTITIONED BY (COUNTRY)
LOCATION cos://us-geo/sql/customers_partitioned.csv

-- attach table partitions by scanning the location of the table
ALTER TABLE customers_partitioned RECOVER PARTITIONS

An alternative way to create a table definition is to use the automatic schema detection where you do not need to specify any columns.

-- create a definition for the table shippers with automatic schema detection
CREATE TABLE shippers
USING parquet
LOCATION cos://us-geo/sql/shippers.parquet

tableProperty

The tableProperty option consists of one or more key and value pairs.

tableProperty option.
Key Value Default Table format Description
HEADER true or false true CSV Use the HEADER option to specify whether your CSV object has a header included.
DELIMITER single (possibly escaped) character , (comma) CSV Use the DELIMITER option to specify the used delimiter in your CSV objects. All single Unicode characters are allowed as delimiters.
MULTILINE true or false false JSON Use the MULITLINE option to specify whether the JSON object is single or multiline.
MODE 'FAILFAST' or 'PERMISSIVE' 'FAILFAST' JSON FAILFAST mode causes an error when any JSON records are accessed that are either syntactically invalid or do not conform to the specicied table schema. In PERMISSIVE mode, illegal records or fields are implicitly converted to NULLs and do not cause the SQL statement to fail.
-- Example of creating a table definition in the catalog for a CSV data without header line:
CREATE TABLE LAX_CARGO
(Extract_Date timestamp, Report_Date timestamp, Arrival_Departure string,
Domestic_International string, Cargo_Type string, Air_Cargo_Tons int)
USING csv
LOCATION cos://us-geo/sql/LAX_Cargo.csv
options(HEADER=false)

When you create a JSON table, the query engine expects that the data contains only syntactically valid records and that all fields and sub-fields match the schema that you have declared for the table. Any access to invalid data causes the query to fail. You can opt for more relaxed processing by specifying mode='PERMISSIVE' as part of the table options. In permissive mode, invalid JSON data does not fail the query, instead one of the following happens.

  • Fields or sub-fields that do not match the declared schema type are are dropped from the JSON records, so that their value becomes NULL.
  • JSON records that contain a syntax error, such as a misplaced comma or quote, are converted to an empty record, so that all their fields become NULL.

Note that query input from a COS URI with STORED AS JSON format (where the schema is always inferred implicitly) is always processed in permissive mode. So, for COS URIs, invalid records are converted to NULLs. Only when you define a catalog table for more formalized access to the same data, the default handling changes to the stricter failfast mode, in order to avoid silent skipping of invalid data that might cause "data loss" problems. For a catalog table, you need to explicitly specify permissive mode to opt-in to relaxed processing that skips invalid data.

Drop Table

dropTable

Drop a table definition from the catalog. If the table does not exist, you receive an error, unless the IF EXISTS option is specified.

This command does not delete any data in Object Storage. It removes only the table definition from the catalog.

-- drop a definition for the table customer
DROP TABLE customers

Create view

createView

identifierComment

Create a view definition in the catalog, based on existing table and view definitions. If a table or view with the same name exists in the same Data Engine instance, you receive an error, unless the IF NOT EXISTS clause is specified.

The query definition is mandatory. It automatically specifies the SQL query that is used, whenever you use the view in a FROM clause of a query. You can hide some complexity of your data model by creating views on top of your tables. It is also possible to define views on top of other views.

-- create a view on top of table customer
CREATE VIEW CUSTOMER_STATISTICS AS
    SELECT country, region, count(*) customers
        FROM CUSTOMERS
        WHERE region is NOT NULL
        GROUP BY country, region

Drop view

dropView

Drop a view definition from the catalog. If the view does not exist, you receive an error, unless the IF EXISTS option is specified.

Note: This command does not delete any data in Object Storage. It removes only the view definition from the catalog.

-- drop a view definition for the view customer_statistics
DROP VIEW customer_statistics

Alter table partitions

alterTablePartitions

Use alter table to modify the definition of the partitions or to automatically discover the available partitions.

Use the RECOVER PARTITIONS option to automatically replace the table partition metadata with the structure that is detected from Object Storage data by using the location prefix that is specified for the table.

-- replace the table partitions by scanning the location of the table
ALTER TABLE customers_partitioned RECOVER PARTITIONS

partitionSpecification

To add or remove partitions individually, use the ADD PARTITION or DROP PARTITION options.

With the ADD PARTITION option, you can specify an explicit location for the new partition. This way, you can construct a table from object locations that do not share a common Object Storage prefix, or are even located in separate buckets. If the partition location is not specified, it is inferred from the location of the table and the value(s) of the partitioning column(s). ADD PARTITION does not validate the specified or inferred location.

-- alter the table partitions by adding a partition
ALTER TABLE customers_partitioned ADD IF NOT EXISTS PARTITION ( COUNTRY = 'Spain') LOCATION cos://us-geo/sql/customers_partitioned.csv/COUNTRY=Spain
-- alter the table partitions by dropping a partition
ALTER TABLE customers_partitioned DROP IF EXISTS PARTITION ( COUNTRY = 'Nowhere')

The SET LOCATION option can be used to change the location of an existing partition.

-- modify the location of an existing partition
ALTER TABLE customers_partitioned PARTITION (country = 'Spain') SET LOCATION cos://eu-de/sql/customers_partitioned.csv/COUNTRY=Spain

Use the EXISTS option to avoid getting errors during ADD or DROP.

Alter table columns

alterTableColumns

Use alter table to add new columns to the schema of a catalog table.

Adding columns to the schema has no effect on the objects in Object Storage for the table. If some or all of these objects do not contain a column, values for the corresponding rows are treated as NULL. You can add new objects to the table location (for nonpartitioned tables) or new partitions (for partitioned tables) that contain the new columns to evolve the table schema.

-- create a partitioned sample table in PARQUET format
CREATE TABLE customers_addcol USING PARQUET LOCATION cos://us-geo/sql/customers_partitioned.parquet
-- add a new column that does not yet occur in existing partitions. new partitions can add data for that column
ALTER TABLE customers_addcol ADD COLUMNS (priority INTEGER)

Do not use the ADD COLUMNS option with CSV tables. The CSV data format identifies columns by order (not by name), so any schema change leads to a schema mismatch with existing data.

Alternatively, you can perform schema changes by dropping and re-creating catalog tables. It does not affect the stored data in Object Storage. This allows you to reexecute the automatic schema detection when the underlying data is extended with new objects containing more columns. You can also use this method to remove columns from the schema that you do not want to appear in the catalog.

Alter table set table properties

alterTableSetTableProperties

Use alter table set table properties to set new properties to the Hive table. The property is set in key value format. Currently, only the comment property is supported.

-- set property comment to the table CUSTOMERS_PARTITIONED
ALTER TABLE CUSTOMERS_PARTITIONED SET TBLPROPERTIES ('comment' = 'A table comment')

Describe table

describeTable

Return the schema (column names and data types) of a table or view definition. If the table or view does not exist, you receive an error.

-- returns detailed information about the customer table
DESCRIBE TABLE customers_partitioned

Show tables

showTables

Returns the list of the defined tables and views in the catalog. The LIKE option allows to filter for an indicated pattern. Use * as wildcard character.

-- returns all defined tables in the catalog for this instance
SHOW TABLES

Show Partitions

showPartitions

List the defined partitions of a table when a table was created as partitioned. You can filter the returned partitions by using the partitionSpec option.

-- returns all partitions for the table customers_partitioned
SHOW PARTITIONS customers_partitioned

Index management

With the following commands, you can create indexes for data skipping during SQL execution to improve performance and lower the costs of your SQL queries. The indexes store summary metadata for each partition of your table to avoid scanning data that is not needed for the query execution. For more information, see index management.

Create index

createIndex

Create an index on the objects in the specified Object Storage location or on the specified table. Define the required index type for each column that you want to calculate the summary metadata for. Create the index on columns that are used for predicates in the SQL statements.

metaindexIndextype

  • MINMAX: Stores minimum or maximum values for a column for all types, except for complex types.
  • VALUELIST: Stores the list of unique values for the column for all types if the distinct values in that column are low.
  • BLOOMFILTER: Uses bloom filter technique for byte, string, long, integer, or short types if the distinct values in that column are high.
  • GEOSPATIAL: Stores a geospatial bounding box for geometry types.
-- create an index on the columns temp, lat, lng, vid and city of the metergen sample table
CREATE METAINDEX
MINMAX FOR temp,
MINMAX FOR lat,
MINMAX FOR lng,
BLOOMFILTER FOR vid,
VALUELIST FOR city
ON cos://us-geo/sql/metergen STORED AS parquet
-- create an index on the columns customerID and city of the sample table CUSTOMERS_PARTITIONED
CREATE METAINDEX
VALUELIST for city,
BLOOMFILTER for customerID
ON TABLE CUSTOMERS_PARTITIONED
-- create a geospatial index on the column location of the hospitals sample table
CREATE METAINDEX
GEOSPATIAL FOR location
ON cos://us-geo/sql/hospitals_geometry.parquet STORED AS parquet

Before you start to use data skipping index management commands, ensure that you set the base location in Object Storage, where you want the metadata to be stored. Use the following command:

-- set the default location for all indexes
ALTER METAINDEX SET LOCATION cos://us-south/<mybucket>/<mypath>

Drop index

dropIndex

Drop an existing index based on the objects in the specified Object Storage location or on the specified table. Use the following command when the index is no longer needed:

-- drop the index based on the metergen sample data set
DROP METAINDEX ON cos://us-geo/sql/metergen STORED AS parquet

Refresh index

refreshIndex

Refresh an existing index based on the objects in the specified Object Storage location or on the specified table. Use the following command if the data changed and you need to update the index:

-- refresh the index based on metergen sample data set
REFRESH METAINDEX ON cos://us-geo/sql/metergen STORED AS parquet

Describe index

describeIndex

Describe an existing index based on the objects in the specified Object Storage location or on the specified table. Use the following command to receive information of the index, such as index status, types that are used, location where it is stored, or number of objects processed.

-- describe the index based on the metergen sample data set
DESCRIBE METAINDEX ON cos://us-geo/sql/metergen STORED AS parquet

Show indexes

showIndexes

List all stored indexes in the base location. Tables with a different index location are not displayed in the list.

-- list all Metaindexes in the base location
SHOW METAINDEXES

Alter index

alterIndex

You must alter the Object Storage location for all indexes only once to define the base location. If you change it later, Data Engine cannot find the index metadata anymore. Existing index metadata on previous location is not dropped. Therefore, you can always switch back to the old location when needed.

-- set the default location for all indexes
ALTER METAINDEX SET LOCATION cos://us-south/<mybucket>/<mypath>/

Alter table set location

alterTableSetLocation

With this command, you can define a location for this specified Hive table. If you change it later, Data Engine does not find the index metadata anymore. Existing index metadata on previous location is not dropped, therefore you can always switch back to the old location when needed.

-- set the index location for the table CUSTOMERS_PARTITIONED
ALTER TABLE CUSTOMERS_PARTITIONED SET METAINDEX LOCATION cos://us-south/<mybucket>/<mypath>

Alter table drop location

alterTableDropLocation

With this command, you can drop a location for the specified table. Use this command if the index metadata is to be fetched from the base location. The metadata for the index that is stored in Object Storage is not dropped and must be cleaned up manually.

-- set the index location for the table CUSTOMERS_PARTITIONED
ALTER TABLE CUSTOMERS_PARTITIONED DROP METAINDEX LOCATION

IndexAsset

metaindexAsset

The indexAsset is either based on a table or Cloud Object Storage location.

The metaindexAssetLocation is a subset of the externalTableSpec.

The metaindexAssetHiveTable refers to a Hive table.

Miscellaneous definitions

COSURI

A Cloud Object Storage Uniform Resource Identifier (COS URI) is a string of characters that uniquely identifies an object on Cloud Object Storage. By definition URIs are case-insensitive.

The syntax of a Cloud Object Storage URI is thoroughly described in section Table unique resource identifier.

CRN_URI

A database table CRN is a unique identifier that consists of the CRN of a database service instance and a specific table name that instance. The user must have access to this service instance and its credentials.

The syntax of a table CRN is thoroughly described in section Table unique resource identifier.

DB2_TABLE_URI

A Db2 table URI is a string of characters that uniquely identifies a table in an IBM® Db2® on Cloud and IBM® Db2® Warehouse on Cloud instance. The instance must be enabled for IAM and the IBMid if the user must be added as a database user.

The syntax of a Db2 Table URI is thoroughly described in section Table unique resource identifier.

Identifier

An identifier is a name that uniquely identifies an entity. The two types of identifiers are unquoted identifiers and back quoted identifiers.

Unquoted identifier

An unquoted identifier is at least one character long. The following valid characters can be used:

  • Digits 0-9
  • Letters a-z, A-Z
  • Underscore _

Back quoted identifier

It is an identifier that is embraced by grave accent ` characters. A back quoted identifier can contain any character. That includes the grave accent character that must be escaped like this ``.

The following example shows how to add a column name that contains a special character:

SELECT col1 as `Lösung` FROM VALUES 1, 2 ,3

Table Identifier

A table identifier uniquely identifies a table or view in the catalog of the Data Engine instance. The following valid characters can be used:

  • Digits 0-9
  • Letters a-z, A-Z
  • Underscore _

Number

A number is either a signed or unsigned integer or a decimal number.

Unsigned number

An unsigned number is an integer or decimal number without sign or type suffix.

>Unsigned integer number

An unsigned integer is an integer number without sign or type suffix.

Integer number

An integer number is represented by a sequence of at least one digit, thus 0 to 9. The integer number can have a suffix denoting the type of integer number. The three types of integer numbers are:

  • Y: tiny integer number
  • S: small integer number
  • L: large integer number

For more information about data types, see dataType.

Decimal number

The following is a decimal number:

  • A sequence of at least one digit followed by a positive or negative exponent, for example, 10E2 represents integer 1000 and 1E-1 represents 0.1.
  • A decimal number, for example, 3.14.
  • A decimal number followed by a positive or negative exponent, for example, 3.14E+3 represents 3140.00 or 3.14E-3 represents 0.00314.

The decimal number can have a suffix denoting the type of decimal number. The two types of decimal numbers are:

  • D: double decimal number
  • BD: large decimal number

For more information about data types, see dataType.

String

A string is a sequence of arbitrary characters that includes escaped characters, for example, \t, either enclosed in single quotation marks ', or double quotation marks ". To include any quotation mark characters in the string, they must be escaped as \\' or \\".