Skip to content

Bodo SQL

Bodo SQL provides high performance and scalable SQL query execution using Bodo's HPC capabilities and optimizations. It also provides native Python/SQL integration as well as SQL to Pandas conversion for the first time. BodoSQL is in early stages and its capabilities are expanding rapidly.

Getting Started

Installation

Bodo SQL is currently in Beta. Install it using:

conda install bodosql -c bodo.ai -c conda-forge

Using Bodo SQL

The example below demonstrates using Bodo SQL in Python programs. It loads data into a dataframe, runs a SQL query on the data, and runs Python/Pandas code on query results:

import pandas as pd
import bodo
import bodosql

@bodo.jit
def f(filename):
    df1 = pd.read_parquet(filename)
    bc = bodosql.BodoSQLContext({"table1": df1})
    df2 = bc.sql("SELECT A FROM table1 WHERE B > 4")
    print(df2.A.sum())

f("my_data.pq")

This program is fully type checked, optimized and parallelized by Bodo end-to-end. BodoSQLContext creates a SQL environment with tables created from dataframes. BodoSQLContext.sql() runs a SQL query and returns the results as a dataframe. BodoSQLContext can be used outside Bodo JIT functions if necessary as well.

You can run this example by creating my_data.pq:

import pandas as pd
import numpy as np

NUM_GROUPS = 30
NUM_ROWS = 20_000_000
df = pd.DataFrame({
    "A": np.arange(NUM_ROWS) % NUM_GROUPS,
    "B": np.arange(NUM_ROWS)
})
df.to_parquet("my_data.pq")

To run the example, save it in a file called example.py and run it using mpiexec, e.g.:

mpiexec -n 8 python example.py

Aliasing

In all but the most trivial cases, Bodo SQL generates internal names to avoid conflicts in the intermediate dataframes. By default, Bodo SQL does not rename the columns for the final output of a query using a consistent approach. For example the query:

bc.sql("SELECT SUM(A) FROM table1 WHERE B > 4")
Results in an output column named $EXPR0. To reliably reference this column later in your code, we highly recommend using aliases for all columns that are the final outputs of a query, such as:

bc.sql("SELECT SUM(A) as sum_col FROM table1 WHERE B > 4")

Note

BodoSQL supports using aliases generated in SELECT inside GROUP BY and HAVING in the same query, but you cannot do so with WHERE.

Supported Operations

We currently support the following SQL query statements and clauses with Bodo SQL, and are continuously adding support towards completeness. Note that Bodo SQL ignores casing of keywords, and column and table names, except for the final output column name. Therefore, select a from table1 is treated the same as SELECT A FROM Table1, except for the names of the final output columns (a vs A).

SELECT

The SELECT statement is used to select data in the form of columns. The data returned from Bodo SQL is stored in a dataframe.

SELECT <COLUMN_NAMESFROM <TABLE_NAME>

For Instance:

SELECT A FROM customers

Example Usage:

>>>@bodo.jit
... def g(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name FROM customers"
...    res = bc.sql(query)
...    return res

>>>customers_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez",
...                 "Demetrius Chavez","Weston Jefferson","Jonathon Middleton",
...                 "Shawn Winters","Keely Hutchinson", "Darryl Rosales",],
...     "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43, 68.34, 12764.50, 3489.25, 654.24, 25645.39]
... })

>>>g(customers_df)
                name
0       Deangelo Todd
1        Nikolai Kent
2          Eden Heath
3    Taliyah Martinez
4    Demetrius Chavez
5    Weston Jefferson
6  Jonathon Middleton
7       Shawn Winters
8    Keely Hutchinson
9      Darryl Rosales

SELECT DISTINCT

The SELECT DISTINCT statement is used to return only distinct (different) values:

SELECT DISTINCT <COLUMN_NAMESFROM <TABLE_NAME>

DISTINCT can be used in a SELECT statement or inside an aggregate function. For example:

SELECT DISTINCT A FROM table1

SELECT COUNT DISTINCT A FROM table1

Example Usage

>>>@bodo.jit
... def g(df):
...    bc = bodosql.BodoSQLContext({"payments":df})
...    query = "SELECT DISTINCT paymentType FROM payments"
...    res = bc.sql(query)
...    return res

>>>payment_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "paymentType": ["VISA", "VISA", "AMEX", "VISA", "WIRE", "VISA", "VISA", "WIRE", "VISA", "AMEX"],
... })

>>>g(payment_df) # inside SELECT
paymentType
0        VISA
2        AMEX
4        WIRE

>>>def g(df):
...    bc = bodosql.BodoSQLContext({"payments":df})
...    query = "SELECT COUNT(DISTINCT paymentType) as num_payment_types FROM payments"
...    res = bc.sql(query)
...    return res

>>>g(payment_df) # inside aggregate
num_payment_types
0          3

WHERE

The WHERE clause on columns can be used to filter records that satisfy specific conditions:

SELECT <COLUMN_NAMESFROM <TABLE_NAMEWHERE <CONDITION>

For Example:

SELECT A FROM table1 WHERE B > 4

Example Usage

>>>@bodo.jit
... def g(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name FROM customers WHERE balance 3000"
...    res = bc.sql(query)
...    return res

>>>customers_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez",
...                 "Demetrius Chavez","Weston Jefferson","Jonathon Middleton",
...                 "Shawn Winters","Keely Hutchinson", "Darryl Rosales",],
...     "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43, 68.34, 12764.50, 3489.25, 654.24, 25645.39]
... })

>>>g(customers_df)
                name
3    Taliyah Martinez
6  Jonathon Middleton
7       Shawn Winters
9      Darryl Rosales

ORDER BY

The ORDER BY keyword sorts the resulting dataframe in ascending or descending order, with NULL values either at the start or end of the column. By default, it sorts the records in ascending order with null values at the end. For descending order and nulls at the front, the DESC and NULLS FIRST keywords can be used:

SELECT <COLUMN_NAMES>
FROM <TABLE_NAME>
ORDER BY <ORDERED_COLUMN_NAMES[ASC|DESC] [NULLS FIRST|LAST]

For Example:

SELECT A, B FROM table1 ORDER BY B, A DESC NULLS FIRST

Example Usage

>>>@bodo.jit
... def g(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name, balance FROM customers ORDER BY balance"
...    res = bc.sql(query)
...    return res

>>>customers_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez",
...                 "Demetrius Chavez","Weston Jefferson","Jonathon Middleton",
...                 "Shawn Winters","Keely Hutchinson", "Darryl Rosales",],
...     "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43, 68.34, 12764.50, 3489.25, 654.24, 25645.39]
... })

>>>g(customers_df)
                name   balance
2          Eden Heath     23.58
5    Weston Jefferson     68.34
8    Keely Hutchinson    654.24
4    Demetrius Chavez    943.43
0       Deangelo Todd   1123.34
1        Nikolai Kent   2133.43
7       Shawn Winters   3489.25
3    Taliyah Martinez   8345.15
6  Jonathon Middleton  12764.50
9      Darryl Rosales  25645.39

LIMIT

Bodo SQL supports the LIMIT keyword to select a limited number of rows. This keyword can optionally include an offset:

SELECT <COLUMN_NAMES>
FROM <TABLE_NAME>
WHERE <CONDITION>
LIMIT <LIMIT_NUMBEROFFSET <OFFSET_NUMBER>
For Example:

SELECT A FROM table1 LIMIT 5

SELECT B FROM table2 LIMIT 8 OFFSET 3
Specifying a limit and offset can be also be written as:

LIMIT <OFFSET_NUMBER>, <LIMIT_NUMBER>
For Example:

SELECT B FROM table2 LIMIT 3, 8
Example Usage

>>>@bodo.jit
... def g1(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name FROM customers LIMIT 4"
...    res = bc.sql(query)
...    return res

>>>@bodo.jit
... def g2(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name FROM customers LIMIT 4 OFFSET 2"
...    res = bc.sql(query)
...    return res

>>>customers_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez",
...                 "Demetrius Chavez","Weston Jefferson","Jonathon Middleton",
...                 "Shawn Winters","Keely Hutchinson", "Darryl Rosales",],
...     "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43, 68.34, 12764.50, 3489.25, 654.24, 25645.39]
... })

>>>g1(customers_df) # LIMIT 4
               name
0     Deangelo Todd
1      Nikolai Kent
2        Eden Heath
3  Taliyah Martinez

>>>g2(customers_df) # LIMIT 4 OFFSET 2
               name
2        Eden Heath
3  Taliyah Martinez
4  Demetrius Chavez
5  Weston Jefferson

NOT IN

The IN determines if a value can be chosen a list of options. Currently we support lists of literals or columns with matching types:

SELECT <COLUMN_NAMES>
FROM <TABLE_NAME>
WHERE <COLUMN_NAMEIN (<val1>, <val2>, ... <valN>)
For example:
SELECT A FROM table1 WHERE A IN (5, 10, 15, 20, 25)
Example Usage
>>>@bodo.jit
... def g1(df):
...    bc = bodosql.BodoSQLContext({"payments":df})
...    query = "SELECT customerID FROM payments WHERE paymentType IN ('AMEX', 'WIRE')"
...    res = bc.sql(query)
...    return res

>>>@bodo.jit
... def g2(df):
...    bc = bodosql.BodoSQLContext({"payments":df})
...    query = "SELECT customerID FROM payments WHERE paymentType NOT IN ('AMEX', 'VISA')"
...    res = bc.sql(query)
...    return res

>>>payment_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "paymentType": ["VISA", "VISA", "AMEX", "VISA", "WIRE", "VISA", "VISA", "WIRE", "VISA", "AMEX"],
... })

>>>g1(payment_df) # IN
   customerID
2           2
4           4
7           7
9           9

>>>g2(payment_df) # NOT IN
   customerID
4           4
7           7

NOT BETWEEN

The BETWEEN operator selects values within a given range. The values can be numbers, text, or datetimes. The BETWEEN operator is inclusive: begin and end values are included:

SELECT <COLUMN_NAMES>
FROM <TABLE_NAME>
WHERE <COLUMN_NAMEBETWEEN <VALUE1AND <VALUE2>
For example:
SELECT A FROM table1 WHERE A BETWEEN 10 AND 100
Example Usage
>>>@bodo.jit
... def g(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name, balance FROM customers WHERE balance BETWEEN 1000 and 5000"
...    res = bc.sql(query)
...    return res

>>>@bodo.jit
... def g2(df):
...    bc = bodosql.BodoSQLContext({"customers":df})
...    query = "SELECT name, balance FROM customers WHERE balance NOT BETWEEN 100 and 10000"
...    res = bc.sql(query)
...    return res

>>>customers_df = pd.DataFrame({
...     "customerID": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
...     "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez",
...                 "Demetrius Chavez","Weston Jefferson","Jonathon Middleton",
...                 "Shawn Winters","Keely Hutchinson", "Darryl Rosales",],
...     "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43, 68.34, 12764.50, 3489.25, 654.24, 25645.39]
... })

>>>g1(payment_df) # BETWEEN
            name  balance
0  Deangelo Todd  1123.34
1   Nikolai Kent  2133.43
7  Shawn Winters  3489.25

>>>g2(payment_df) # NOT BETWEEN
                 name   balance
2          Eden Heath     23.58
5    Weston Jefferson     68.34
6  Jonathon Middleton  12764.50
9      Darryl Rosales  25645.39

CAST

THE CAST operator converts an input from one type to another. In many cases casts are created implicitly, but this operator can be used to force a type conversion.

The following casts are currently supported. Please refer to supported_dataframe_data_types for the Python types for each type keyword:

From To Notes
VARCHAR VARCHAR
VARCHAR TINYINT/SMALLINT/INTEGER/BIGINT
VARCHAR FLOAT/DOUBLE
VARCHAR DECIMAL Equivalent to DOUBLE. This may change in the future.
VARCHAR TIMESTAMP
VARCHAR DATE Truncates to date but is still Timestamp type. This may change in the future.
TINYINT/SMALLINT/INTEGER/BIGINT VARCHAR
TINYINT/SMALLINT/INTEGER/BIGINT TINYINT/SMALLINT/INTEGER/BIGINT
TINYINT/SMALLINT/INTEGER/BIGINT FLOAT/DOUBLE
TINYINT/SMALLINT/INTEGER/BIGINT DECIMAL Equivalent to DOUBLE. This may change in the future.
TINYINT/SMALLINT/INTEGER/BIGINT TIMESTAMP
FLOAT/DOUBLE VARCHAR
FLOAT/DOUBLE TINYINT/SMALLINT/INTEGER/BIGINT
FLOAT/DOUBLE FLOAT/DOUBLE
FLOAT/DOUBLE DECIMAL Equivalent to DOUBLE. This may change in the future
TIMESTAMP VARCHAR
TIMESTAMP TINYINT/SMALLINT/INTEGER/BIGINT
TIMESTAMP TIMESTAMP
TIMESTAMP DATE Truncates to date but is still Timestamp type. This may change in the future.

Note

CAST correctness can often not be determined at compile time. Users are responsible for ensuring that conversion is possible (e.g. CAST(str_col as INTEGER)).

JOIN

A JOIN clause is used to combine rows from two or more tables, based on a related column between them:

SELECT <COLUMN_NAMES>
  FROM <LEFT_TABLE_NAME>
  <JOIN_TYPE<RIGHT_TABLE_NAME>
  ON <LEFT_TABLE_COLUMN_NAME= <RIGHT_TABLE_COLUMN_NAME>
For example:
SELECT table1.A, table1.B FROM table1 JOIN table2 on table1.A = table2.C
Here are the different types of the joins in SQL:

  • (INNER) JOIN: returns records that have matching values in both tables
  • LEFT (OUTER) JOIN: returns all records from the left table, and the matched records from the right table
  • RIGHT (OUTER) JOIN: returns all records from the right table, and the matched records from the left table
  • FULL (OUTER) JOIN: returns all records when there is a match in either left or right table

Bodo SQL currently supports inner join on all conditions, but all outer joins are only supported on an equality between columns.

Example Usage

>>>@bodo.jit
... def g1(df1, df2):
...    bc = bodosql.BodoSQLContext({"customers":df1, "payments":df2})
...    query = "SELECT name, paymentType FROM customers JOIN payments ON customers.customerID = payments.customerID"
...    res = bc.sql(query)
...    return res

>>>@bodo.jit
... def g2(df1, df2):
...    bc = bodosql.BodoSQLContext({"customers":df1, "payments":df2})
...    query = "SELECT name, paymentType FROM customers FULL JOIN payments ON customers.customerID = payments.customerID"
...    res = bc.sql(query)
...    return res

>>>customer_df = pd.DataFrame({
...    "customerID": [0, 2, 4, 5, 7,],
...    "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez","Demetrius Chavez",],
...    "address": ["223 Iroquois LanenWest New York, NJ 07093","37 Depot StreetnTaunton, MA 02780",
...                "639 Maple St.nNorth Kingstown, RI 02852","93 Bowman Rd.nChester, PA 19013",
...                "513 Manchester Ave.nWindsor, CT 06095",],
...    "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43,]
... })
>>>payment_df = pd.DataFrame({
...     "customerID": [0, 1, 4, 6, 7],
...     "paymentType": ["VISA", "VISA", "AMEX", "VISA", "WIRE",],
... })

>>>g1(customer_df, payment_df) # INNER JOIN
               name paymentType
0     Deangelo Todd        VISA
1        Eden Heath        AMEX
2  Demetrius Chavez        WIRE

>>>g2(customer_df, payment_df) # OUTER JOIN
               name paymentType
0     Deangelo Todd        VISA
1      Nikolai Kent         NaN
2        Eden Heath        AMEX
3  Taliyah Martinez         NaN
4  Demetrius Chavez        WIRE
5               NaN        VISA
6               NaN        VISA

UNION

The UNION operator is used to combine the result-set of two SELECT statements:

SELECT <COLUMN_NAMESFROM <TABLE1>
UNION
SELECT <COLUMN_NAMESFROM <TABLE2>
Each SELECT statement within the UNION clause must have the same number of columns. The columns must also have similar data types. The output of the UNION is the set of rows which are present in either of the input SELECT statements.

The UNION operator selects only the distinct values from the inputs by default. To allow duplicate values, use UNION ALL:

SELECT <COLUMN_NAMESFROM <TABLE1>
UNION ALL
SELECT <COLUMN_NAMESFROM <TABLE2>

Example Usage

>>>@bodo.jit
... def g1(df):
...    bc = bodosql.BodoSQLContext({"customers":df1, "payments":df2})
...    query = "SELECT name, paymentType FROM customers JOIN payments ON customers.customerID = payments.customerID WHERE paymentType in ('WIRE')
...             UNION SELECT name, paymentType FROM customers JOIN payments ON customers.customerID = payments.customerID WHERE balance < 1000"
...    res = bc.sql(query)
...    return res

>>>@bodo.jit
... def g2(df):
...    bc = bodosql.BodoSQLContext({"customers":df1, "payments":df2})
...    query = "SELECT name, paymentType FROM customers JOIN payments ON customers.customerID = payments.customerID WHERE paymentType in ('WIRE')
...             UNION ALL SELECT name, paymentType FROM customers JOIN payments ON customers.customerID = payments.customerID WHERE balance < 1000"
...    res = bc.sql(query)
...    return res

>>>customer_df = pd.DataFrame({
...    "customerID": [0, 2, 4, 5, 7,],
...    "name": ["Deangelo Todd","Nikolai Kent","Eden Heath", "Taliyah Martinez","Demetrius Chavez",],
...    "address": ["223 Iroquois LanenWest New York, NJ 07093","37 Depot StreetnTaunton, MA 02780",
...                "639 Maple St.nNorth Kingstown, RI 02852","93 Bowman Rd.nChester, PA 19013",
...                "513 Manchester Ave.nWindsor, CT 06095",],
...    "balance": [1123.34, 2133.43, 23.58, 8345.15, 943.43,]
... })
>>>payment_df = pd.DataFrame({
...     "customerID": [0, 1, 4, 6, 7],
...     "paymentType": ["VISA", "VISA", "AMEX", "VISA", "WIRE",],
... })

>>>g1(customer_df, payment_df) # UNION
           name paymentType  balance
0  Demetrius Chavez        WIRE   943.43
0        Eden Heath        AMEX    23.58

>>>g2(customer_df, payment_df) # UNION ALL
            name paymentType  balance
0  Demetrius Chavez        WIRE   943.43
0        Eden Heath        AMEX    23.58
1  Demetrius Chavez        WIRE   943.43

INTERSECT

The INTERSECT operator is used to calculate the intersection of two SELECT statements:

SELECT <COLUMN_NAMESFROM <TABLE1>
INTERSECT
SELECT <COLUMN_NAMESFROM <TABLE2>

Each SELECT statement within the INTERSECT clause must have the same number of columns. The columns must also have similar data types. The output of the INTERSECT is the set of rows which are present in both of the input SELECT statements. The INTERSECT operator selects only the distinct values from the inputs.

GROUP BY

The GROUP BY statement groups rows that have the same values into summary rows, like "find the number of customers in each country". The GROUP BY statement is often used with aggregate functions to group the result-set by one or more columns:

SELECT <COLUMN_NAMES>
FROM <TABLE_NAME>
WHERE <CONDITION>
GROUP BY <GROUP_EXPRESION>
ORDER BY <COLUMN_NAMES>

For example:

SELECT MAX(A) FROM table1 GROUP BY B
GROUP BY statements also referring to columns by alias or column number:
SELECT MAX(A), B - 1 as val FROM table1 GROUP BY val
SELECT MAX(A), B FROM table1 GROUP BY 2

BodoSQL supports several subclauses that enable grouping by multiple different sets of columns in the same SELECT statment. GROUPING SETS is the first. It is equivalent to performing a group by for each specified set (seting each column not present in the grouping set to null), and unioning the results. For example:

SELECT MAX(A), B, C FROM table1 GROUP BY GROUPING SETS (B, B, (B, C), ())

This is equivalent to:

SELECT * FROM
    (SELECT MAX(A), B, null FROM table1 GROUP BY B)
UNION
    (SELECT MAX(A), B, null FROM table1 GROUP BY B)
UNION
    (SELECT MAX(A), B, C FROM table1 GROUP BY B, C)
UNION
    (SELECT MAX(A), null, null FROM table1)

Note

The above example is not valid BodoSQL code, as we do not support null literals. It is used only to show the null filling behavior.

CUBE is equivalent to grouping by all possible permutations of the specified set. For example:

SELECT MAX(A), B, C FROM table1 GROUP BY CUBE(B, C)

Is equivalent to

SELECT MAX(A), B, C FROM table1 GROUP BY GROUPING SETS ((B, C), (B), (C), ())

ROLLUP is equivalent to grouping by n + 1 grouping sets, where each set is constructed by dropping the rightmost element from the previous set, until no elements remain in the grouping set. For example:

SELECT MAX(A), B, C FROM table1 GROUP BY ROLLUP(B, C, D)

Is equivalent to

SELECT MAX(A), B, C FROM table1 GROUP BY GROUPING SETS ((B, C, D), (B, C), (B), ())

CUBE and ROLLUP can be nested into a GROUPING SETS clause. For example:

SELECT MAX(A), B, C GROUP BY GROUPING SETS (ROLLUP(B, C, D), CUBE(B, C), (A))

Which is equivalent to

SELECT MAX(A), B, C GROUP BY GROUPING SETS ((B, C, D), (B, C), (B), (), (B, C), (B), (C), (), (A))

HAVING

The HAVING clause is used for filtering with GROUP BY. HAVING applies the filter after generating the groups, whereas WHERE applies the filter before generating any groups:

SELECT column_name(s)
FROM table_name
WHERE condition
GROUP BY column_name(s)
HAVING condition
For example:
SELECT MAX(A) FROM table1 GROUP BY B HAVING C < 0
HAVING statements also referring to columns by aliases used in the GROUP BY:
SELECT MAX(A), B - 1 as val FROM table1 GROUP BY val HAVING val 5

QUALIFY

QUALIFY is similar to HAVING, except it applies filters after computing the results of one or more window functions. This occurs after WHERE and HAVING. For example:

SELECT column_name(s),
FROM table_name
WHERE condition
GROUP BY column_name(s)
HAVING condition
QUALIFY MAX(A) OVER (PARTITION BY B ORDER BY C ROWS BETWEEN 1 FOLLOWING AND 1 PRECEDING) > 1

Is equivalent to

SELECT column_name(s) FROM
    (SELECT column_name(s), MAX(A) OVER (PARTITION BY B ORDER BY C ROWS BETWEEN 1 FOLLOWING AND 1 PRECEDING) as window_output
    FROM table_name
    WHERE condition
    GROUP BY column_name(s)
    HAVING condition)
WHERE window_output > 1

CASE

The CASE statement goes through conditions and returns a value when the first condition is met:

SELECT CASE WHEN cond1 THEN value1 WHEN cond2 THEN value2 ... ELSE valueN END
For example:
SELECT (CASE WHEN A 1 THEN A ELSE B END) as mycol FROM table1
If the types of the possible return values are different, BodoSQL will attempt to cast them all to a common type, which is currently undefined behavior. The last else clause can optionally be excluded, in which case, the CASE statement will return null if none of the conditions are met. For example:
SELECT (CASE WHEN A < 0 THEN 0 END) as mycol FROM table1
is equivalent to:
SELECT (CASE WHEN A < 0 THEN 0 ELSE NULL END) as mycol FROM table1

LIKE

The LIKE clause is used to filter the strings in a column to those that match a pattern:

SELECT column_name(s) FROM table_name WHERE column LIKE pattern
In the pattern we support the wildcards % and _. For example:
SELECT A FROM table1 WHERE B LIKE '%py'

GREATEST

The GREATEST clause is used to return the largest value from a list of columns:

SELECT GREATEST(col1, col2, ..., colN) FROM table_name
For example:
SELECT GREATEST(A, B, C) FROM table1

LEAST

The LEAST clause is used to return the smallest value from a list of columns:

SELECT LEAST(col1, col2, ..., colN) FROM table_name
For example:
SELECT LEAST(A, B, C) FROM table1

PIVOT

The PIVOT clause is used to transpose specific data rows in one or more columns into a set of columns in a new DataFrame:

SELECT col1, ..., colN FROM table_name PIVOT (
    AGG_FUNC_1(colName or pivotVar) AS alias1, ...,  AGG_FUNC_N(colName or pivotVar) as aliasN
    FOR pivotVar IN (ROW_VALUE_1 as row_alias_1, ..., ROW_VALUE_N as row_alias_N)
)
PIVOT produces a new column for each pair of pivotVar and aggregation functions.

For example:

SELECT single_sum_a, single_avg_c, triple_sum_a, triple_avg_c FROM table1 PIVOT (
    SUM(A) AS sum_a, AVG(C) AS avg_c
    FOR A IN (1 as single, 3 as triple)
)
Here single_sum_a will contain sum(A) where A = 1, single_avg_c will contain AVG(C) where A = 1 etc.

If you explicitly specify other columns as the output, those columns will be used to group the pivot columns. For example:

SELECT B, single_sum_a, single_avg_c, triple_sum_a, triple_avg_c FROM table1 PIVOT (
    SUM(A) AS sum_a, AVG(C) AS avg_c
    FOR A IN (1 as single, 3 as triple)
)
Contains 1 row for each unique group in B. The pivotVar can also require values to match in multiple columns. For example:
SELECT * FROM table1 PIVOT (
    SUM(A) AS sum_a, AVG(C) AS avg_c
    FOR (A, B) IN ((1, 4) as col1, (2, 5) as col2)
)

WITH

The WITH clause can be used to name subqueries:

WITH sub_table AS (SELECT column_name(s) FROM table_name)
SELECT column_name(s) FROM sub_table
For example:
WITH subtable as (SELECT MAX(A) as max_al FROM table1 GROUP BY B)
SELECT MAX(max_val) FROM subtable

Aliasing

SQL aliases are used to give a table, or a column in a table, a temporary name:

SELECT <COLUMN_NAMEAS <ALIAS>
FROM <TABLE_NAME>

For example:

Select SUM(A) as total FROM table1

We strongly recommend using aliases for the final outputs of any queries to ensure all column names are predictable.

Operators

Arithmetic

  • Bodo SQL currently supports the following arithmetic operators:

    • + (addition)
    • - (subtraction)
    • * (multiplication)
    • / (true division)
    • % (modulo)

Comparison

  • Bodo SQL currently supports the following comparison operators:

    • = (equal to)
    • > (greater than)
    • < (less than)
    • >= (greater than or equal to)
    • <= (less than or equal to)
    • <> (not equal to)
    • != (not equal to)
    • <=> (equal to or both inputs are null)

Logical

  • Bodo SQL currently supports the following logical operators:

    • AND
    • OR
    • NOT

String

  • Bodo SQL currently supports the following string operators:

    • || (string concatenation)

Numeric Functions

Except where otherwise specified, the inputs to each of these functions can be any numeric type, column or scalar. Here is an example using MOD:

SELECT MOD(12.2, A) FROM table1

Bodo SQL Currently supports the following Numeric Functions:

ABS

  • ABS(n)

    Returns the absolute value of n

COS

  • COS(n)

    Calculates the Cosine of n

SIN

  • SIN(n)

    Calculates the Sine of n

TAN

  • TAN(n)

    Calculates the Tangent of n

ACOS

  • ACOS(n)

    Calculates the Arccosine of n

ASIN

  • ASIN(n)

    Calculates the Arcsine of n

ATAN

  • ATAN(n)

    Calculates the Arctangent of n

ATAN2

  • ATAN2(A, B)

    Calculates the Arctangent of A divided by B

COTAN

  • COTAN(X)

    Calculates the Cotangent of X

CEIL

  • CEIL(X)

    Converts X to an integer, rounding towards positive infinity

CEILING

  • CEILING(X)

    Equivalent to CEIL

FLOOR

  • FLOOR(X)

    Converts X to an integer, rounding towards negative infinity

DEGREES

  • DEGREES(X)

    Converts a value in radians to the corresponding value in degrees

RADIANS

  • RADIANS(X)

    Converts a value in radians to the corresponding value in degrees

LOG10

  • LOG10(X)

    Computes Log base 10 of x. Returns NaN for negative inputs, and -inf for 0 inputs.

LOG

  • LOG(X)

    Equivalent to LOG10(x)

LOG10

  • LOG10(X)

    Computes Log base 2 of x. Returns NaN for negative inputs, and -inf for 0 inputs.

LN

  • LN(X)

    Computes the natural log of x. Returns NaN for negative inputs, and -inf for 0 inputs.

MOD

  • MOD(A,B)

    Computes A modulo B.

CONV

  • CONV(X, current_base, new_base)

    CONV takes a string representation of an integer value, it's current_base, and the base to convert that argument to. CONV returns a new string, that represents the value in the new base. CONV is only supported for converting to/from base 2, 8, 10, and 16.

    For example:

    CONV('10', 10, 2) =='1010'
    CONV('10', 2, 10) =='2'
    CONV('FA', 16, 10) =='250'
    

    SQRT

    • SQRT(X)

    Computes the square root of x. Returns NaN for negative inputs, and -inf for 0 inputs.

PI

  • PI()

    Returns the value of PI

POW, POWER

  • POW(A, B), POWER(A, B)

    Returns A to the power of B. Returns NaN if A is negative, and B is a float. POW(0,0) is 1

EXP

  • EXP(X)

    Returns e to the power of X

SIGN

  • SIGN(X)

    Returns 1 if X 0, -1 if X < 0, and 0 if X = 0

ROUND

  • ROUND(X, num_decimal_places)

    Rounds X to the specified number of decimal places

TRUNCATE

  • TRUNCATE(X, num_decimal_places)

    Equivalent to ROUND(X, num_decimal_places)

BITAND

  • BITAND(A, B)

    Returns the bitwise-and of its inputs.

BITOR

  • BITOR(A, B)

    Returns the bitwise-or of its inputs.

BITXOR

  • BITOR(A, B)

    Returns the bitwise-xor of its inputs.

BITNOT

  • BITNOT(A)

    Returns the bitwise-negation of its input.

BITSHIFTLEFT

  • BITSHIFTLEFT(A, B)

    Returns the bitwise-leftshift of its inputs. Note: the output is always of type int64. Undefined behavior when B is negative or too large.

BITSHIFTRIGHT

  • BITSHIFTRIHGT(A, B)

    Returns the bitwise-rightshift of its inputs. Undefined behavior when B is negative or too large.

GETBIT

  • GETBIT(A, B)

    Returns the bit of A corresponding to location B, where 0 is the rightmost bit. Undefined behavior when B is negative or too large.

BOOLAND

  • BOOLAND(A, B)

    Returns true when A and B are both non-null non-zero. Returns false when one of the arguments is zero and the other is either zero or NULL. Returns NULL otherwise.

BOOLOR

  • BOOLOR(A, B)

    Returns true if either A or B is non-null and non-zero. Returns false if both A and B are zero. Returns NULL otherwise.

BOOLXOR

  • BOOLXOR(A, B)

    Returns true if one of A and B is zero and the other is non-zero. Returns false if A and B are both zero or both non-zero. Returns NULL if either A or B is NULL.

BOOLNOT

  • BOOLNOT(A)

    Returns true if A is zero. Returns false if A is non-zero. Returns NULL if A is NULL.

REGR_VALX

  • REGR_VALX(Y, X)

    Returns NULL if either input is NULL, otherwise X

REGR_VALY

  • REGR_VALY(Y, X)

    Returns NULL if either input is NULL, otherwise Y

Aggregation Functions

Bodo SQL Currently supports the following Aggregation Functions on all types:

COUNT

  • COUNT

    Count the number of elements in a column or group.

In addition, Bodo SQL also supports the following functions on numeric types

AVG

  • AVG

    Compute the mean for a column.

MAX

  • MAX

    Compute the max value for a column.

MIN

  • MIN

    Compute the min value for a column.

STDDEV

  • STDDEV

    Compute the standard deviation for a column with N - 1 degrees of freedom.

STDDEV_SAMP

  • STDDEV_SAMP

    Compute the standard deviation for a column with N - 1 degrees of freedom.

STDDEV_POP

  • STDDEV_POP

    Compute the standard deviation for a column with N degrees of freedom.

SUM

  • SUM

    Compute the sum for a column.

VARIANCE

  • VARIANCE

    Compute the variance for a column with N - 1 degrees of freedom.

VAR_SAMP

  • VAR_SAMP

    Compute the variance for a column with N - 1 degrees of freedom.

VAR_POP

  • VAR_POP

    Compute the variance for a column with N degrees of freedom.

All aggregate functions have the syntax:

SELECT AGGREGATE_FUNCTION(<COLUMN_EXPRESSION>)
FROM <TABLE_NAME>
GROUP BY <COLUMN_NAMES>

These functions can be used either in a groupby clause, where they will be computed for each group, or by itself on an entire column expression. For example:

SELECT AVG(A) FROM table1 GROUP BY B

SELECT COUNT(Distinct A) FROM table1

Timestamp Functions

Bodo SQL currently supports the following Timestamp functions:

DATEDIFF

  • DATEDIFF(timestamp_val1, timestamp_val2)

    Computes the difference in days between two Timestamp values

STR_TO_DATE

  • STR_TO_DATE(str_val, literal_format_string)

    Converts a string value to a Timestamp value given a literal format string. If a year, month, and day value is not specified, they default to 1900, 01, and 01 respectively. Will throw a runtime error if the string cannot be parsed into the expected values. See DATE_FORMAT for recognized formatting characters.

    For example:

    STR_TO_DATE('2020 01 12', '%Y %m %d') ==Timestamp '2020-01-12'
    STR_TO_DATE('01 12', '%m %d') ==Timestamp '1900-01-12'
    STR_TO_DATE('hello world', '%Y %m %d') ==RUNTIME ERROR
    

DATE_FORMAT

  • DATE_FORMAT(timestamp_val, literal_format_string)

    Converts a timestamp value to a String value given a scalar format string.

    Recognized formatting characters:

    • %i Minutes, zero padded (00 to 59)
    • %M Full month name (January to December)
    • %r Time in format in the format (hh:mm:ss AM/PM)
    • %s Seconds, zero padded (00 to 59)
    • %T Time in format in the format (hh:mm:ss)
    • %T Time in format in the format (hh:mm:ss)
    • %u week of year, where monday is the first day of the week(00 to 53)
    • %a Abbreviated weekday name (sun-sat)
    • %b Abbreviated month name (jan-dec)
    • %f Microseconds, left padded with 0's, (000000 to 999999)
    • %H Hour, zero padded (00 to 23)
    • %j Day Of Year, left padded with 0's (001 to 366)
    • %m Month number (00 to 12)
    • %p AM or PM, depending on the time of day
    • %d Day of month, zero padded (01 to 31)
    • %Y Year as a 4 digit value
    • %y Year as a 2 digit value, zero padded (00 to 99)
    • %U Week of year where sunday is the first day of the week (00 to 53)
    • %S Seconds, zero padded (00 to 59)

    For example:

    DATE_FORMAT(Timestamp '2020-01-12', '%Y %m %d') =='2020 01 12'
    DATE_FORMAT(Timestamp '2020-01-12 13:39:12', 'The time was %T %p. It was a %u') =='The time was 13:39:12 PM. It was a Sunday'
    

DATE_ADD

  • DATE_ADD(timestamp_val, interval)

    Computes a timestamp column by adding an interval column/scalar to a timestamp value. If the first argument is a string representation of a timestamp, Bodo will cast the value to a timestamp.

DATE_SUB

  • DATE_SUB(timestamp_val, interval)

    Computes a timestamp column by subtracting an interval column/scalar to a timestamp value. If the first argument is a string representation of a timestamp, Bodo will cast the value to a timestamp.

DATE_TRUNC

  • DATE_TRUNC(str_literal, timestamp_val)

    Truncates a timestamp to the provided str_literal field. str_literal must be a compile time constant and one of:

    • "MONTH"
    • "WEEK"
    • "DAY"
    • "HOUR"
    • "MINUTE"
    • "SECOND"
    • "MILLISECOND"
    • "MICROSECOND"
    • "NANOSECOND"

NOW

  • NOW()

    Computes a timestamp equal to the current system time

LOCALTIMESTAMP

  • LOCALTIMESTAMP()

    Equivalent to NOW

CURDATE

  • CURDATE()

    Computes a timestamp equal to the current system time, excluding the time information

CURRENT_DATE

  • CURRENT_DATE()

    Equivalent to CURDATE

EXTRACT

  • EXTRACT(TimeUnit from timestamp_val)

    Extracts the specified TimeUnit from the supplied date.

    Allowed TimeUnits are:

    • MICROSECOND
    • MINUTE
    • HOUR
    • DAY (Day of Month)
    • DOY (Day of Year)
    • DOW (Day of week)
    • WEEK
    • MONTH
    • QUARTER
    • YEAR

    TimeUnits are not case sensitive.

MICROSECOND

  • MICROSECOND(timestamp_val)

    Equivalent to EXTRACT(MICROSECOND from timestamp_val)

SECOND

  • SECOND(timestamp_val)

    Equivalent to EXTRACT(SECOND from timestamp_val)

MINUTE

  • MINUTE(timestamp_val)

    Equivalent to EXTRACT(MINUTE from timestamp_val)

HOUR

  • HOUR(timestamp_val)

    Equivalent to EXTRACT(HOUR from timestamp_val)

WEEK

  • WEEK(timestamp_val)

    Equivalent to EXTRACT(WEEK from timestamp_val)

WEEKOFYEAR

  • WEEKOFYEAR(timestamp_val)

    Equivalent to EXTRACT(WEEK from timestamp_val)

MONTH

  • MONTH(timestamp_val)

    Equivalent to EXTRACT(MONTH from timestamp_val)

QUARTER

  • QUARTER(timestamp_val)

    Equivalent to EXTRACT(QUARTER from timestamp_val)

YEAR

  • YEAR(timestamp_val)

    Equivalent to EXTRACT(YEAR from timestamp_val)

WEEKISO

  • WEEKISO(timestamp_val)

    Computes the ISO week for the provided timestamp value.

YEAROFWEEKISO

  • YEAROFWEEKISO(timestamp_val)

    Computes the ISO year for the provided timestamp value.

MAKEDATE

  • MAKEDATE(integer_years_val, integer_days_val)

    Computes a timestamp value that is the specified number of days after the specified year.

DAYNAME

  • DAYNAME(timestamp_val)

    Computes the string name of the day of the timestamp value.

MONTHNAME

  • MONTHNAME(timestamp_val)

    Computes the string name of the month of the timestamp value.

TO_DAYS

  • TO_DAYS(timestamp_val)

    Computes the difference in days between the input timestamp, and year 0 of the Gregorian calendar

TO_SECONDS

  • TO_SECONDS(timestamp_val)

    Computes the number of seconds since year 0 of the Gregorian calendar

FROM_DAYS

  • FROM_DAYS(n)

    Returns a timestamp values that is n days after year 0 of the Gregorian calendar

UNIX_TIMESTAMP

  • UNIX_TIMESTAMP()

    Computes the number of seconds since the unix epoch

FROM_UNIXTIME

  • FROM_UNIXTIME(n)

    Returns a Timestamp value that is n seconds after the unix epoch

ADDDATE

  • ADDDATE(timestamp_val, interval)

    Same as DATE_ADD

SUBDATE

  • SUBDATE(timestamp_val, interval)

    Same as DATE_SUB

TIMESTAMPDIFF

  • TIMESTAMPDIFF(unit, timestamp_val1, timestamp_val2)

    Returns timestamp_val1 - timestamp_val2 rounded down to the provided unit.

WEEKDAY

  • WEEKDAY(timestamp_val)

    Returns the weekday number for timestamp_val.

    Note

    Monday = 0, Sunday=6

YEARWEEK

  • YEARWEEK(timestamp_val)

    Returns the year and week number for the provided timestamp_val concatenated as a single number. For example:

    YEARWEEK(TIMESTAMP '2021-08-30::00:00:00')
    202135
    

LAST_DAY

  • LAST_DAY(timestamp_val)

    Given a timestamp value, returns a timestamp value that is the last day in the same month as timestamp_val.

UTC_TIMESTAMP

  • UTC_TIMESTAMP()

    Returns the current UTC date and time as a timestamp value.

UTC_DATE

  • UTC_DATE()

    Returns the current UTC date as a Timestamp value.

TO_DATE

  • TO_DATE(col_expr)

    Casts the col_expr to a timestamp column truncated to the date portion.

String Functions

Bodo SQL currently supports the following string functions:

LOWER

  • LOWER(str)

    Converts the string scalar/column to lower case.

LCASE

  • LCASE(str)

    Same as LOWER.

UPPER

  • UPPER(str)

    Converts the string scalar/column to upper case.

UCASE

  • UCASE(str)

    Same as UPPER.

CONCAT

  • CONCAT(str_0, str_1, ...)

    Concatenates the strings together. Requires at least two arguments.

CONCAT_WS

  • CONCAT_WS(str_separator, str_0, str_1, ...)

    Concatenates the strings together, with the specified separator. Requires at least three arguments

SUBSTRING

  • SUBSTRING(str, start_index, len)

    Takes a substring of the specified string, starting at the specified index, of the specified length. start_index = 1 specifies the first character of the string, start_index = -1 specifies the last character of the string. start_index = 0 causes the function to return empty string. If start_index is positive and greater then the length of the string, returns an empty string. If start_index is negative, and has an absolute value greater then the length of the string, the behavior is equivalent to start_index = 1.

    For example:

    SUBSTRING('hello world', 1, 5) =='hello'
    SUBSTRING('hello world', -5, 7) =='world'
    SUBSTRING('hello world', -20, 8) =='hello wo'
    SUBSTRING('hello world', 0, 10) ==''
    

    MID

    • MID(str, start_index, len)

    Equivalent to SUBSTRING

SUBSTR

  • SUBSTR(str, start_index, len)

    Equivalent to SUBSTRING

LEFT

  • LEFT(str, n)

    Takes a substring of the specified string consisting of the leftmost n characters

  • RIGHT(str, n)

    Takes a substring of the specified string consisting of the rightmost n characters

REPEAT

  • REPEAT(str, len)

    Extends the specified string to the specified length by repeating the string. Will truncate the string If the string's length is less then the len argument

    For example:

    REPEAT('abc', 7) =='abcabca'
    REPEAT('hello world', 5) =='hello'
    

STRCMP

  • STRCMP(str1, str2)

    Compares the two strings lexicographically. If str1 > str2, return 1. If str1 < str2, returns -1. If str1 == str2, returns 0.

REVERSE

  • REVERSE(str)

    Returns the reversed string.

ORD

  • ORD(str)

    Returns the integer value of the unicode representation of the first character of the input string. returns 0 when passed the empty string

CHAR

  • CHAR(int)

    Returns the character of the corresponding unicode value. Currently only supported for ASCII characters (0 to 127, inclusive)

SPACE

  • SPACE(int)

    Returns a string containing the specified number of spaces.

LTRIM

  • LTRIM(str)

    returns the input string, will remove all spaces from the left of the string

RTRIM

  • RTRIM(str)

    returns the input string, will remove all spaces from the right of the string

TRIM

  • TRIM(str)

    returns the input string, will remove all spaces from the left and right of the string

SUBSTRING_INDEX

  • SUBSTRING_INDEX(str, delimiter_str, n)

    Returns a substring of the input string, which contains all characters that occur before n occurrences of the delimiter string. if n is negative, it will return all characters that occur after the last n occurrences of the delimiter string. If num_occurrences is 0, it will return the empty string

    For example:

    SUBSTRING_INDEX('1,2,3,4,5', ',', 2) =='1,2'
    SUBSTRING_INDEX('1,2,3,4,5', ',', -2) =='4,5'
    SUBSTRING_INDEX('1,2,3,4,5', ',', 0) ==''
    

LPAD

  • LPAD(string, len, padstring)

    Extends the input string to the specified length, by appending copies of the padstring to the left of the string. If the input string's length is less then the len argument, it will truncate the input string.

    For example:

    LPAD('hello', 10, 'abc') =='abcabhello'
    LPAD('hello', 1, 'abc') =='h'
    

RPAD

  • RPAD(string, len, padstring)

    Extends the input string to the specified length, by appending copies of the padstring to the right of the string. If the input string's length is less then the len argument, it will truncate the input string.

    For example:

    RPAD('hello', 10, 'abc') =='helloabcab'
    RPAD('hello', 1, 'abc') =='h'
    

REPLACE

  • REPLACE(base_string, substring_to_remove, string_to_substitute)

    Replaces all occurrences of the specified substring with the substitute string.

    For example:

    REPLACE('hello world', 'hello' 'hi') =='hi world'
    

LENGTH

  • LENGTH(string)

    Returns the number of characters in the given string.

EDITDISTANCE

  • EDITDISTANCE(string0, string1[, max_distance])

    Returns the minimum edit distance between string0 and string1 according to Levenshtein distance. Optionally accepts a third argument specifying a maximum distance value. If the minimum edit distance between the two strings exceeds this value, then this value is returned instead. If it is negative, zero is returned.

SPLIT_PART

  • SPLIT_PART(source, delimeter, part)

    Returns the substring of the source between certain occurrence of the delimeter string, the occurrence being specified by the part. I.e. if part=1, returns the substring before the first occurrence, and if part=2, returns the substring between the first and second occurrence. Zero is treated like 1. Negative indicies are allowed. If the delimeter is empty, the source is treated like a single token. If the part is out of bounds, '' is returned.

#### STRTOK - STRTOK(source[, delimeter[, part]])

Tokenizes the source string by occurrences of any character in the
delimeter string and returns the occurrence specified by the part.
I.e. if part=1, returns the substring before the first occurrence,
and if part=2, returns the substring between the first and second
occurrence. Zero and negative indices are not allowed. Empty tokens
are always skipped in favor of the next non-empty token. In any
case where the only possible output is '', the output is `NULL`.
The delimeter is optional and defaults to ' '. The part is optional
and defaults to 1.

Control Flow Functions

DECODE

  • DECODE(Arg0, Arg1, Arg2, ...)

    When Arg0 is Arg1, outputs Arg2. When Arg0 is Arg3, outputs Arg4. Repeats until it runs out of pairs of arguments. At this point, if there is one remaining argument, this is used as a default value. If not, then the output is NULL. Note: treats NULL as a literal value that can be matched on.

    Therefore, the following:

    DECODE(A, NULL, 0, 'x', 1, 'y', 2, 'z', 3, -1)
    

    Is logically equivalent to:

    CASE WHEN A IS NULL THEN 0
         WHEN A = 'x' THEN 1
         WHEN A = 'y' THEN 2
         WHEN A = 'z' THEN 3
         ELSE -1 END
    

EQUAL_NULL

  • EQUAL_NULL(A, B)

    Returns true if A and B are both NULL, or both non-null and equal to each other.

IF

  • IF(Cond, TrueValue, FalseValue)

    Returns TrueValue if cond is true, and FalseValue if cond is false. Logically equivalent to:

    CASE WHEN Cond THEN TrueValue ELSE FalseValue END
    

IFNULL

  • IFNULL(Arg0, Arg1)

    Returns Arg1 if Arg0 is null, and otherwise returns Arg1. If arguments do not have the same type, Bodo SQL will attempt to cast them all to a common type, which is currently undefined behavior.

NVL

  • NVL(Arg0, Arg1)

    Equivalent to IFNULL

NULLIF

  • NULLIF(Arg0, Arg1)

    Returns null if the Arg0 evaluates to true, and otherwise returns Arg1

COALESCE

  • COALESCE(A, B, C, ...)

    Returns the first non NULL argument, or NULL if no non NULL argument is found. Requires at least two arguments. If Arguments do not have the same type, Bodo SQL will attempt

COUNT

  • COUNT(*)

    Compute the number of entries in a window.

SUM

  • SUM(COLUMN_EXPRESSION)

    Compute the sum over the window or NULL if the window is empty.

AVG

  • AVG(COLUMN_EXPRESSION)

    Compute the average over the window or NULL if the window is empty.

STDDEV

  • STDDEV(COLUMN_EXPRESSION)

    Compute the standard deviation for a sample over the window or NULL if the window is empty.

STDDEV_POP

  • STDDEV_POP(COLUMN_EXPRESSION)

    Compute the standard deviation for a population over the window or NULL if the window is empty.

VARIANCE

  • VARIANCE(COLUMN_EXPRESSION)

    Compute the variance for a sample over the window or NULL if the window is empty.

VAR_POP

  • VAR_POP(COLUMN_EXPRESSION)

    Compute the variance for a population over the window or NULL if the window is empty.

Window Functions

Window functions can be used to compute an aggregation across a row and its surrounding rows. Most window functions have the following syntax:

SELECT WINDOW_FN(ARG1, ..., ARGN) OVER (PARTITION BY PARTITION_COLUMN_1, ..., PARTITION_COLUMN_N ORDER BY SORT_COLUMN_1, ..., SORT_COLUMN_N ROWS BETWEEN <LOWER_BOUND AND <UPPER_BOUND>) FROM table_name
The ROWS BETWEEN ROWS BETWEEN <LOWER_BOUNDAND <UPPER_BOUND> section is used to specify the window over which to compute the function. A bound can can come before the current row, using PRECEDING or after the current row, using FOLLOWING. The bounds can be relative (i.e. N PRECEDING) or they can be absolute using the UNBOUNDED keyword. These bounds are inclusive.

For example:

SELECT SUM(A) OVER (PARTITION BY B ORDER BY C ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) FROM table1
Computes a sum for each row of the current row, the row preceding, and the row following. In contrast:
SELECT SUM(A) OVER (PARTITION BY B ORDER BY C ROWS BETWEEN UNBOUNDED PRECEDING AND 0 FOLLOWING) FROM table1
Computes the cumulative sum because the window always starts at the first row and grows by 1 for each subsequent row.

Window functions execute by performing a series of steps which influences the final output.

  1. Partition by the PARTITION_COLUMN. This effectively performs a groupby on the provided PARTITION_COLUMN.
  2. Sort each group according to the Order By clause.
  3. Apply the function over the "window" given by the window.
  4. Shuffle the data back to the original ordering.

For BodoSQL, PARTITION BY and ORDER BY are required, but ROWS BETWEEN is optional. If ROWS BETWEEN is not specified then it defaults to computing the result over the entire window. Currently BodoSQL supports the following Window functions:

MAX

  • MAX(COLUMN_EXPRESSION)

    Compute the maximum value over the window or NULL if the window is empty.

MIN

  • MIN(COLUMN_EXPRESSION)

    Compute the minimum value over the window or NULL if the window is empty.

COUNT

  • COUNT(COLUMN_EXPRESSION)

    Compute the number of non-NULL entries in a window.

LEAD

  • LEAD(COLUMN_EXPRESSION, N)

    Returns the row that follows the current row by N. If there are fewer than N rows the follow the current row in the window, it returns NULL. N must be a literal non-negative integer.

    This function cannot be used with ROWS BETWEEN.

LAG

  • LAG(COLUMN_EXPRESSION, N)

    Returns the row that precedes the current row by N. If there are fewer than N rows the precede the current row in the window, it returns NULL. N must be a literal non-negative integer.

    This function cannot be used with ROWS BETWEEN.

FIRST_VALUE

  • FIRST_VALUE(COLUMN_EXPRESSION)

    Select the first value in the window or NULL if the window is empty.

LAST_VALUE

  • LAST_VALUE(COLUMN_EXPRESSION)

    Select the last value in the window or NULL if the window is empty.

NTH_VALUE

  • NTH_VALUE(COLUMN_EXPRESSION, N)

    Select the Nth value in the window (1-indexed) or NULL if the window is empty. If N is greater or than the window size, this returns NULL.

NTILE

  • NTILE(N)

    Divides the partitioned groups into N buckets based on ordering. For example if N=3 and there are 30 rows in a partition, the first 10 are assigned 1, the next 10 are assigned 2, and the final 10 are assigned 3.

RANK

  • RANK()

    Compute the rank of each row based on the value(s) in the row relative to all value(s) within the partition. The rank begins with 1 and increments by one for each succeeding value. Duplicate value(s) will produce the same rank, producing gaps in the rank (compare with `DENSE_RANK).

DENSE_RANK

  • DENSE_RANK()

    Compute the rank of each row based on the value(s) in the row relative to all value(s) within the partition without producing gaps in the rank (compare with RANK). The rank begins with 1 and increments by one for each succeeding value. Rows with the same value(s) produce the same rank.

Note

To compare RANK and DENSE_RANK, on input array ['a', 'b', 'b', 'c'], RANK will output [1, 2, 2, 4] while DENSE_RANK outputs [1, 2, 2, 3].

PERCENT_RANK

  • PERCENT_RANK()

    Compute the percentage ranking of the value(s) in each row based on the value(s) relative to all value(s) within the window partition. Ranking calcuated using RANK() divided by the number of rows in the window partition minus one. Paritions with one row have PERCENT_RANK() of 0.

CUME_DIST

  • CUME_DIST()

    Compute the cumulative distribution of the value(s) in each row based on the value(s) relative to all value(s) within the window partition.

ROW_NUMBER

  • ROW_NUMBER()

    Compute an increasing row number (starting at 1) for each row. This function cannot be used with ROWS BETWEEN.to cast them to a common datatype, which is currently undefined behavior.

Supported DataFrame Data Types

BodoSQL uses Pandas DataFrames to represent SQL tables in memory and converts SQL types to corresponding Python types which are used by Bodo. Below is a table mapping SQL types used in BodoSQL to their respective Python types and Bodo data types.

SQL Type(s) Equivalent Python Type Bodo Data Type
TINYINT np.int8 bodo.int8
SMALLINT np.int16 bodo.int16
INT np.int32 bodo.int32
BIGINT np.int64 bodo.int64
FLOAT np.float32 bodo.float32
DECIMAL, DOUBLE np.float64 bodo.float64
VARCHAR, CHAR str bodo.string_type
TIMESTAMP, DATE np.datetime64[ns] bodo.datetime64ns
INTERVAL(day-time) np.timedelta64[ns] bodo.timedelta64ns
BOOLEAN np.bool_ bodo.bool_

BodoSQL can also process DataFrames that contain Categorical or Date columns. However, Bodo will convert these columns to one of the supported types, which incurs a performance cost. We recommend restricting your DataFrames to the directly supported types when possible.

Nullable and Unsigned Types

Although SQL does not explicitly support unsigned types, by default, BodoSQL maintains the exact types of the existing DataFrames registered in a [BodoSQLContext], including unsigned and non-nullable type behavior. If an operation has the possibility of creating null values or requires casting data, BodoSQL will convert the input of that operation to a nullable, signed version of the type.

Supported Literals

BodoSQL supports the following literal types:

  • boolean_literal
  • datetime_literal
  • float_literal
  • integer_literal
  • interval_literal
  • string_literal

Boolean Literal

Syntax:

TRUE | FALSE

Boolean literals are case insensitive.

Datetime Literal

Syntax:

DATE 'yyyy-mm-dd' |
TIMESTAMP 'yyyy-mm-dd' |
TIMESTAMP 'yyyy-mm-dd HH:mm:ss'

Float Literal

Syntax:

[ + | - ] { digit [ ... ] . [ digit [ ... ] ] | . digit [ ... ] }

where digit is any numeral from 0 to 9

Integer Literal

Syntax:

[ + | - ] digit [ ... ]

where digit is any numeral from 0 to 9

Interval Literal

Syntax:

INTERVAL integer_literal interval_type

Where integer_literal is a valid integer literal and interval type is one of:

DAY[S] | HOUR[S] | MINUTE[S] | SECOND[S]

In addition we also have limited support for YEAR[S] and MONTH[S]. These literals cannot be stored in columns and currently are only supported for operations involving add and sub.

String Literal

Syntax:

'char [ ... ]'

Where char is a character literal in a Python string.

NULL Semantics

Bodo SQL converts SQL queries to Pandas code that executes inside Bodo. As a result, NULL behavior aligns with Pandas and may be slightly different than other SQL systems. This is currently an area of active development to ensure compatibility with other SQL systems.

Most operators with a NULL input return NULL. However, there a couple notable places where Bodo SQL may not match other SQL systems:

  • Bodo SQL treats NaN the same as NULL
  • Is (NOT) False and Is (NOT) True return NULL when used on a null expression
  • AND will return NULL if any of the inputs is NULL

BodoSQL Caching & Parameterized Queries

BodoSQL can reuse Bodo caching to avoid recompilation when used inside a JIT function. BodoSQL caching works the same as Bodo, so for example:

@bodo.jit(cache=True)
def f(filename):
    df1 = pd.read_parquet(filename)
    bc = bodosql.BodoSQLContext({"table1": df1})
    df2 = bc.sql("SELECT A FROM table1 WHERE B > 4")
    print(df2.A.sum())

This will avoid recompilation so long as the DataFrame scheme stored in filename has the same schema and the code does not change.

To enable caching for queries with scalar parameters that you may want to adjust between runs, we introduce a feature called parameterized queries. In a parameterized query, the SQL query replaces a constant/scalar value with a variable, which we call a named parameter. In addition, the query is passed a dictionary of parameters which maps each name to a corresponding Python variable.

For example, if in the above SQL query we wanted to replace 4 with other integers, we could rewrite our query as:

bc.sql("SELECT A FROM table1 WHERE B @var", {"var": python_var})

Now anywhere that @var is used, the value of python_var at runtime will be used instead. This can be used in caching, because python_var can be provided as an argument to the JIT function itself, thus enabling changing the filter without recompiling. The full example looks like this:

@bodo.jit(cache=True)
def f(filename, python_var):
    df1 = pd.read_parquet(filename)
    bc = bodosql.BodoSQLContext({"table1": df1})
    df2 = bc.sql("SELECT A FROM table1 WHERE B @var", {"var": python_var})
    print(df2.A.sum())

Named parameters cannot be used in places that require a constant value to generate the correct implementation (e.g. TimeUnit in EXTRACT).

IO Handling

BodoSQL is great for compute based SQL queries, but you cannot yet access external storage directly from SQL. Instead, you can load and store data using Bodo and various Python APIs. Here we explain a couple common methods for loading data.

Pandas IO in JIT function with SQL Query

The most common way to load data is to first use Pandas APIs to load a DataFrame inside a JIT function and then to use that DataFrame inside a BodoSQLContext.

def f(f1, f2):
    df1 = pd.read_parquet(f1)
    df2 = pd.read_parquet(f2)
    bc = bodosql.BodoSQLContext(
        {
            "t1": df1,
            "t2": df2,
        }
    )
    return bc.sql("select t1.A, t2.B from t1, t2 where t1.C > 5 and t1.D = t2.D")

Pandas IO in a JIT Function Separate from Query

The previous approach works well for most individual queries. However, when running several queries on the same dataset, it should ideally be loaded once for all queries. To do this, you can structure your JIT code to contain a single load function at the beginning. For example:

@bodo.jit
def load_data(f1, f2):
    df1 = pd.read_parquet(f1)
    df2 = pd.read_parquet(f2)
    return df1, df2

def q1(df1, df2):
    bc = bodosql.BodoSQLContext(
        {
            "t1": df1,
            "t2": df2,
        }
    )
    return bc.sql("select t1.A, t2.B from t1, t2 where t1.C > 5 and t1.D = t2.D")

...

@bodo.jit
def run_queries(f1, f2):
    df1, df2 = load_data(f1, f2)
    print(q1(df1, df2))
    print(q2(df2))
    print(q3(df1))
    ...

run_queries(f1, f2)

This approach prevents certain optimizations, such as filter pushdown. However, the assumption here is that you will use the entire DataFrame across the various benchmarks, so no optimization is useful by itself. In addition, any optimizations that can apply to all queries can be done explicitly inside load_data. For example, if all queries are operate on a single day's data with df1, you can write that filter in load_data to limit IO and filter pushdown will be performed.

@bodo.jit
def load_data(f1, f2, target_date):
    df1 = pd.read_parquet(f1)
    # Applying this filter limits how much data is loaded.
    df1 = df1[df1.date_val == target_date]
    df2 = pd.read_parquet(f2)
    return df1, df2

@bodo.jit
def run_queries(f1, f2, target_date):
    df1, df2 = load_data(f1, f2, target_date)
    ...

run_queries(f1, f2, target_date)

TablePath API

The TablePath API a general purpose IO interface to specify IO sources. It is meant as an alternative to loading the exact tables that you are using inside your JIT function. The TablePath API indicates how to find a table if needed. BodoSQL will then load that table when needed within a single query. For example, here is some sample code that loads two DataFrames from parquet using the TablePath API.

@bodo.jit
def f(f1, f2):
    bc = bodosql.BodoSQLContext(
        {
            "t1": bodosql.TablePath(f1, "parquet"),
            "t2": bodosql.TablePath(f2, "parquet"),
        }
    )
    return bc.sql("select t1.A, t2.B from t1, t2 where t1.C > 5 and t1.D = t2.D")

Here, the TablePath constructor doesn't load any data (even when used in regular Python). Instead, a BodoSQLContext will generate code to load that data once it sees the contents are used inside the query. This allows you to declare your tables once without worrying about memory usage resulting from loading an entire dataset. For example:

t1 = bodosql.TablePath(f1, "parquet")
t2 = bodosql.TablePath(f2, "parquet")

@bodo.jit
def f1(t1):
    bc = bodosql.BodoSQLContext(
        {
            "t1": t1
        }
    )
    ...


@bodo.jit
def f2(t2):
    bc = bodosql.BodoSQLContext(
        {
            "t2": t2,
        }
    )
    ...

Additionally, TablePath provides the best filter pushdown support on individual BodoSQL queries. Note that it loads all used tables from IO on every query. If you reuse the same table and columns across multiple queries, you should consider loading the DataFrame once in a separate JIT function.

API Reference

  • bodosql.TablePath(file_path, file_type, *, conn_str=None, reorder_io=None)

    Specifies how a DataFrame should be loaded from IO by a BodoSQL query. This can only load data when used with a BodoSQLContext constructor.

    Arguments

    • file_path: Path to IO file or name of the table for SQL. This must constant at compile time if used inside JIT.

    • file_type: Type of file to load as a string. Supported values are "parquet" and "sql". This must constant at compile time if used inside JIT.

    • conn_str: Connection string used to connect to a SQL DataBase, equivalent to the conn argument to pandas.read_sql. This must be constant at compile time if used inside JIT and must be None if not loading from a SQL DataBase.

  • reorder_io: Boolean flag determining when to load IO. If False, all used tables are loaded before executing any of the query. If True, tables are loaded just before first use inside the query, which often results in decreased peak memory usage as each table is partially processed before loading the next table. The default value, None, behaves like True, but this may change in the future. This must be constant at compile time if used inside JIT.

Back to top