GROUP BY

Group the SELECT results by the values of the specified columns or expressions. GROUP BY is often combined with aggregate functions (COUNT, MAX, MIN, SUM, AVG) to perform calculations in each group.

If GROUP BY is present in the query, then when selecting columns (between SELECT ... FROM), you can use the following constructs:

  1. Columns by which grouping is performed (included in the GROUP BY argument).
  2. Aggregate functions (see the next section). Columns that are not used for grouping can only be included as arguments for an aggregate function.
  3. Functions that return the start and end times of the current window (HOP_START and HOP_END).
  4. Arbitrary calculations combining items 1–3.

You can group by the result of an arbitrary expression computed from the source columns. In this case, to access the result of this expression, we recommend assigning a name to it using AS. See the second example.

Syntax

SELECT                             -- In SELECT, you can use:
    column1,                       -- key columns specified in GROUP BY
    key_n,                         -- named expressions specified in GROUP BY
    column1 + key_n,               -- arbitrary non-aggregate functions on them
    Aggr_Func1( column2 ),         -- aggregate functions containing any columns as arguments,
    Aggr_Func2( key_n + column2 ), -- including named expressions specified in GROUP BY
    ...
FROM table
GROUP BY
    column1, column2, ...,
    <expr> AS key_n           -- When grouping by expression, you can set a name for it using AS,
                              -- and use it in SELECT

The query in the format SELECT * FROM table GROUP BY k1, k2, ... returns all columns listed in GROUP BY, i.e., is equivalent to SELECT DISTINCT k1, k2, ... FROM table.

An asterisk can also be used as an argument for the COUNT aggregate function. COUNT(*) means "the count of rows in the group".

Note

Aggregate functions ignore NULL in their arguments, except for COUNT.

YQL also provides aggregation factories implemented by the functions AGGREGATION_FACTORY and AGGREGATE_BY.

Examples

SELECT key, COUNT(*) FROM my_table
GROUP BY key;

SELECT double_key, COUNT(*) FROM my_table
GROUP BY key + key AS double_key;

SELECT
   double_key,                           -- OK: A key column
   COUNT(*) AS group_size,               -- OK: COUNT(*)
   SUM(key + subkey) AS sum1,            -- OK: An aggregate function
   CAST(SUM(1 + 2) AS String) AS sum2,   -- OK: An aggregate function with a constant argument
   SUM(SUM(1) + key) AS sum3,            -- ERROR: Nested aggregations are not allowed
   key AS k1,                            -- ERROR: Using a non-key column named key without aggregation
   key * 2 AS dk1,                       -- ERROR in YQL: using a non-key column named key without aggregation
FROM my_table
GROUP BY
  key * 2 AS double_key,
  subkey as sk,

Warning

Specifying a name for a column or expression in GROUP BY .. AS foo it is an extension on top of YQL. Such a name becomes visible in WHERE despite the fact that filtering by WHERE is executed before the grouping. For example, if the T table includes two columns, foo and bar, then the query SELECT foo FROM T WHERE foo > 0 GROUP BY bar AS foo would actually filter data by the bar column from the source table.

GROUP BY ... SessionWindow()

YQL supports grouping by session. To standard expressions in GROUP BY, you can add a special SessionWindow function:

SELECT
  user,
  session_start,
  SessionStart() AS same_session_start, -- It's same as session_start
  COUNT(*) AS session_size,
  SUM(value) AS sum_over_session,
FROM my_table
GROUP BY user, SessionWindow(<time_expr>, <timeout_expr>) AS session_start

The following happens in this case:

  1. The input table is partitioned by the grouping keys specified in GROUP BY, ignoring SessionWindow (in this case, it's based on user).

    If GROUP BY includes nothing more than SessionWindow, then the input table gets into one partition.

  2. Each partition is split into disjoint subsets of rows (sessions).

    For this, the partition is sorted in the ascending order of the time_expr expression.
    The session limits are drawn between neighboring items of the partition, that differ in their time_expr values by more than timeout_expr.

  3. The sessions obtained in this way are the final partitions on which aggregate functions are calculated.

The SessionWindow() key column (in the example, it's session_start) has the value "the minimum time_expr in the session".
If GROUP BY includes SessionWindow(), you can use a special aggregate function
SessionStart.

An extended version of SessionWindow with four arguments is also supported:

SessionWindow(<order_expr>, <init_lambda>, <update_lambda>, <calculate_lambda>)

Where:

  • <order_expr>: An expression used to sort the source partition
  • <init_lambda>: A lambda function to initialize the state of session calculation. It has the signature (TableRow())->State. It's called once for the first (following the sorting order) element of the source partition
  • <update_lambda>: A lambda function to update the status of session calculation and define the session limits. It has the signature (TableRow(), State)->Tuple<Bool, State>. It's called for every item of the source partition, except the first one. The new value of state is calculated based on the current row of the table and the previous state. If the first item in the return tuple is True, then a new session starts from the current row. The key of the new session is obtained by applying <calculate_lambda> to the second item in the tuple.
  • <calculate_lambda>: A lambda function for calculating the session key (the "value" of SessionWindow() that is also accessible via SessionStart()). The function has the signature (TableRow(), State)->SessionKey. It's called for the first item in the partition (after <init_lambda>) and those items for which <update_lambda> has returned True in the first item in the tuple. Please note that to start a new session, you should make sure that <calculate_lambda> has returned a value different from the previous session key. Sessions having the same keys are not merged. For example, if <calculate_lambda> returns the sequence 0, 1, 0, 1, then there will be four different sessions.

Using the extended version of SessionWindow, you can, for example, do the following: divide a partition into sessions, as in the SessionWindow use case with two arguments, but with the maximum session length limited by a certain constant:

Example

$max_len = 1000; -- is the maximum session length.
$timeout = 100; -- is the timeout (timeout_expr in a simplified version of SessionWindow).

$init = ($row) -> (AsTuple($row.ts, $row.ts)); -- is the session status: tuple from 1) value of the temporary column ts in the session's first line and 2) in the current line
$update = ($row, $state) -> {
  $is_end_session = $row.ts - $state.0 > $max_len OR $row.ts - $state.1 > $timeout;
  $new_state = AsTuple(IF($is_end_session, $row.ts, $state.0), $row.ts);
  return AsTuple($is_end_session, $new_state);
};
$calculate = ($row, $state) -> ($row.ts);
SELECT
  user,
  session_start,
  SessionStart() AS same_session_start, -- It's same as session_start
  COUNT(*) AS session_size,
  SUM(value) AS sum_over_session,
FROM my_table
GROUP BY user, SessionWindow(ts, $init, $update, $calculate) AS session_start

You can use SessionWindow in GROUP BY only once.

ROLLUP, CUBE, and GROUPING SETS

The results of calculating the aggregate function as subtotals for the groups and overall totals over individual columns or whole table.

Syntax

SELECT
    c1, c2, -- the columns to group by

AGGREGATE_FUNCTION(c3) AS outcome_c  -- an aggregate function (SUM, AVG, MIN, MAX, COUNT)

FROM table_name

GROUP BY
    GROUP_BY_EXTENSION(c1, c2)       -- an extension of GROUP BY: ROLLUP, CUBE, or GROUPING SETS
  • ROLLUP groups the column values in the order they are listed in the arguments (strictly from left to right), generates subtotals for each group and the overall total.
  • CUBE groups the values for every possible combination of columns, generates the subtotals for each group and the overall total.
  • GROUPING SETS sets the groups for subtotals.

You can combine ROLLUP, CUBE and GROUPING SETS, separating them by commas.

GROUPING

The values of columns not used in calculations are replaced with NULL in the subtotal. In the overall total, the values of all columns are replaced by NULL. GROUPING: A function that allows you to distinguish the source NULL values from the NULL values added while calculating subtotals and overall totals.

GROUPING returns a bit mask:

  • 0: If NULL is used for the original empty value.
  • 1: If NULL is added for a subtotal or overall total.

Example

SELECT
    column1,
    column2,
    column3,

    CASE GROUPING(
        column1,
        column2,
        column3,
    )
        WHEN 1  THEN "Subtotal: column1 and column2"
        WHEN 3  THEN "Subtotal: column1"
        WHEN 4  THEN "Subtotal: column2 and column3"
        WHEN 6  THEN "Subtotal: column3"
        WHEN 7  THEN "Grand total"
        ELSE         "Individual group"
    END AS subtotal,

    COUNT(*) AS rows_count

FROM my_table

GROUP BY
    ROLLUP(
        column1,
        column2,
        column3
    ),
    GROUPING SETS(
        (column2, column3),
        (column3)
        -- if you add here (column2) as well, then together
        -- the ROLLUP and GROUPING SETS would produce a result
        -- similar to CUBE
    )
;

DISTINCT

Applying aggregate functions only to distinct values of the column.

Note

Applying DISTINCT to calculated values is not currently implemented. For this purpose, you can use a subquery or the expression GROUP BY ... AS ....

Example

SELECT
  key,
  COUNT (DISTINCT value) AS count -- top 3 keys by the number of unique values
FROM my_table
GROUP BY key
ORDER BY count DESC
LIMIT 3;

You can also use DISTINCT to fetch distinct rows using SELECT DISTINCT.

GROUP COMPACT BY

Improves aggregation efficiency if the query author knows in advance that none of aggregation keys finds large amounts of data (i.e., with the order of magnitude exceeding a gigabyte or a million of rows). If this assumption fails to materialize, then the operation may fail with Out of Memory error or start running much slower compared to the non-COMPACT version.

Unlike the usual GROUP BY, the Map-side combiner stage and additional Reduce are disabled for each field with DISTINCT aggregation.

Example

SELECT
  key,
  COUNT (DISTINCT value) AS count -- top 3 keys by the number of unique values
FROM my_table
GROUP COMPACT BY key
ORDER BY count DESC
LIMIT 3;

GROUP BY ... HOP

Group the table by the values of the specified columns or expressions and subsets by time (the time window).

Among the columns used for grouping, make sure to use the HOP construct to define the time window for grouping.

HOP(time_extractor, hop, interval, delay)

The implemented version of the time window is called the hopping window. This is a window that moves forward in discrete intervals (the hop parameter). The total duration of the window is set by the interval parameter. To determine the time of each input event, the time_extractor parameter is used. This expression depends only on the input values of the columns and must have the Timestamp type. It specifies where to extract the time value from data.

The following happens in this case:

  1. The input table is partitioned by the grouping keys specified in GROUP BY, ignoring HOP. If GROUP BY includes nothing more than HOP, then the input table gets into one partition.
  2. Each partition is sorted in ascending order of the expression time_extractor.
  3. Each partition is split into subsets of rows (possibly intersecting), on which aggregate functions are calculated.

In each partition defined by the values of all the grouping columns, the window moves forward independently of other streams. The advancement of the window depends entirely on the latest event in the partition.

The interval and delay parameters must be multiples of the hop parameter. Non-multiple intervals are prohibited in the current implementation.
The interval and hop parameters must be positive.

The delay parameter is ignored in the current implementation because the data in one partition is already sorted.

To set hop, interval, and delay, use a string expression compliant with ISO 8601. This format is used to construct the built-in Interval type from a string.

When selecting columns (between SELECT ... FROM) you can use the HOP_START and HOP_END functions (without parameters), which return a value of Timestamp type, corresponding to the start and end of the current window.

The tumbling window, known in other systems, is a special case of a hopping window where interval == hop.

Examples

SELECT
    key,
    COUNT(*)
FROM my_table
GROUP BY
    HOP(CAST(subkey AS Timestamp), "PT10S", "PT1M", "PT30S"),
    key;
-- hop = 10 seconds
-- interval = 1 minute
-- delay = 30 seconds

SELECT
    double_key,
    HOP_END() as time,
    COUNT(*) as count
FROM my_table
GROUP BY
    key + key AS double_key,
    HOP(ts, "PT1M", "PT1M", "PT1M");

HAVING

Filtering a SELECT based on the calculation results of aggregate functions. The syntax is similar to WHERE.

Example

SELECT
    key
FROM my_table
GROUP BY key
HAVING COUNT(value) > 100;
Previous
FLATTEN
Next
JOIN