6.1. Synopsis¶

SELECT { <expression> | * } [, ... ]

  FROM { <table_name> | <find_expression> } [, ... ]

  [ <asof_type> ASOF JOIN { <table_name> | <find_expression> } [, ... ] [ PREWHERE <condition> ]]

  [ ASOF JOIN <range_generator> [ PREWHERE <condition> ] ]

  [ IN { RANGE <range_spec> | '[' RANGE <range_spec> [, ...] ']' }
    [ WITH MONTHS IN ( <month>, <month> ) ]
    [ WITH DAYS IN ( <day>, <day> ) ]
    [ WITH TIME IN ( <time>, <time> ) ] ]

  [ WHERE <condition> ]

  [ GROUP BY <group> [, ... ] ]

  [ PIVOT <column> ]

  [ HAVING <condition> ]

  [ ORDER BY { <expression> [ { ASC | DESC } ] } [, ... ] ]

  [ LIMIT <limit> ]

  [ OFFSET <offset> ]

asof_type ::= FULL | LEFT | RIGHT

expression ::=
  <column_name>
  | <function> ( <column_name> )
  | ( <sub_query> )

range_spec ::=
  ( <timestamp>, <timestamp> )
  ( <timestamp>, <time_offset> )

range_generator ::=
  ( <timestamp>, <timestamp>, <time_offset> )
  ( <timestamp>, <time_offset>, <time_offset> )

group ::=
  <expression>
  | <duration>

find_expression ::=
  FIND ( { <tag_expression> | NOT <tag_expression> } [ AND ... ] )

tag_expression ::=
  TAG = 'tag_name'

sub_query ::=
  SELECT <function> ( <column_name> ) FROM <table_name> IN RANGE <sub_range_spec>

sub_range_spec ::=
  ( <timestamp> | <column_name>, <timestamp> | <column_name> )
  ( <timestamp> | <column_name>, <time_offset> )

6.2. Description¶

SELECT retrieves rows from one or more tables. A SELECT statement can perform calculations on rows prior to returning the result which are performed server-side and distributed over the entire cluster.

6.3. Parameters¶

table_name

The name of the table to retrieve rows from.

find_expression

When your tables are tagged, you can use inline key/value lookups to perform your query over multiple tables. To match all tables that have the tag “stocks”, you can use FIND(tag='stocks' AND type=ts).

asof_type

The type of ASOF JOIN to perform, it can be FULL, LEFT, or RIGHT.

column_name

A column name to read data from. Must be part of the table’s schema or will throw an error otherwise.

function

An aggregate function to apply over. Can only be used in combination with GROUP BY. For valid functions, please refer to the function reference.

condition

A WHERE or HAVING condition is any expression that evaluates to a boolean. Any row that does not match this predicate will be filtered from the results. For an overview of valid operators, please refer to the comparison operators reference. The WHERE condition is applied before the SELECT, whereas the HAVING condition is applied on the SELECT results.

range_generator

An expression to generate range intervals.

limit

Accepts a non-negative integer to limit the maximum number of rows returned by the query. Typical use cases include pagination and returning the top-X results.

offset

Accepts a non-negative integer to skip rows of the result. Most common use cases include pagination in combination with limit.

timestamp

An absolute timestamp. This can be either a date or a date + time. Supports precision for days, seconds or nanoseconds. For more information, please refer to the documentation for timestamps.

time_offset

A relative offset, can only be used in combination with an absolute_timestamp. For more information, please refer to the documentation for timestamps.

month

A month. Can be one of jan, feb, mar, apr, may, jun, jul, aug, sep, oct, nov or dec. When defining a subrange using WITH MONTHS IN, is both left and right inclusive: WITH MONTHS IN (jan, apr) will match January to April.

day

A weekday. Can be one of sun, mon, tue, wed, thu, fri or sat. When defining a subrange using WITH DAYS IN, is both left and right inclusive: WITH DAYS IN (mon, wed) will match Monday to Wednesday.

time

Time of day, having a precision of either minutes, seconds or nanoseconds. When defining a subrange using WITH TIME IN, is left inclusive and right exclusive: WITH TIME IN (09:22:00, 09:22:01) will not match a row whose time is exactly 09:22:01 AM.

duration

A time duration to group aggregates by. Valid values are any durations as specified in the documentation for durations.

6.4. Examples¶

Select all columns from all rows:

SELECT * FROM example

Get everything between January 1st 2007 and January 1st 2008 (left inclusive) for the time series “stocks.apple”:

SELECT * FROM stocks.apple IN RANGE(2007, 2008)

Get everything between November 3rd 2017, 20:01:22 and December 2nd, 2017, 06:20:10 (left inclusive) for the time series “stocks.apple”:

SELECT * FROM stocks.apple IN RANGE(2017-11-03T20:01:22, 2017-12-02T06:20:10)

Get the first 10 days of 2007 for “stocks.apple”:

SELECT * FROM stocks.apple IN RANGE(2007, +10d)

Get the last second of 2016 for “stocks.apple”:

SELECT * FROM stocks.apple IN RANGE(2017, -1s)

Get the close and open values that are greater than 3 of “stocks.apple” for the first 10 days of 2016:

SELECT close, open FROM stocks.apple IN RANGE(2016, +10d) WHERE (close > 3) AND (open > 3)

Get the last close value for March 28th 2016:

SELECT LAST(close) FROM stocks.apple IN RANGE(2016-03-28, +1d)

6.5. Advanced examples¶

Get the open and close value when volume is greater than 0 for the first 5 days of 2016 and 2017 for “stocks.apple”:

SELECT open, close FROM stocks.apple IN [range(2016, +5d), range(2017, +5d)] WHERE volume > 0

Get the hourly arithmetic mean of volume exchanged for all nasdaq stocks for yesterday:

SELECT arithmetic_mean(volume) FROM find(tag='nasdaq' AND type=ts) IN RANGE(yesterday, +1d) GROUP BY hour

Get the sum of volumes for every Friday of January 2008 between 16:00 and 17:00 for “stocks.apple”:

SELECT sum(volume) FROM stocks.apple IN RANGE(2008, +month) WITH DAYS IN (fri, fri) WITH TIME IN (16:00, 17:00)

Get the daily open, high, low, close, volume for “stocks.apple” for the last 30 days:

SELECT first(open), max(high), min(low), last(close), sum(volume) FROM stocks.apple IN RANGE(today, -30d) GROUP BY day

Get the sum of volume and the number of lines for the last hour by 10 seconds group:

SELECT sum(volume), count(volume) FROM stocks.apple IN RANGE(now, -1h) GROUP BY 10s

Get the sum of volumes for “stocks.apple” the year 2008 and 2010, grouped by month (gregorian calendar):

SELECT sum(volume) FROM stocks.apple IN [RANGE(2008, +1y), RANGE(2010, +1y)] GROUP BY month

If we assume we have an additional ‘deal_timestamp’ column on our timeseries, we apply a filter on it:

SELECT sum(volume) FROM stocks.apple IN RANGE(now, -1h) WHERE deal_timestamp=datetime(2009-11-23T09:30)

6.6. IN FIND example¶

Example: tagging stock tables by exchange:

CREATE TABLE stocks (symbol INT64, open DOUBLE, close DOUBLE, high DOUBLE, low DOUBLE, volume INT64)

timestamp                      symbol   open             close            high             low              volume
-----------------------------------------------------------------------------------------------------------------------------
2018-01-01T00:00:00.000000000Z 1        10.240000        10.170000        10.360000        10.070000        1027389
2018-01-01T00:00:00.000000000Z 2        902.750000       909.510000       925.000000       890.120000       8437
2018-01-01T00:00:00.000000000Z 3        45.770000        49.750000        53.000000        42.820000        9820

And with the following int keys (added through the key/value API):

Key	Value	Tags
‘symbol.apple’	1	‘nasdaq’
‘symbol.facebook’	2	‘nasdaq’
‘symbol.tesco’	3	‘lse’

The following query:

select * from stocks where symbol in find(tag='nasdaq')

Will output:

timestamp                      symbol   open             close            high             low              volume
-----------------------------------------------------------------------------------------------------------------------------
2018-01-01T00:00:00.000000000Z 1        10.240000        10.170000        10.360000        10.070000        1027389
2018-01-01T00:00:00.000000000Z 2        902.750000       909.510000       925.000000       890.120000       8437

6.7. SUBQUERY example¶

A subquery performs an aggregation for each resulting row of the outer query.

Assuming these tables

parent

$timestamp	contract_end
2021-01-01T00:00:00	2021-01-02T00:00:00
2021-01-02T00:00:00	2021-01-04T00:00:00

child

$timestamp	c
2021-01-01T00:00:00	1
2021-01-02T00:00:00	2
2021-01-03T00:00:00	3
2021-01-04T00:00:00	4
2021-01-05T00:00:00	5

The following query:

SELECT $timestamp, contract_end, (SELECT SUM(c) FROM child IN RANGE (parent.$timestamp, parent.contract_end)) AS sum FROM parent;

Will output:

timestamp                                        contract_end              sum
-------------------------------------------------------------------------------
2021-01-01T00:00:00.000000000Z 2021-01-02T00:00:00.000000000Z                1
2021-01-02T00:00:00.000000000Z 2021-01-04T00:00:00.000000000Z                5

6.8. ASOF joins¶

An ASOF joins several tables based on their timestamps. ASOF joins use the last value for the merge, based in the range of the query. If values exist before the range of the select, they will be ignored.

There are four types of ASOF joins:

• Left: The left table will be used as a reference. The timestamps for the right table will be using the last value for each timestamps of all columns of the left table.

• Right: The right table will be used as a reference. The timestamps for the left table will be using the last value for each timestamps of all columns of the right table.

• Full: Both table will be used as references.

• Range: Joins the table on the left against a generated range.

Currently, ASOF joins support joining against exactly one table. However, the number of tables to join is not limited. This means, for example, that if you are doing a LEFT ASOF JOIN, only one table may be specified on the left, but several can be specified on the right.

This is correct:

SELECT * FROM table_left LEFT ASOF JOIN table_right1, table_right2;

This is not supported:

SELECT * FROM table_left1, table_left2 LEFT ASOF JOIN table_right;

It is possible to join the result of a find. However, the result of the FIND must yield exactly one value for the table to join against. The number of results for the tables being joined is not limited.

This is correct if FIND(tag='left_tag') returns exactly one table, and FIND(tag='right_tag') returns at least one table:

SELECT * FROM FIND(tag='left_tag') LEFT ASOF JOIN FIND(tag='right_tag');

6.9. PIVOT¶

A PIVOT creates a new table whose columns are selected timestamps and the product (pivot column value) X (selected columns). It’s often combined with ASOF, to treat pivot each column value as a table and align their content to the ASOF reference table. PIVOT queries supports lookup for pivot and pivoted columns. Usual filters (PREWHERE, WHERE and HAVING) and GROUP BY clauses are supported.