12. Select#

12.1. Synopsis#

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

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

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

  [ ASOF JOIN <range_generator> [ <interpolation> ] [ 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> ]

  [ RESTRICT TO <restriction> ]

  [ 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> )

interpolation ::=
  INTERPOLATE (method=<interpolation_method> [, limit_direction=<interpolation_limit_direction>])

interpolation_limit_direction ::=
 FORWARD | BACKWARD | BOTH | NONE

interpolation_method ::=
  CONSTANT | LINEAR | CUBIC | BARYCENTRIC | LOGARITHMIC

lookback ::=
  WITH LOOKBACK <duration>

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

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

restriction ::=
  <function> ( <column_name> )

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> )

12.2. 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.

12.3. 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.

restriction

A RESTRICT TO restriction selects only one row from each set of rows with duplicated $timestamp values. The restriction function can be one of first, last, min, max, abs_min, abs_max. For a description of these functions, please refer to the function reference.

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.

interpolation Specifies the mathematical technique used to estimate values between two known data points.

12.4. Examples#

Select all columns from all rows:

SELECT * FROM example

Note

The examples below assume typical open, high, low, close, volume stocks time series.

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);

Filter based on an exact timestamp rather than a range:

SELECT * FROM stocks.apple WHERE $timestamp = DATETIME(2024-01-01);

12.5. Advanced examples#

Note

The examples below assume typical open, high, low, close, volume stocks time series.

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 1hour

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, +1month) 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 1day

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 1month

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)

12.6. 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

12.7. 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

12.8. JSON operator examples#

QuasarDB supports use of JSON operator on column types STRING and SYMBOL. Using JSON operator allows you to access objects and arrays in JSON strings. All values accessed by JSON operator are returned as string. If the value is not found, or column value can’t be parsed as JSON the result is null. Currently we don’t support GROUP BY on JSON operator.

There are three JSON operators: - ::JSON operator: Interprets column value as JSON. - ->> operator: Access JSON object value. - -> operator: Access JSON object.

Assuming table data with the following data:

$timestamp

raw

2025-02-12T00:00:00

{“key_1”:”value_1”, “key_2”:{“key1”:1, “key2”:[3,4,5]}}

2025-02-13T00:00:00

{“key_1”:”value_2”, “key_2”:{“key1”:6, “key2”:[8,9,10]}}

We can access value of key_1 and key2 using JSON operator with query:

SELECT $timestamp, raw::JSON->>'key_1', raw::JSON->>'key_2' FROM data;

This will output:

$timestamp

raw::JSON->>’key_1’

raw::JSON->>’key_2’

2025-02-12T00:00:00

“value_1”

{“key1”:1, “key2”:[3,4,5]}

2025-02-13T00:00:00

“value_2”

{“key1”:6, “key2”:[8,9,10]}

To access values inside key_2 object, we can use -> operator:

SELECT $timestamp, raw::JSON->'key_2'->>'key1', raw::JSON->'key_2'->>'key2' FROM data;

This query will output:

$timestamp

raw::JSON->’key_2’->>’key1’

raw::JSON->’key_2’->>’key2’

2025-02-12T00:00:00

1

[3,4,5]

2025-02-13T00:00:00

6

[8,9,10]

To access values inside key_2 array, we can use ->> operator, and index of the array:

SELECT $timestamp, raw::JSON->'key_2'->'key2'->>0 FROM data;

This query will output:

$timestamp

raw::JSON->’key_2’->’key2’->>0

2025-02-12T00:00:00

3

2025-02-13T00:00:00

8

You can also use JSON operator in WHERE clauses:

SELECT $timestamp, raw::JSON->'key_2'->'key2'->>0 FROM data where raw::JSON->>'key_1'='value_1';

This query returns the first element of key2 array for rows where key_1 is equal to value_1:

$timestamp

raw::JSON->’key_2’->’key2’->>0

2025-02-12T00:00:00

3

12.9. 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');

12.9.1. Filtering ASOF join results#

The WHERE clause is applied after the ASOF JOIN is executed. What if you want to filter the data before you join the two tables? QuasarDB has a specific PREWHERE clause to solve that problem.

This query will filter out negative values before joining the two tables:

SELECT * FROM table_left LEFT ASOF JOIN table_right PREWHERE table_left.value >= 0

This query will filter out negative values after joining the two tables:

SELECT * FROM table_left LEFT ASOF JOIN table_right WHERE table_left.value >= 0

This query will filter out negative value before joining the two tables and filter out values greater than 10 after joining the two tables:

SELECT * FROM table_left LEFT ASOF JOIN table_right PREWHERE table_left.value >= 0 WHERE table_left.value < 10

12.9.2. Examples#

Assuming these tables

table_left

Timestamp

Pressure

2019-11-23T13:02:01

100

2019-11-23T13:03:03

110

2019-11-23T13:03:59

105

2019-11-23T13:05:00

115

table_right

Timestamp

Temperature

2019-11-23T13:01:58

56

2019-11-23T13:03:03

59

2019-11-23T13:04:02

58

2019-11-23T13:05:02

56

2019-11-23T13:05:22

57

Here are the results for the three possible ASOF joins:

SELECT $timestamp, pressure, temperature FROM table_left LEFT ASOF JOIN table_right;

Timestamp

Pressure

Temperature

2019-11-23T13:02:01

100

56

2019-11-23T13:03:03

110

59

2019-11-23T13:03:59

105

59

2019-11-23T13:05:00

115

58

SELECT $timestamp, pressure, temperature FROM table_left RIGHT ASOF JOIN table_right;

Timestamp

Pressure

Temperature

2019-11-23T13:01:58

(void)

56

2019-11-23T13:03:03

110

59

2019-11-23T13:04:02

105

58

2019-11-23T13:05:02

115

56

2019-11-23T13:05:22

115

57

SELECT $timestamp, pressure, temperature FROM table_left FULL ASOF JOIN table_right;

Timestamp

Pressure

Temperature

2019-11-23T13:01:58

(void)

56

2019-11-23T13:02:01

100

56

2019-11-23T13:03:03

110

59

2019-11-23T13:03:59

105

59

2019-11-23T13:04:02

105

58

2019-11-23T13:05:00

115

58

2019-11-23T13:05:02

115

56

2019-11-23T13:05:22

115

57

SELECT $timestamp, pressure FROM table_left ASOF JOIN RANGE(2019-11-23T13:02:00, +5min, +1min);

Timestamp

Pressure

2019-11-23T13:02:00

(void)

2019-11-23T13:03:00

100

2019-11-23T13:04:00

105

2019-11-23T13:05:00

115

2019-11-23T13:06:00

115

Using interpolation, we could obtain the following results:

SELECT $timestamp, temperature FROM table_left FULL ASOF JOIN table_right INTERPOLATE(method=linear);

Timestamp

Temperature

2019-11-23T13:01:58

56

2019-11-23T13:02:01

56.13846154

2019-11-23T13:03:03

59

2019-11-23T13:03:59

58.05084746

2019-11-23T13:04:02

58

2019-11-23T13:05:00

56.06666667

2019-11-23T13:05:02

56

2019-11-23T13:05:22

57

SELECT $timestamp, pressure FROM %{:1} FULL ASOF JOIN %{:2} INTERPOLATE(method=linear, limit_direction=none);

Timestamp

Pressure

2019-11-23T13:01:58

(void)

2019-11-23T13:02:01

100

2019-11-23T13:03:03

110

2019-11-23T13:03:59

105

2019-11-23T13:04:02

105.4918033

2019-11-23T13:05:00

115

2019-11-23T13:05:02

(void)

2019-11-23T13:05:22

(void)

12.10. 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.

12.10.1. Examples#

Assuming this table

sells

Timestamp

Country

Id

2010-01-01

France

14

2011-01-01

Italy

42

2012-01-01

Italy

30

2013-01-01

France

28

2014-01-01

France

65

A first simple query can be done:

SELECT $timestamp, Country, Id FROM sells PIVOT ON Country ORDER BY Id;

Timestamp

France.Id

Italy.Id

2010-01-01

14

(void)

2013-01-01

28

(void)

2012-01-01

(void)

30

2011-01-01

(void)

42

2014-01-01

65

(void)

Now, assuming these three tables

ref_table

Timestamp

Label

2017-01-01

A

2018-01-01

B

2019-01-01

C

north_sells

Timestamp

Vendor

Benefits

2017-01-01

Bob

10

2018-01-01

Alice

20

2019-01-01

Bob

15

2020-01-01

Alice

40

east_sells

Timestamp

Vendor

Benefits

2017-01-01

Carol

30

2018-01-01

Dave

5

2019-01-01

Dave

10

A PIVOT ASOF query can be done:

SELECT $timestamp, label, vendor, benefits FROM ref_table LEFT ASOF JOIN north_sells, east_sells PIVOT ON vendor;

Timestamp

Label

Alice.benefits

Bob.benefits

Carol.benefits

Dave.benefits

2017-01-01

A

(void)

10

30

(void)

2018-01-01

B

20

10

30

5

2019-01-01

C

20

15

30

10

With a PREWHERE and GROUP BY clause, we can get average benefits on several years of some vendors:

SELECT $timestamp, vendor, avg(benefits) FROM ref_table LEFT ASOF JOIN north_sells, east_sells IN RANGE(2016, +4y) PREWHERE vendor != 'Alice' GROUP BY 2y, vendor PIVOT ON vendor;

Timestamp

arithmetic_mean(Bob.benefits)

arithmetic_mean(Carol.benefits)

arithmetic_mean(Dave.benefits)

2016-01-01

10

30

(void)

2018-01-01

12.5

30

7.5

12.11. Interpolation#

Interpolation is a mathematical technique used to estimate values between two known data points. In QuasarDB, it is primarily employed to make predictions or fill in gaps in time-series data. Interpolation helps smooth time-series visualizations and makes predictions between known data points.

12.11.1. Why Use Interpolation?#

Interpolation is useful when you need to estimate values between known data points. It’s particularly valuable when data changes discretely meaning that the data points are distinct and separate, and there are no values between those points, and you want to approximate values between two points. The choice of interpolation method depends on your understanding of how the underlying data behaves over time. Different interpolation methods are available, including constant, linear, cubic, barycentric, and logarithmic.

It is typically applied to arithmetic values, such as temperature, pressure, or other numeric measurements. For non-numeric columns, the previous row’s value is usually copied. Any null values are subject to interpolation.

12.11.2. Interpolation Methods#

QuasarDB supports various interpolation methods:

  • Constant (default): Takes the previous value.

  • Linear: Estimates values by drawing an imaginary line between two data points.

  • Cubic: Draws a 3rd-degree curve based on up to four data points.

  • Barycentric: Uses all selected points to draw a smoother, higher-degree curve. It can produce “stray” oscillations.

  • Logarithmic: Performs linear interpolation on logarithms of the values, ideal for data changing exponentially with time.

All these methods require at least two data points, except for constant, which requires only one.

So, which method to choose? It depends on the nature of your data and how accurately you know it.

Constant: If you’re okay with values staying constant until you have new data, this is the method for you. No predictions, just consistency.

Linear: Use this when you have a clear idea that your data is steadily changing over time. It’s like connecting the dots.

Cubic and Barycentric: When you want super-smooth curves through your data, choose one of these methods. They’re great for data that has a gentle flow but might have some small “bumps” in it.

Logarithmic: If your data behaves exponentially, this method is your friend. It’s like understanding the exponential growth.

12.11.3. Interpolation Limit Directions#

When using interpolation, you can specify a limit direction to guide how the estimation is made. This direction sets the boundaries for data estimation between known points. Let’s explore these directions more profoundly:

Forward Direction:

In the forward direction, interpolation predicts future values based on the data you already have. It’s like peering ahead to anticipate what might come next. Use this when your objective is to forecast future data points.

Backward Direction:

In contrast, the backward direction examines past data to infer what might have occurred earlier. Choose this when you’re interested in understanding the events that led to the present state of your data. It helps you reconstruct the past based on the available information.

Both Directions:

The “both” direction is a comprehensive approach that involves looking into both the past and the future. This direction is valuable when you want a complete view, covering both past events and future predictions.

None Direction:

When you select “none,” you’re keeping it simple. It strictly relies on the data points you already have and doesn’t make any predictions.

12.11.4. Examples#

Insert some sample data into the left table
INSERT INTO table_left ($timestamp, a) VALUES
 (2019-01-01T00:00, 10),
 (2019-01-01T01:00, 11),
 (2019-01-01T05:00, 15)
Insert some sample data into the right table
INSERT INTO table_right ($timestamp, b) VALUES
 (2019-01-01T00:00, 110),
 (2019-01-01T02:00, 112),
 (2019-01-01T03:00, 113),
 (2019-01-01T04:00, 114)
Linear: It provides a straightforward way to estimate values between two data points assuming a constant rate of change.
SELECT $timestamp, table_left.a, table_right.b FROM table_left LEFT ASOF JOIN table_right INTERPOLATE(method=linear);

$timestamp                       table_left.a       table_right.b
------------------------------------------------------------------
2019-01-01T00:00:00.000000000Z               10               110
2019-01-01T01:00:00.000000000Z               11               111
2019-01-01T05:00:00.000000000Z               15               114

Returned 3 rows in 3,165 us
Scanned 7 points in 3,165 us (2,211 rows/sec)
Logarithmic:
SELECT $timestamp, table_left.a, table_right.b FROM table_left LEFT ASOF JOIN table_right INTERPOLATE(method=logarithmic);

$timestamp                       table_left.a        table_right.b
------------------------------------------------------------------
2019-01-01T00:00:00.000000000Z               10               110
2019-01-01T01:00:00.000000000Z               11           110.995
2019-01-01T05:00:00.000000000Z               15               114

Returned 3 rows in 2,562 us
Scanned 7 points in 2,562 us (2,731 rows/sec)

Let’s dessect the example:

  • Timestamp 2019-01-01T00:00:00.000000000Z:

For “a” (from “table_left”), the value is 10. For “b” (from “table_right”), the value is 110. This is the starting point of your data. At this timestamp, “a” is 10, and “b” is 110.

  • Timestamp 2019-01-01T01:00:00.000000000Z:

For “a” (from “table_left”), the value is interpolated to be 11. For “b” (from “table_right”), the value is interpolated to be approximately 110.995. Here, you can see a non-linear change in the data. While “a” increases by 1, “b” increases only slightly, which is characteristic of logarithmic data changes. The interpolation method accurately captures this non-linear growth.

  • Timestamp 2019-01-01T05:00:00.000000000Z:

For “a” (from “table_left”), the value is interpolated to be 15. For “b” (from “table_right”), the value is 114. By the final timestamp, you observe that “a” has increased to 15, which is consistent with the previous non-linear trend. Meanwhile, “b” has also increased, but the exact value at this point is 114.

The key takeaway here is that the data changes in a non-linear, exponential fashion. As time progresses, small changes in “a” lead to relatively larger changes in “b.” Logarithmic interpolation is well-suited to capture these non-linear relationships in the data.

Cubic:
SELECT $timestamp, table_left.a, table_right.b FROM table_left LEFT ASOF JOIN table_right INTERPOLATE(method=cubic);
$timestamp                       table_left.a   table_right.b

------------------------------------------------------------------
2019-01-01T00:00:00.000000000Z               10               110
2019-01-01T01:00:00.000000000Z               11               111
2019-01-01T05:00:00.000000000Z               15               114

Returned 3 rows in 2,880 us
Scanned 7 points in 2,880 us (2,429 rows/sec)
ASOF JOIN RANGE interpolation within a single table
CREATE TABLE single_table (a DOUBLE)

INSERT INTO single_table ($timestamp, a) VALUES (2019-01-01T01:00, 11), (2019-01-01T05:00, 17);

SELECT * FROM single_table ASOF JOIN RANGE(2019-01-01T00:00:00, +7h, 1h) INTERPOLATE(method=linear);

$timestamp                           $table   single_table.a
-------------------------------------------------------------
2019-01-01T00:00:00.000000000Z $multitables           (void)
2019-01-01T01:00:00.000000000Z $multitables               11
2019-01-01T02:00:00.000000000Z $multitables             12.5
2019-01-01T03:00:00.000000000Z $multitables               14
2019-01-01T04:00:00.000000000Z $multitables             15.5
2019-01-01T05:00:00.000000000Z $multitables               17
2019-01-01T06:00:00.000000000Z $multitables               17

Returned 7 rows in 11,604 us
Scanned 2 points in 11,604 us (172 rows/sec)

Interpolation can work on a single table, and it can be used to estimate values within that table. It’s particularly useful when you want to fill in missing or interpolated values within the same dataset, often due to gaps in the data.

12.11.5. Automatic Handling of Data Types#

In most cases, interpolation methods can only be applied to numeric columns, such as DOUBLE or INT64, while non-numeric columns, like BLOB, STRING, or SYMBOL, are typically not interpolated. For these non-numeric columns, our system applies the “constant” method, which essentially copies the previous available value.

This approach ensures that interpolation is automatically applied to numeric values, without the need for users to specify which columns should be interpolated. For non-numeric columns, the previous value is retained to maintain data consistency.

Example:

Consider the following example with a table named test containing various data types:
CREATE TABLE test (cb BLOB, cd DOUBLE, ci INT64, cs STRING, csy SYMBOL(sym_csy2), ct TIMESTAMP)

INSERT INTO test ($timestamp, cb, cd, ci, cs, csy, ct) VALUES
  (2019-01-01T00:00, 'd', 2.0, 2, 'D', 'ddd', 2001-01-02),
  (2019-01-01T04:00, 'e', 5.0, 20, 'E', 'eee', 2001-01-04),
  (2019-01-01T06:00, 'f', 6.0, 200, 'F', 'fff', 2001-01-07)

SELECT $timestamp, test.* FROM test ASOF JOIN RANGE(2019-01-01T00:00, +5h, 1h) INTERPOLATE(method=linear)

$timestamp                       test.cb          test.cd   test.ci   test.cs   test.csy                        test.ct
------------------------------------------------------------------------------------------------------------------------
2019-01-01T00:00:00.000000000Z         d                2         2         D        ddd 2001-01-02T00:00:00.000000000Z
2019-01-01T01:00:00.000000000Z         d             2.75         7         D        ddd 2001-01-02T00:00:00.000000000Z
2019-01-01T02:00:00.000000000Z         d              3.5        11         D        ddd 2001-01-02T00:00:00.000000000Z
2019-01-01T03:00:00.000000000Z         d             4.25        16         D        ddd 2001-01-02T00:00:00.000000000Z
2019-01-01T04:00:00.000000000Z         e                5        20         E        eee 2001-01-04T00:00:00.000000000Z

In this example, the query interpolates values for all columns within the specified time range. Numeric columns like “cd” and “ci” are interpolated using the “linear” method, while non-numeric columns like “cb,” “cs,” and “csy” receive constant values, copying the previous available value.

This approach ensures that interpolation is performed on all columns automatically, simplifying the experience. There’s no need to specify which columns should be interpolated, making the process more accessible for those who are new to interpolation methods.