Skip to main content

Write SQL queries

Transform data from the CDF staging area into a data model using built-in and custom Spark SQL queries. Select Switch to SQL editor on the Transform data page to create a transformation in Spark SQL. This article describes the queries and explains how you can load data incrementally.

Tip

The SQL editor offers built-in code completion and built-in Spark SQL functions and Cognite custom SQL functions.

Note

Your changes won't be kept if you switch from the SQL editor to the mapping editor.

Read from a CDF staging table

To select data from a CDF staging table, use the syntax mydb.mytable:

select
*
from
database-name.table-name

If your database or table name contains special characters, enclose the name in backticks, for example `my-db`.`my table`.

Avoid schema inference

Transformations infer schemas in the CDF staging table, but this process only uses a subset of all the rows in the table. You can avoid schema inference and write a schema fitted your data.

To avoid schema inference:

  select
*
from
cdf_raw("database-name", "table-name")

This returns data with the schema key:STRING, lastUpdatedTime:TIMESTAMP, columns:STRING, where the columns string contains the JSON value encoded as a string.

Here's an example of how to enforce a user-defined schema:

    select

get_json_object(columns, '$.externalId') AS externalId,

timestamp(get_json_object(columns, '$.timestamp')) AS timestamp,

double(get_json_object(columns, '$.value')) AS value

from
cdf_raw("database-name", "table-name")

Read from other CDF resource types

To select other CDF resource types, use the syntax _cdf.resource_type.

select * from _cdf.events

The supported resource types are:

  • _cdf.events
  • _cdf.assets
  • _cdf.files
  • _cdf.timeseries
  • _cdf.sequences
  • _cdf_sequences.<sequence_externalId>
  • _cdf.datapoints
  • _cdf.stringdatapoints
  • _cdf.labels
  • _cdf.relationships

Load data incrementally

When reading data, your transformation jobs will run much faster and more efficiently if it only has to read the data that has changed since the last time the transformation job ran. This reduces the time to run a transformation, and allows for more frequent runs. One way to achieve this, is to filter on the lastUpdatedTime column to query for the rows that have changed after a specific timestamp. The filter on lastUpdatedTime is pushed to the underlying resource type (if supported) to run the query more efficiently and only return changes.

There are some minor syntax differences between some of the resource types for this filtering, but for example, when reading from staging tables it could look like this:

select * from mydb.mytable where lastUpdatedTime > to_timestamp(123456).

Instead of encoding the timestamp in the query and keeping it up to date every time new data has been processed, you can use the is_new function to do this for you automatically. The function returns true when a row has changed since the last time the transformation was run and false otherwise. This filters out older rows before the results are processed.

The first time you run a transformation using the query below, all the rows of mytable will be processed:

select * from mydb.mytable where is_new("mydb_mytable", lastUpdatedTime)

If the transformation completes successfully, the second run will only process rows that have changed since the first run.

If the transformation fails, is_new filters the same rows the next time the transformation is run. This ensures that there is no data loss in the transformation from source to destination.

note

Incremental load is disabled when previewing query results. That is, is_new will always return true for all rows.

Each is_new filter is identified by a name and can be set to any constant string (for example,"mydb_mytable" in the above example). This allows you to differentiate between multiple calls to is_new in the same query and use is_new to filter on multiple tables. To easily identify the different filters, we recommend that you use the name of the table as the name of the is_new filter. The name is stored with the transformation and must be unique for the specific transformation. If you use the same name in two different transformations, they're stored separately to not interfere with each other.

note

It's not common to use multiple is_new filters in the same query. Instead, it's more likely you'll use is_new on the main resource you're accessing. Then, you can join in different resources with data to improve any new entries from the main table or resource. If you use multiple is_new filters, they are applied to each source separately before any join operations are evaluated. This means that for the join to work correctly in this case, both sources have to be updated at the same time.

Resource types supporting incremental data loading on the lastUpdatedTime column

Incremental data loading is supported by filtering on lastUpdatedTime for the following resource types in addition to staging:

  • _cdf.assets
  • _cdf.events
  • _cdf.files
  • _cdf.timeseries
  • _cdf.relationships
  • _cdf.datasets

Incremental data loading when using Data Modeling

For data modeling, we don't recommend filtering on timestamp or int64 columns. Instead, it is more efficient to use the variant of the is_new function that uses the sync API to read all changes since the last time the transformation was successfully run. This is_new function is used when it references the cdf_nodes(), cdf_edges() or cdf_data_models() functions instead of a single column like lastUpdatedTime.

This could look like this:

select * from cdf_nodes() where is_new('my_nodes')

where is_new will filter on the output of cdf_nodes

If you have multiple sources in the same query, you must specify which source the is_new is referencing. This is done by providing an alias on the source function, like this:

select * from cdf_nodes() a, cdf_edges() b where is_new('a_nodes', a)

Here the query defines an alias for the cdf_nodes() function, and then specifies to apply the is_new filter on this alias. This is different than how is_new is used for other resource types, where the specification is to a specific column in the source.

The source can be any of the cdf_nodes, cdf_edges or cdf_data_models functions, and can reference a specific view, such as:

select * from cdf_data_models('space', 'datamodel_external_id', 'version', 'view') where is_new('datamodel_view')

is_new translates the query to filter on a cursor that tracks all changes. The cursor is updated every time the transformation is successfully run, in the same way as is_new for other resource types. You don't need to explicitly model support for this filtering in your data model, as it is inherently supported in data modeling. You can also combine this with other filters (where clauses), and it will use any matching indexes set up in data modeling to ensure performance of any optional filters.

Warning

When using is_new with data modeling, the transformation must run at least once per three days to ensure it can find the difference between the last run and the new run. If it doesn't run for three days or less, the transformation falls back to read all of the input data.

Backfill

To process all the data even if it hasn't changed since the last transformation, change the name of the is_new filter, for example, by adding a postfix with an incrementing number (e.g. "mydb_mytable_1").

This is especially useful when the logic of the query changes and data that has already been imported needs to be updated accordingly.

Custom SQL functions

In addition to the built-in Spark SQL functions, we also provide a set of custom SQL functions to help you write efficient transformations.

note

When a function expects var_args, it allows a variable number of arguments of any type, including star *.

get_names

  • get_names(var_args): Array[String]

Returns an array of the field names of a struct or row.

Example

select get_names(*) from mydb.mytable
-- Returns the column names of 'mydb.mytable'
select get_names(some_struct.*) from mydb.mytable
-- Returns the field names of 'some_struct'

cast_to_strings

  • cast_to_strings(var_args): Array[String]

Casts the arguments to an array of strings. It handles array, struct and map types by casting it to JSON strings.

Example

select cast_to_strings(*) from mydb.mytable
-- Returns the values of all columns in 'mydb.mytable' as strings

to_metadata

  • to_metadata(var_args): Map[String, String]

Creates metadata compatible type from the arguments. In practice it does map_from_arrays(get_names(var_args), cast_to_strings(var_args)). Use this function when you want to transform your columns or structures into a format that fits the metadata field in CDF.

Example

select to_metadata(*) from mydb.mytable
-- Creates a metadata structure from all the columns found in 'mydb.mytable'

to_metadata_except

  • to_metadata_except(excludeFilter: Array[String], var_args)

Returns a metadata structure (Map[String, String]) where strings found in excludeFilter will exclude keys from var_args.

Use this function when you want to put most, but not all, columns into metadata, for example to_metadata_except(array("someColumnToExclude"), *)

Example

select to_metadata_except(array("myCol"), myCol, testCol) from mydb.mytable
-- Creates a map where myCol is filtered out.
-- The result in this case will be Map("testCol" -> testCol.value.toString)

asset_ids

Attempts to find asset names under the given criteria and return the IDs of the matching assets. Three variations are available.

Attempts to find given assetNames in all assets.

  • asset_ids(assetNames: Array[String]): Array[BigInt]

Attempts to find assetNames in the asset hierarchy with rootAssetName as their root asset.

  • asset_ids(assetNames: Array[String], rootAssetName: String): Array[BigInt]

Attempts to find assetNames that belong to the datasetIds.

  • asset_ids(assetNames: Array[String], datasetIds: Array[Long]): Array[BigInt]

Attempts to find assetNames that belong to the datasetIds under the rootAssetName.

  • asset_ids(assetNames: Array[String], rootAssetName: String, datasetIds: Array[Long]): Array[BigInt]

See Assets for more information about assets in CDF.

important

The entire job will be aborted if asset_ids() did not find any matching assets.

Example

select asset_ids(array("PV10", "PV11"))
select asset_ids(array("PV10", "PV11"), "MyBoat")
select asset_ids(array("PV10", "PV11"), array(254343, 23433, 54343))
select asset_ids(array("PV10", "PV11"), array(dataset_id("pv-254343-ext-id"), 23433, 54343))
select asset_ids(array("PV10", "PV11"), "MyBoat", array(dataset_id("pv-254343-ext-id"), 23433, 54343))

is_new

  • is_new(name: String, version: long)

Returns true if the version provided is higher than the version found with the specified name, based on the last time the transformation was run. version can be any column of dataype long with only incremental values ingested. A popular example is the lastUpdatedTime column.

  • If the transformation completes successfully, the next transformation job only processes rows that have changed since the start of the last successfully completed transformation job.

  • If the transformation fails, is_new processes all rows that have changed since the start of the last successful run. This ensures no data loss in the transformation from source to destination. See also Load data incrementally.

Tip

If you're using more than one occurrence of is_new() in one transformation, we recommend that you use different variable names. This guarantees that subqueries within one transformation don't override the lastUpdatedTime record before the transformation is completed.

Example

select * from mydb.mytable where is_new("mydb_mytable_version", lastUpdatedTime)
-- Returns only rows that have changed since the last successful run

dataset_id

  • dataset_id(externalId: String): BigInt

Attempts to find the id of the given data set by externalId and returns the id if the externalId exists.

Example

select dataset_id("EXAMPLE_DATASET") as dataSetId

cdf_assetSubtree

  • cdf_assetSubtree(externalId: String or id: BigInt): Table[Asset]

Returns an asset subtree under a specific asset in an asset hierarchy, that is, all the child assets for a specific asset in an asset hierarchy are returned.

important

If the total size of subtree exceeds 100,000 assets, an error will be returned.

Example

select * from cdf_assetSubtree('externalId of an asset')
select * from cdf_assetSubtree('id of an asset')

cdf_nodes

  • cdf_nodes(space of the view: String, externalId of the view: String, version of the view: String): Table[Nodes]
  • cdf_nodes(): Table[Nodes]

Returns nodes in the CDF project as a table.

  • cdf_nodes() returns space and externalId of all nodes in the CDF project.
  • cdf_nodes("space of the view: String", "externalId of the view: String"," version of the view: String") returns a table with nodes ingested with view as reference.
    The table contains space and externalId columns and columns for each property in the view.

Example

select * from cdf_nodes('space of the view: String', 'externalId of the view: String', 'version of the view: String')
select * from cdf_nodes()

cdf_edges

  • cdf_edges("space of the view: String", "externalId of the view: String", "version of the view: String"): Table[Edges]
  • cdf_edges(): Table[Edges]

Returns edges in the CDF project as a table.

  • cdf_edges() returns space, externalId, startNode, endNode, and type of all edges in a CDF project.

  • cdf_edges(space of the view: String, externalId of the view: String, version of the view: String) returns a table with edges ingested with view as reference.
    The table contains space, externalId, startNode, endNode, and type columns and columns for each property in the view.

Example

select * from cdf_edges('space of the view: String', 'externalId of the view: String', 'version of the view: String')
select * from cdf_edges()

node_reference

  • node_reference("space: String", "externalId: String"): STRUCT<"space:string", "externalId:string">
  • node_reference("externalId: String"): STRUCT<"space:String", "externalId:String">

To reference a node, you need the space externalId of the node and the node externalId. Typically, you reference a node when writing or filtering edges based on startNode and endNode.

node_reference accepts the single parameter externalId of the node. The target/instance space set at the transformation is used as the space externalId of the node.

important

If you're using node_reference for filtering i.e. in your where clause, you must add the space externalId and the node externalId.

Example

select node_reference('space externalId of a node', 'externalId of a node') as startNode, node_reference('space externalId of a node', 'externalId of a node') as endNode, ... from mydb.mytable
select node_reference('externalId of a node') as startNode, node_reference('externalId of a node') as endNode, ... from mydb.mytable
select * from cdf_edges('space of the view: String', 'externalId of the view: String', 'version of the view: String') where startNode = node_reference('space externalId of a node', 'externalId of a node') or node_reference('space externalId of a node', 'externalId of a node')

type_reference

  • type_reference("space: String", "externalId: String"): STRUCT<"space:String", "externalId:String">
  • type_reference("externalId: String"): STRUCT<"space:String", "externalId:String">

All edges have type. To filter edges based on type, use type_reference and provide the space externalId and the edge type externalId. If you're writing edges with a view reference, you must specify the edge type using type_reference.

type_reference accepts the single parameter externalId of the edge type. The target/instance space set at the transformation is used as the space externalId of the edge type.

important

If you're using type_reference for filtering i.e. in your where clause, you must add the space externalId and the edge type externalId.

Example

select node_reference('space externalId of a node', 'externalId of a node') as startNode, type_reference('space externalId of a node', 'externalId of a node') as endNode, ... from mydb.mytable
select * from cdf_edges('space of the view: String', 'externalId of the view: String', 'version of the view: String') where type = type_reference('space externalId of a node', 'externalId of a node') or type_reference('space externalId of a node', 'externalId of a node')
select * from cdf_edges() where type = type_reference('space externalId of a node', 'externalId of a node') or type_reference('space externalId of a node', 'externalId of a node')

cdf_data_models

  • cdf_data_models("data model space: String", "data model externalId: String", "data model version: String", "type external id: String" ): Table[Nodes]
  • cdf_data_models("data model space: String", "data model externalId: String", "data model version: String", "type external id: String", "property in type containing the relationship: String" ): Table[Edges]

These functions follow the data model UI lingo and make it easy to retrieve the data written to types and relationship.

To retrieve data from a type in your data model, provide the data model's space, externalId, version and the externalId of the type as input parameters to cdf_data_models.

To retrieve data from a relationship in your data model, provide the data model's space, externalId, version,the externalId of the type containing the relationship and the name of the relationship property in the type as input parameters to cdf_data_models.

Example

select * from cdf_data_models('data model space: String', 'data model externalId: String', 'data model version: String', 'type external id: String')
select * from cdf_data_models('data model space: String', 'data model externalId: String', 'data model version: String', 'type external id: String', 'property in type where relationship is defined: String')

Disabled Spark SQL functions

We currently don't support using these Spark SQL functions when you transform data:

xpath

xpath_boolean

xpath_double

xpath_float

xpath_int

xpath_number

xpath_short

xpath_string

xpath_long

java_method

reflect