Name resolution

  • Article

Applies to: check marked yes Databricks SQL check marked yes Databricks Runtime

Name resolution is the process by which identifiers are resolved to specific column-, field-, parameter-, or table-references.

Column, field, parameter, and variable resolution

Identifiers in expressions can be references to any one of the following:

  • Column name based on a view, table, common table expression (CTE), or a column_alias.
  • Field name or map key within a struct or map. Fields and keys can never be unqualified.
  • Parameter name of a SQL User Defined Function.
  • Variable name.
  • A special function such as current_user or current_date which does not require the usage of ().
  • The DEFAULT keyword which is used in the context of INSERT, UPDATE, MERGE, or SET VARIABLE to set a column or variable value to its default.

Name resolution applies the following principles:

  • The closest matching reference wins, and
  • Columns and parameter win over fields and keys.

In detail, resolution of identifiers to a specific reference follows these rules in order:

  1. Local references

    1. Column reference

      Match the identifier, which may be qualified, to a column name in a table reference of the FROM clause.

      If there is more than one such match, raise an AMBIGUOUS_COLUMN_OR_FIELD error.

    2. Parameterless function reference

      If the identifier is unqualified and matches current_user, current_date, or current_timestamp: Resolve it as one of these functions.

    3. Column DEFAULT specification

      If the identifier is unqualified, matches default and makes up the entire expression in the context of an UPDATE SET, INSERT VALUES, or MERGE WHEN [NOT] MATCHED: Resolve as the respective DEFAULT value of the target table of the INSERT, UPDATE or MERGE.

    4. Struct field or map key reference

      If the identifier is qualified, then endeavor to match it to a field or map key according to the following steps:

      A. Remove the last identifier and treat it as a field or key. B. Match the remainder to a column in table reference of the FROM clause.

      If there is more than one such match, raise an AMBIGUOUS_COLUMN_OR_FIELD error.

      If there is a match and the column is a:

      • STRUCT: Match the field.

        If the field cannot be matched, raise a FIELD_NOT_FOUND error.

        If there is more than one field, raise a AMBIGUOUS_COLUMN_OR_FIELD error.

      • MAP: Raise an error if the key is qualified.

        A runtime eror may occur if the key is not actually present in the map.

      • Any other type: Raise an error. C. Repeat the preceding step to remove the trailing identifier as a field. Apply rules (A) and (B) while there is an identifier left to interpret as a column.

  2. Lateral column aliasing

    Applies to: check marked yes Databricks SQL check marked yes Databricks Runtime 12.2 LTS and above

    If the expression is within a SELECT list, match the leading identifier to a preceding column alias in that SELECT list.

    If there is more than one such match, raise an AMBIGUOUS_LATERAL_COLUMN_ALIAS error.

    Match each remaining identifier as a field or a map key, and raise FIELD_NOT_FOUND or AMBIGUOUS_COLUMN_OR_FIELD error if they cannot be matched.

  3. Correlation

    • LATERAL

      If the query is preceded by a LATERAL keyword, apply rules 1.a and 1.d considering the table references in the FROM containing the query and preceding the LATERAL.

    • Regular

      If the query is a scalar subquery, IN, or EXISTS subquery apply rules 1.a, 1.d, and 2 considering the table references in the containing query's FROM clause.

  4. Nested correlation

    Re-apply rule 3 iterating over the nesting levels of the query.

  5. Routine parameters

    If the expression is part of a CREATE FUNCTION statement:

    1. Match the identifier to a parameter name. If the identifier is qualified, the qualifier must match the name of the function.
    2. If the identifier is qualified, match to a field or map key of a parameter following rule 1.c

  6. Variables

    1. Match the identifier to a variable name. If the identifier is qualified, the qualifier must be session or system.session.
    2. If the identifier is qualified, match to a field or map key of a variable following rule 1.c

Limitations

To prevent execution of potentially expensive correlated queries, Azure Databricks limits supported correlation to one level. This restriction also applies to parameter references in SQL functions.

Examples

-- Differentiating columns and fields
> SELECT a FROM VALUES(1) AS t(a);
 1

> SELECT t.a FROM VALUES(1) AS t(a);
 1

> SELECT t.a FROM VALUES(named_struct('a', 1)) AS t(t);
 1

-- A column takes precendece over a field
> SELECT t.a FROM VALUES(named_struct('a', 1), 2) AS t(t, a);
 2

-- Implict lateral column alias
> SELECT c1 AS a, a + c1 FROM VALUES(2) AS T(c1);
 2  4

-- A local column reference takes precedence, over a lateral column alias
> SELECT c1 AS a, a + c1 FROM VALUES(2, 3) AS T(c1, a);
 2  5

-- A scalar subquery correlation to S.c3
> SELECT (SELECT c1 FROM VALUES(1, 2) AS t(c1, c2)
           WHERE t.c2 * 2 = c3)
    FROM VALUES(4) AS s(c3);
 1

-- A local reference takes precedence over correlation
> SELECT (SELECT c1 FROM VALUES(1, 2, 2) AS t(c1, c2, c3)
           WHERE t.c2 * 2 = c3)
    FROM VALUES(4) AS s(c3);
  NULL

-- An explicit scalar subquery correlation to s.c3
> SELECT (SELECT c1 FROM VALUES(1, 2, 2) AS t(c1, c2, c3)
           WHERE t.c2 * 2 = s.c3)
    FROM VALUES(4) AS s(c3);
 1

-- Correlation from an EXISTS predicate to t.c2
> SELECT c1 FROM VALUES(1, 2) AS T(c1, c2)
    WHERE EXISTS(SELECT 1 FROM VALUES(2) AS S(c2)
                  WHERE S.c2 = T.c2);
 1

-- Attempt a lateral correlation to t.c2
> SELECT c1, c2, c3
    FROM VALUES(1, 2) AS t(c1, c2),
         (SELECT c3 FROM VALUES(3, 4) AS s(c3, c4)
           WHERE c4 = c2 * 2);
 [UNRESOLVED_COLUMN] `c2`

-- Successsful usage of lateral correlation with keyword LATERAL
> SELECT c1, c2, c3
    FROM VALUES(1, 2) AS t(c1, c2),
         LATERAL(SELECT c3 FROM VALUES(3, 4) AS s(c3, c4)
                  WHERE c4 = c2 * 2);
 1  2  3

-- Referencing a parameter of a SQL function
> CREATE OR REPLACE TEMPORARY FUNCTION func(a INT) RETURNS INT
    RETURN (SELECT c1 FROM VALUES(1) AS T(c1) WHERE c1 = a);
> SELECT func(1), func(2);
 1  NULL

-- A column takes precedence over a parameter
> CREATE OR REPLACE TEMPORARY FUNCTION func(a INT) RETURNS INT
    RETURN (SELECT a FROM VALUES(1) AS T(a) WHERE t.a = a);
> SELECT func(1), func(2);
 1  1

-- Qualify the parameter with the function name
> CREATE OR REPLACE TEMPORARY FUNCTION func(a INT) RETURNS INT
    RETURN (SELECT a FROM VALUES(1) AS T(a) WHERE t.a = func.a);
> SELECT func(1), func(2);
 1  NULL

-- Lateral alias takes precedence over correlated reference
> SELECT (SELECT c2 FROM (SELECT 1 AS c1, c1 AS c2) WHERE c2 > 5)
    FROM VALUES(6) AS t(c1)
  NULL

-- Lateral alias takes precedence over function parameters
> CREATE OR REPLACE TEMPORARY FUNCTION func(x INT)
    RETURNS TABLE (a INT, b INT, c DOUBLE)
    RETURN SELECT x + 1 AS x, x
> SELECT * FROM func(1)
  2 2

-- All together now
> CREATE OR REPLACE TEMPORARY VIEW lat(a, b) AS VALUES('lat.a', 'lat.b');

> CREATE OR REPLACE TEMPORARY VIEW frm(a) AS VALUES('frm.a');

> CREATE OR REPLACE TEMPORARY FUNCTION func(a INT, b int, c int)
  RETURNS TABLE
  RETURN SELECT t.*
    FROM lat,
         LATERAL(SELECT a, b, c
                   FROM frm) AS t;

> VALUES func('func.a', 'func.b', 'func.c');
  a      b      c
  ----- -----  ------
  frm.a  lat.b  func.c

Table and view resolution

An identifier in table-reference can be any one of the following:

  • Persistent table or view in Unity Catalog or the Hive Metastore
  • Common table expression (CTE)
  • Temporary view

Resolution of an identifier depends on whether it is qualified:

  • Qualified

    If the identifier is fully qualified with three parts: catalog.schema.relation, it is unique.

    If the identifier consists of two parts: schema.relation, it is further qualified with the result of SELECT current_catalog() to make it unique.

  • Unqualified

    1. Common table expression

      If the reference is within the scope of a WITH clause, match the identifier to a CTE starting with the immediately containing WITH clause and moving outwards from there.

    2. Temporary view

      Match the identifier to any temporary view defined within the current session.

    3. Persisted table

      Fully qualify the identifier by pre-pending the result of SELECT current_catalog() and SELECT current_schema() and look it up as a persistent relation.

If the relation cannot be resolved to any table, view, or CTE, Databricks raises a TABLE_OR_VIEW_NOT_FOUND error.

Examples

-- Setting up a scenario
> USE CATALOG spark_catalog;
> USE SCHEMA default;

> CREATE TABLE rel(c1 int);
> INSERT INTO rel VALUES(1);

-- An fully qualified reference to rel:
> SELECT c1 FROM spark_catalog.default.rel;
 1

-- A partially qualified reference to rel:
> SELECT c1 FROM default.rel;
 1

-- An unqualified reference to rel:
> SELECT c1 FROM rel;
 1

-- Add a temporary view with a conflicting name:
> CREATE TEMPORARY VIEW rel(c1) AS VALUES(2);

-- For unqualified references the temporary view takes precedence over the persisted table:
> SELECT c1 FROM rel;
 2

-- Temporary views cannot be qualified, so qualifiecation resolved to the table:
> SELECT c1 FROM default.rel;
 1

-- An unqualified reference to a common table expression wins even over a temporary view:
> WITH rel(c1) AS (VALUES(3))
    SELECT * FROM rel;
 3

-- If CTEs are nested, the match nearest to the table reference takes precedence.
> WITH rel(c1) AS (VALUES(3))
    (WITH rel(c1) AS (VALUES(4))
      SELECT * FROM rel);
  4

-- To resolve the table instead of the CTE, qualify it:
> WITH rel(c1) AS (VALUES(3))
    (WITH rel(c1) AS (VALUES(4))
      SELECT * FROM default.rel);
  1

-- For a CTE to be visible it must contain the query
> SELECT * FROM (WITH cte(c1) AS (VALUES(1))
                   SELECT 1),
                cte;
  [TABLE_OR_VIEW_NOT_FOUND] The table or view `cte` cannot be found.

Function resolution

A function reference is recognized by the mandatory trailing set of parentheses.

It can resolve to:

  • A builtin function provided by Azure Databricks,
  • A temporary user defined function scoped to the current session, or
  • A persistent user defined function stored in the hive metastore or Unity Catalog.

Resolution of a function name depends on whether it is qualified:

  • Qualified

    If the name is fully qualified with three parts: catalog.schema.function, it is unique.

    If the name consists of two parts: schema.function, it is further qualified with the result of SELECT current_catalog() to make it unique.

    The function is then looked up in the catalog.

  • Unqualified

    For unqualified function names Azure Databricks follows a fixed order of precedence (PATH):

    1. Builtin function

      If a function by this name exists among the set of built-in functions, that function is chosen.

    2. Temporary function

      If a function by this name exists among the set of temporary functions, that function is chosen.

    3. Persisted function

      Fully qualify the function name by pre-pending the result of SELECT current_catalog() and SELECT current_schema() and look it up as a persistent function.

If the function cannot be resolved Azure Databricks raises an UNRESOLVED_ROUTINE error.

Examples

> USE CATALOG spark_catalog;
> USE SCHEMA default;

-- Create a function with the same name as a builtin
> CREATE FUNCTION concat(a STRING, b STRING) RETURNS STRING
    RETURN b || a;

-- unqualified reference resolves to the builtin CONCAT
> SELECT concat('hello', 'world');
 helloworld

-- Qualified reference resolves to the persistent function
> SELECT default.concat('hello', 'world');
 worldhello

-- Create a persistent function
> CREATE FUNCTION func(a INT, b INT) RETURNS INT
    RETURN a + b;

-- The persistent function is resolved without qualifying it
> SELECT func(4, 2);
 6

-- Create a conflicting temporary function
> CREATE FUNCTION func(a INT, b INT) RETURNS INT
    RETURN a / b;

-- The temporary function takes precedent
> SELECT func(4, 2);
 2

-- To resolve the persistent function it now needs qualification
> SELECT spark_catalog.default.func(4, 3);
 6