Use SQL with Table Sources

Use SQL expressions to manage your source data.

Limited Availability

Overview

Sources are the building blocks for data in Visier. They are the raw materials to map your data to Visier's analytic model.

Visier supports the following source types:

• File sources: File-based containers that use regex and override behavior to manage data. For more information about acceptable files, see Data File Guidelines.
• Table sources: Row-based containers that support incremental updates and SQL transformations during sync. For more information, see Use SQL with Table Sources.

The following table describes the differences between the behavior of each source type.

You can create table sources:

• Using data connectors. For more information, see Data In Connectors.
• With Visier APIs. For more information, see Table Source API.
• Manually. For more information, see Manually create a table source.

Difference from file sources

You can use Spark SQL to query, upsert, replace, commit, rollback, and process data. SQL expressions tell Visier how to map and replace data. File sources use Visier Extraction Language (VEL) and Override behavior.

A table source temporarily stores data while you commit the data to Visier, then discards the files. Table sources are better for near real-time data processing because they do not track individual files indefinitely and can efficiently handle updates and deletions. As a result, the volume of data to process is much lower.

About transactions

To bring data into a table source, start a transaction. A transaction is a temporary container that stores data and SQL actions, such as inserting or replacing data in the table source. An open transaction is a write transaction or a read transaction.

• Write transaction: The transaction makes changes to a table source. One write transaction can be open at a time. Close a transaction by committing the transaction and processing the data. In an API call, specify tablesToModify to lock tables for writing during the transaction.
• Read transaction: The transaction can view data within a table source. While a SQL job processes changes to a table source, a read transaction will only see the data available prior to the job. In an API call, omit tablesToModify to make the transaction read-only.

About input files

A data file that you upload to a table source transaction is an input file. An input file can contain data to insert into your sources, IDs to match against to change or delete rows, or any other input for your SQL query.

SQL expressions reference input files in the transaction by the id field (set in the job's inputs array), not the filename.

Add data to a table source

Prerequisites: Connect your source data to objects in Visier. For more information, see Add a Mapping. If mapping by formula, use SQL expressions instead of VEL.

Use APIs to add data

1. Start a transaction: POST /v1alpha/data/table-sources/transactions. For more information and code samples, see Start a transaction.
2. Upload a file to the transaction: PUT /v1alpha/data/table-sources/transactions/{transactionId}/inputs/{filename}. For more information and code samples, see Upload an input file to a transaction.
3. Apply SQL expressions and process the data: POST /v1alpha/data/table-sources/transactions/{transactionId}/jobs. For more information and code samples, see Run a SQL job within a transaction.

Manually add data

1. On the global navigation bar, click Sources.
2. Select the table source and click Records.
3. In Temporary Data, click Add Files.
4. Select the files to upload.
5. In SQL Commands, type the SQL expressions to apply to the source. For an example of potential SQL expressions, see SQL examples.
6. Click Execute SQL. The expression moves to Current Transaction.
7. Optional: To remove SQL expressions, in Current Transaction, click Rollback.
8. When finished, click Commit.
9. To process the data, run a job. For more information, see Run a Job.

Define column settings

Assign each column a data type and optionally designate one column as the primary timestamp.

Use APIs to define columns

Define the columns array in any of the following API calls:

1. Create a table source: POST /v1alpha/data/table-sources/tables. For more information and code samples, see Create a table source.
2. Update a table source: PUT /v1alpha/data/table-sources/tables/{tableId}. For more information and code samples, see Update a table source.

Manually define columns

When you access a table source in the Sources room, you can create columns, assign column data types, and choose the primary timestamp. For more information, see Manually create a table source.

Edit key-value pairs

View and change the user-defined key-value pairs for a table source. In addition to key-value pairs, a table source's Table Properties tab displays the source's number of records, last updated date, and data size.

Use APIs to edit key-value pairs

1. Retrieve a table source's existing key-value pairs: GET /v1alpha/data/table-sources/tables/{tableId}/metadata. In the response, look for metadata. For more information and code samples, see Retrieve a table source's metadata.
2. Run a SQL job with the setMetadata action. POST /v1alpha/data/table-sources/transactions/{transactionId}/jobs. For more information and code samples, see setMetadata.

Manually edit key-value pairs

1. On the global navigation bar, click Sources.
2. Select the table source and click Table Properties.
3. In Key/Value Metadata, do any of the following:
   1. To add a new key-value pair, click Add Pair. Type a key and value.
   2. To change an existing key-value pair, type a new key or value.
   3. To delete a key-value pair, click the Delete button .
4. To revert your changes without saving, click Reset.
5. To save your changes, click Commit. This executes the changes to the table source.

SQL examples

This section provides sample SQL expressions to use in table source queries and mapping formulas. For more information about SQL functions, see Apache Spark SQL Functions.

Table source mapping formulas have protective limitations. They cannot use functions that change row counts or depend on table order. For a list of SQL functions that are not supported, see Unsupported SQL functions.

Samples in this section use the following table.

-- employee_custom_data
-- Columns: employee_id (STRING), badge_number (STRING), office_floor (INTEGER),
--          is_remote (BOOLEAN), effective_date (DATE)

SELECT

1
2
SELECT employee_id, badge_number, office_floor, effective_date
FROM employee_custom_data

1
2
3
4
SELECT employee_id, badge_number, office_floor, effective_date
FROM employee_custom_data
WHERE effective_date >= '2026-01-01'
  AND is_remote = false

1
2
3
4
5
6
7
SELECT office_floor,
       COUNT(*)                                         AS headcount,
       SUM(CASE WHEN is_remote THEN 1 ELSE 0 END)      AS remote_count
FROM employee_custom_data
WHERE effective_date = '2026-01-01'
GROUP BY office_floor
ORDER BY office_floor

1
2
3
4
5
6
7
8
9
SELECT e.employee_id,
       e.badge_number,
       n.office_floor,
       n.is_remote,
       n.effective_date
FROM new_employees n
LEFT JOIN employee_custom_data e
    ON e.employee_id = n.employee_id
   AND e.effective_date = n.effective_date

1
2
3
4
5
6
7
SELECT employee_id, badge_number, office_floor, is_remote, effective_date
FROM (
    SELECT *,
           ROW_NUMBER() OVER (PARTITION BY employee_id ORDER BY effective_date DESC) AS rn
    FROM employee_custom_data
) ranked
WHERE rn = 1

INSERT

1
2
3
INSERT INTO employee_custom_data
SELECT employee_id, badge_number, office_floor, is_remote, effective_date
FROM new_employees

1
2
INSERT INTO employee_custom_data (employee_id, badge_number, office_floor, is_remote, effective_date)
VALUES ('E001', 'B-1042', 3, false, '2026-03-01')

1
2
3
4
5
6
7
8
9
INSERT INTO employee_custom_data
SELECT n.employee_id, n.badge_number, n.office_floor, n.is_remote, n.effective_date
FROM new_employees n
WHERE NOT EXISTS (
    SELECT 1
    FROM employee_custom_data e
    WHERE e.employee_id   = n.employee_id
      AND e.effective_date = n.effective_date
)

UPDATE

1
2
3
UPDATE employee_custom_data
SET is_remote = true
WHERE office_floor = 0

1
2
3
4
5
6
7
8
MERGE INTO employee_custom_data AS target
USING new_employees AS source
    ON target.employee_id    = source.employee_id
   AND target.effective_date = source.effective_date
WHEN MATCHED THEN
    UPDATE SET
        office_floor = source.office_floor,
        is_remote    = source.is_remote

DELETE

1
2
DELETE FROM employee_custom_data
WHERE effective_date = '2025-12-31'

1
2
3
4
DELETE FROM employee_custom_data
WHERE employee_id NOT IN (
    SELECT employee_id FROM new_employees
)

1
2
3
4
5
6
DELETE FROM employee_custom_data
WHERE (employee_id, effective_date) IN (
    SELECT employee_id, effective_date
    FROM new_employees
    WHERE is_deleted = true
)

The above example processes a CDC feed where the input file includes an is_deleted column. This removes target file rows that have a matching key in the source file and are flagged for deletion.

MERGE

1
2
3
4
5
6
7
8
9
10
11
12
MERGE INTO employee_custom_data AS target
USING new_employees AS source
    ON target.employee_id   = source.employee_id
   AND target.effective_date = source.effective_date
WHEN MATCHED THEN
    UPDATE SET
        badge_number = source.badge_number,
        office_floor = source.office_floor,
        is_remote    = source.is_remote
WHEN NOT MATCHED THEN
    INSERT (employee_id, badge_number, office_floor, is_remote, effective_date)
    VALUES (source.employee_id, source.badge_number, source.office_floor, source.is_remote, source.effective_date)

A single MERGE statement can handle updates and inserts but cannot delete rows that are absent from the source. To implement a full sync, run a MERGE for the upsert followed by a separate DELETE for rows no longer present in the source. These can be submitted as two separate query actions in the same job.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
// Step 1: Upsert from the source file
MERGE INTO employee_custom_data AS target
USING new_employees AS source
    ON target.employee_id   = source.employee_id
   AND target.effective_date = source.effective_date
WHEN MATCHED THEN
    UPDATE SET
        badge_number = source.badge_number,
        office_floor = source.office_floor,
        is_remote    = source.is_remote
WHEN NOT MATCHED THEN
    INSERT (employee_id, badge_number, office_floor, is_remote, effective_date)
    VALUES (source.employee_id, source.badge_number, source.office_floor, source.is_remote, source.effective_date)

// Step 2: Delete rows that are no longer in the source file
DELETE FROM employee_custom_data
WHERE employee_id NOT IN (SELECT employee_id FROM new_employees)

The following example processes a CDC feed where the input includes an op column indicating the operation type:

• 'I' for insert.
• 'U' for update.
• 'D' for delete.

This removes rows matched by key that are flagged for deletion, updates other matched rows are updated, and inserts unmatched rows that are not deletions.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
MERGE INTO employee_custom_data AS target
USING new_employees AS source
    ON target.employee_id   = source.employee_id
   AND target.effective_date = source.effective_date
WHEN MATCHED AND source.op = 'D' THEN
    DELETE
WHEN MATCHED AND source.op != 'D' THEN
    UPDATE SET
        badge_number = source.badge_number,
        office_floor = source.office_floor,
        is_remote    = source.is_remote
WHEN NOT MATCHED AND source.op != 'D' THEN
    INSERT (employee_id, badge_number, office_floor, is_remote, effective_date)
    VALUES (source.employee_id, source.badge_number, source.office_floor, source.is_remote, source.effective_date)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
MERGE INTO employee_custom_data AS target
USING new_employees AS source
    ON target.employee_id   = source.employee_id
   AND target.effective_date = source.effective_date
WHEN MATCHED AND (
    target.office_floor <> source.office_floor OR
    target.is_remote    <> source.is_remote
) THEN
    UPDATE SET
        office_floor = source.office_floor,
        is_remote    = source.is_remote
WHEN NOT MATCHED THEN
    INSERT (employee_id, badge_number, office_floor, is_remote, effective_date)
    VALUES (source.employee_id, source.badge_number, source.office_floor, source.is_remote, source.effective_date)

Spark SQL for column mappings

Use these expressions in the SELECT list of a query to transform or derive column values before writing them to a table source.

String functions

1
UPPER(employee_id)                                  AS employee_id

1
TRIM(badge_number)                                  AS badge_number

1
CONCAT(first_name, ' ', last_name)                  AS full_name

CONCAT_WS(' ', first_name, last_name)

1
SUBSTR(employee_id, 1, 3)                           AS id_prefix

1
REGEXP_REPLACE(badge_number, '[^A-Za-z0-9]', '')    AS badge_clean

1
SPLIT(location_code, '-')[0]                        AS region_code

1
COALESCE(department_name, 'Unknown')                AS department_name

1
COALESCE(department_name, 'Unknown')                AS department_name

Date and time functions

1
TO_DATE(hire_date_str, 'yyyy-MM-dd')                AS hire_date

1
DATE_FORMAT(effective_date, 'yyyy-MM')              AS year_month

1
DATE_ADD(effective_date, 90)                        AS review_date

1
DATEDIFF(end_date, start_date)                      AS tenure_days

1
2
3
YEAR(effective_date)                                AS year
MONTH(effective_date)                               AS month
DAYOFWEEK(effective_date)                           AS day_of_week

1
CURRENT_DATE()                                      AS snapshot_date

1
DATE_TRUNC('month', effective_date)                 AS month_start

Conditional expressions

1
IF(is_remote, 'Remote', 'On-site')                  AS work_location

1
2
3
4
5
CASE
    WHEN office_floor >= 10 THEN 'Executive'
    WHEN office_floor >= 5  THEN 'Senior'
    ELSE                         'Standard'
END                                                 AS floor_tier

1
NULLIF(badge_number, 'N/A')                         AS badge_number

1
COALESCE(preferred_name, legal_name, employee_id)   AS display_name

Numeric functions

1
ROUND(salary / 12.0, 2)                             AS monthly_salary

1
ABS(variance)                                       AS abs_variance

1
MOD(employee_seq, 10)                               AS shard_key

1
CAST(headcount_str AS INTEGER)                      AS headcount

1
CASE WHEN total > 0 THEN ROUND(part / total * 100, 1) ELSE 0.0 END AS pct

Type casting

1
CAST('2026-01-01' AS DATE)                          AS start_date

1
CAST(office_floor AS STRING)                        AS floor_label

1
CAST(is_remote_flag AS BOOLEAN)                     AS is_remote

1
CAST(updated_at AS DATE)                            AS updated_date

Hashing and anonymization

1
SHA2(employee_id, 256)                              AS employee_id_hash

1
MD5(CONCAT(employee_id, effective_date))            AS row_fingerprint

Unsupported SQL functions

Table source mapping formulas have protective limitations. They cannot use functions that change row counts or depend on table order.

Unsupported in mapping formulas and source queries

Table sources do not support the following functions in any way.

• Data definition language (DDL): CREATE, DROP, ALTER
• Procedures: CALL
• Table properties: TBLPROPERTIES
• Insecure functions: These functions expose sensitive information or allow arbitrary code execution, and are not supported. 
  • current_catalog
  • current_database
  • current_schema
  • current_user
  • java_method
  • raise_method
  • raise_error
  • reflect
  • try_reflect
  • user
  • version

Unsupported in mapping formulas

You can use the following functions in source queries but not in mapping formulas.

Aggregating functions

• array_agg
• bitmap_or_agg
• bitmap_construct_agg
• count
• count_if
• count_min_sketch
• cume_dist
• histogram_numeric
• hll_sketch_agg
• hll_union_agg
• kurtosis
• grouping
• grouping_id
• sum
• avg
• min
• max
• collect_list
• collect_set
• approx_count_distinct
• first
• last

Set generating functions

• explode
• explode_outer
• inline
• inline_outer
• json_tuple
• posexplode
• posexplode_outer
• stack

Side effect functions

• curdate
• current_date
• current_timestamp
• localtimestamp
• monotonically_increasing_id
• now
• rand
• randn
• random
• session_window
• shuffle

Window functions

• row_number
• rank
• dense_rank
• ntile
• lag
• lead
• cume_dist
• percent_rank