Create a Computed Join
This page documents the Add Computed Join form, then walks through the steps to create one from the datastore container list. For the conceptual overview, see How Computed Join Works.
The form has two editor modes. Pairwise joins two containers on a single key. SQL joins two or more aliased sources with a Spark SQL query you write. The Editor Mode tabs switch between them, and the fields below the tabs change with the mode.
Permissions
You need Editor permission on the parent datastore, or Author permission plus ownership of the new container, and at least Viewer permission on every other datastore an input reads from. The parent datastore is the left container's datastore in pairwise mode, or the first source's datastore in SQL mode. See Computed Join Permissions for the full matrix.
Show me how
The app can walk you through this. Click Show me how in the form's header, or press H while it is open, and the Add a Computed Join walkthrough highlights each step while you fill in the real form. See The How-tos Scope for every way to start one.
Field reference
The Add Computed Join form starts with the fields every join has, then shows the fields for the selected Editor Mode, and ends with the metadata fields.
Common fields
| Field | Required | Type | Description |
|---|---|---|---|
| Name | Text | Unique name for the Computed Join container on the parent datastore. All spaces are replaced by underscores. Max 255 characters. | |
| Editor Mode | Tabs | Pairwise (default) or SQL. Pick the tab below for the fields each mode shows. |
Mode is fixed after save
The Editor Mode tabs are locked once the join is saved. To switch from pairwise to SQL mode (or the reverse), you must delete the join and create a new one. Deletion is destructive: the join's history, quality checks scoped to it, and anomalies detected on it are removed with the container. Rebuild those artifacts on the new join.
Mode fields
Pairwise mode joins a left and a right container on one key each. Every field below is required except the two prefixes and the Filter and Group By clauses.
Join Type
| Field | Required | Type | Description |
|---|---|---|---|
| Join Type | Option | One of Inner Join, Left Join, Right Join, or Full Outer Join. Defaults to Inner Join. |
Left Reference
The second field's label follows the type of the left datastore: Table for schema-based datastores and File for file-based datastores.
| Field | Required | Type | Description |
|---|---|---|---|
| Datastore | Option | Locked to the datastore you opened. The Computed Join lives under this datastore. | |
| Table or File | Option | Profiled container to join from the left side: table, view, file, Computed Table, or Computed File. | |
| Field | Option | The column used as the join key on the left side. Must resolve to the same data type as the right Field; mismatched types surface as a validation error when you click Validate or Save. | |
| Prefix | Text | Prefix applied to every column from the left side in the joined result. Defaults to left. Max 100 characters. |
Right Reference
The second field's label reads Container until a right datastore is selected, then follows that datastore's type in the same way as the left side.
| Field | Required | Type | Description |
|---|---|---|---|
| Datastore | Option | Datastore for the right container. Can be the same as the left datastore or a different one. You need at least Viewer permission on it. | |
| Table, File, or Container | Option | Profiled container to join from the right side. Same supported types as the left side. | |
| Field | Option | The column used as the join key on the right side. Must resolve to the same data type as the left Field. | |
| Prefix | Text | Prefix applied to every column from the right side. Defaults to right. Max 100 characters. Should differ from the left Prefix when both sides share column names; identical prefixes trigger a duplicate-column error at Validate. |
Select Expression, Filter Clause, and Group By Clause
| Field | Required | Type | Description |
|---|---|---|---|
| Select Expression | Text | SQL select clause naming the columns to include in the joined result. Auto-fills with every top-level field from both sides, prefixed. Edit it to keep only what you need, add aliases, or include expressions. | |
| Filter Clause | Text | Post-join WHERE predicate applied to the joined result. |
|
| Group By Clause | Text | Group By columns when the Select Expression uses aggregation functions. Every non-aggregated column in the Select Expression must appear in this clause. |
SQL mode replaces the Join Type, Left/Right Reference, Select Expression, Filter Clause, and Group By Clause with a Sources list and a Query editor.
Sources
A SQL join reads from 2 to 10 sources. Each source row has these fields:
| Field | Required | Type | Description |
|---|---|---|---|
| Add New Source | Button | Adds a source row. The form starts with the 2 required sources and accepts up to 10; at 10 the button is disabled and its tooltip reads "A computed join supports at most 10 sources". Once more than 2 sources exist, every source header shows a remove icon. | |
| Datastore | Option | The datastore the source container lives in. Sources may span datastores; you need at least Viewer permission on each. The Computed Join is created under the first source's datastore. | |
| Table, File, or Container | Option | The container to read. The label reads Container until the source's datastore is selected, then follows that datastore's type. It must already be profiled. Base tables, views, files, Computed Tables, and Computed Files are all accepted; another Computed Join is not. | |
| Alias | Text | The name the query uses to reference this source. Must start with a letter or underscore, contain only letters, digits, and underscores, be unique within the join, and not clash with a CTE name in the query. | |
| Filter Clause | Text | A Spark SQL WHERE expression applied while the source is read, before the query runs. Filtered-out rows are never loaded, so this is the main lever for large sources. |
Query
| Field | Required | Type | Description |
|---|---|---|---|
| Query | Text | A single Spark SQL statement that references the aliases as tables. Supports composite join keys, CTEs, GROUP BY, window functions, and joins across any number of sources. Reference sources by their alias only: qualified or catalog table names, statements that modify data, and arbitrary-code functions (reflect, java_method) are rejected. |
Description, Owner, and Additional Metadata
These fields are the same in both modes.
| Field | Required | Type | Description |
|---|---|---|---|
| Description | Text | Free-text description of the Computed Join, with Markdown formatting (max 8000 characters). Surfaces alongside the container in lists and search results. The form can auto-suggest a description from the join configuration. | |
| Owner | Option | The user who owns the Computed Join. Defaults to the user creating the join. Only Editors can pick a different owner at create time; Authors are locked to themselves. | |
| Additional Metadata | Key-value | Key-value metadata attached to the container. Keys also act as default values for runtime variables in this container's checks. |
Steps
Step 1: Open the datastore where the Computed Join will live and click Add.
Step 2: Choose Computed Join from the dropdown. The Add Computed Join modal opens in pairwise mode.
Step 3: Enter a Name. Under Editor Mode, keep Pairwise or select the SQL tab.
Step 4: Fill in the fields for the selected mode. See the Field reference above for what each field holds.
Step 5: Click Validate. A Validation Successful message appears if the SQL parses and the schema resolves against the input containers.
Step 6: Click Save. A success message appears, the modal closes, and the new Computed Join appears in the datastore's container list.