flyrain commented on code in PR #14117: URL: https://github.com/apache/iceberg/pull/14117#discussion_r2525093400
########## format/udf-spec.md: ########## @@ -0,0 +1,300 @@ +--- +title: "SQL UDF Spec" +--- +<!-- + - Licensed to the Apache Software Foundation (ASF) under one or more + - contributor license agreements. See the NOTICE file distributed with + - this work for additional information regarding copyright ownership. + - The ASF licenses this file to You under the Apache License, Version 2.0 + - (the "License"); you may not use this file except in compliance with + - the License. You may obtain a copy of the License at + - + - http://www.apache.org/licenses/LICENSE-2.0 + - + - Unless required by applicable law or agreed to in writing, software + - distributed under the License is distributed on an "AS IS" BASIS, + - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + - See the License for the specific language governing permissions and + - limitations under the License. + --> + +# Iceberg UDF Spec + +## Background and Motivation + +A SQL user-defined function (UDF or UDTF) is a callable routine that accepts input parameters and executes a function body. +Depending on the function type, the result can be: + +- **Scalar function (UDF)** – returns a scalar value, which may be a primitive type (e.g., `int`, `string`) or a non-primitive type (e.g., `struct`, `list`). +- **Table function (UDTF)** – returns a table with zero or more rows of columns with a uniform schema. + +Many compute engines (e.g., Spark, Trino) already support UDFs, but in different and incompatible ways. Without a common +standard, UDFs cannot be reliably shared across engines or reused in multi-engine environments. + +This specification introduces a standardized metadata format for UDFs in Iceberg. + +## Goals + +* Define a portable metadata format for both scalar and table SQL UDFs. The metadata is self-contained and can be moved across catalogs. +* Support function evolution through versioning and rollback. +* Provide consistent semantics for representing UDFs across engines. + +## Overview + +UDF metadata follows the same design principles as Iceberg table and view metadata: each function is represented by a +**self-contained metadata file**. Metadata captures definitions, parameters, return types, documentation, security, +properties, and engine-specific representations. + +* Any modification (new definition, updated representation, changed properties, etc.) creates a new metadata file, and atomically swaps in the new file as the current metadata. +* Each metadata file includes recent definition versions, enabling rollbacks without external state. + +## Specification + +### UDF Metadata +The UDF metadata file has the following fields: + +| Requirement | Field name | Type | Description | +|-------------|------------------|------------------------|------------------------------------------------------------------| +| *required* | `function-uuid` | `string` | A UUID that identifies the function, generated once at creation. | +| *required* | `format-version` | `int` | Metadata format version (must be `1`). | +| *required* | `definitions` | `list<definition>` | List of function [definition](#definition) entities. | +| *required* | `definition-log` | `list<definition-log>` | History of [definitions snapshots](#definitions-log). | +| *optional* | `location` | `string` | Storage location of metadata files. | +| *optional* | `properties` | `map` | A string to string map of properties. | +| *optional* | `secure` | `boolean` | Whether it is a secure function. Default: `false`. | +| *optional* | `doc` | `string` | Documentation string. | + +Notes: +1. When `secure` is `true`: + - Engines MUST NOT expose the function definition or its body through any form of metadata inspection (e.g., `SHOW FUNCTIONS`). + - Engines MUST prevent leakage of sensitive information during execution via error messages, logs, query plans, or intermediate results. + - Engines MUST NOT perform predicate reordering, short-circuiting, or other optimizations that could change the order or scope of data access. + +### Definition + +Each `definition` represents one function signature (e.g., `add_one(int)` vs `add_one(float)`). + +| Requirement | Field name | Type | Description | +|-------------|----------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| +| *required* | `definition-id` | `string` | An identifier derived from canonical parameter-type tuple (lowercase, no spaces; e.g., `"(int,int,string)"`). If longer than 128 chars, use hashed form `"sig1-<base32(SHA-256(signature))[:26]>"`. | +| *required* | `parameters` | `list<parameter>` | Ordered list of [function parameters](#parameter). Invocation order **must** match this list. | +| *required* | `return-type` | [JSON representation](https://iceberg.apache.org/spec/#appendix-c-json-serialization) of an Iceberg type (`string` for primitives, `object` for complex types) | Type of value returned | +| *optional* | `nullable-return` | `boolean` | Whether the return value is nullable or not. Default: `true`. | +| *required* | `versions` | `list<definition-version>` | [Versioned implementations](#definition-version) of this definition. | +| *required* | `current-version-id` | `int` | Identifier of the current version for this definition. | +| *optional* | `function-type` | `string` (`"udf"` or `"udtf"`, default `"udf"`) | If `"udtf"`, `return-type` must be a `struct` describing the output schema. | +| *optional* | `doc` | `string` | Documentation string. | + +Notes: +1. `sig1-<base32(SHA-256(signature))[:26]>` is a fixed-length hash used when the canonical signature is too long. +It’s generated by taking the SHA-256 of the normalized signature, encoding it in Base32, keeping the first 26 characters, +and prefixing with `sig1-`. This yields a 31-character deterministic ID, easy to verify and future-proof via the `sig1-` +version prefix. + +### Parameter +| Requirement | Field | Type | Description | +|-------------|--------|----------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------| +| *required* | `name` | `string` | Parameter name. | +| *required* | `type` | [JSON representation](https://iceberg.apache.org/spec/#appendix-c-json-serialization) of an Iceberg type (`string` for primitives, `object` for complex types) | Parameter data type. | +| *optional* | `doc` | `string` | Parameter documentation. | + +Notes: +1. Function definitions are identified by the tuple of `type`s and there can be only one definition for a given tuple. +2. Parameter `name`s are immutable since named argument invocation is supported (e.g., `foo(a => 1, b => 2)`). Only `doc` can be updated in place. +3. Variadic (vararg) parameters are not supported. Each definition must declare a fixed number of parameters. +4. Each parameter input MUST be assignable to its declared Iceberg type. For complex types, the value’s + structure must match (correct field names, element/key/value types, and nesting). If a parameter—or any nested + field/element—is marked required, engines MUST reject null at that position (including inside structs, lists, and maps). +5. The `return-type` is immutable. To change it, users must create a new definition and remove the old one. +6. The function MUST return a value assignable to the declared `return-type`, meaning the returned value’s type and + structure must match the declared Iceberg type (including field names, element types, and nesting for complex types), + and any field or element marked as required MUST NOT be null. Engines MUST reject results that violate these rules. + +### Definition-Version + +Each definition can evolve over time by introducing new versions. Review Comment: That's a good point. Updated this section to the intent clearer and more explicit. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
