feat(core): Add incremental scan for appends and positional deletes #3

gbrgr · 2025-10-23T07:59:25Z

Incremental Scan Implementation

Summary

This PR introduces Incremental Scan functionality to the Iceberg Rust implementation, enabling efficient querying of changes between table snapshots. Incremental scans return the net changes (appends and deletes) between two snapshots, which is essential for incremental data processing workflows, change data capture (CDC), and efficient data pipeline operations.

Key Features

Incremental Scan API

New IncrementalScan builder with fluent API for configuring scans between snapshots
Support for scanning changes from a starting snapshot to an ending snapshot
Returns separate streams for appended data and positional deletes
Implements column projection via .select() for efficient data retrieval
Configurable batch size via .with_batch_size() for memory optimization

File Path Tracking

Automatically adds a reserved _file column to all delete record batches containing the source parquet file path
Uses RunEndEncoded arrays for memory-efficient file path storage in non-empty batches
Includes proper field metadata with reserved field ID (-2048)

Net Change Computation

Computes net effects between snapshots: rows added then deleted within the range don't appear in results
Efficiently processes only the data files and delete files that changed between snapshots
Supports complex scenarios including partial deletes, cross-file operations, and multiple snapshots

Implementation Details

Core Components

IncrementalScan Builder (scan/incremental/mod.rs)
- Validates snapshot ranges and table state
- Generates file scan tasks for appends and deletes
- Integrates with existing ArrowReader infrastructure
Streaming Implementation (arrow/incremental.rs)
- StreamsInto trait with .stream() method for converting scan tasks to Arrow record streams
- Separate processing for append tasks (data files) and delete tasks (positional deletes)
- Optimized delete task processing with schema reuse (O(1) allocations instead of O(n_batches))
File Path Column Addition (arrow/reader.rs)
- add_file_path_column() function adds _file column to record batches
- Handles both empty and non-empty batches correctly
- Maintains proper Arrow schema metadata for Iceberg field IDs

Restrictions

Does not support yet deletion of entire parquet files, or overwriting of parquet files.

Testing

Comprehensive test suite added in scan/incremental/tests.rs:

Test Fixture

IncrementalTestFixture - Helper for creating test tables with controlled snapshots
Supports Add operations with custom file names and data
Supports Delete operations with position and file tracking
Verification helper verify_incremental_scan() for asserting expected results

Test Coverage

test_incremental_fixture_simple - Basic append and delete operations
test_incremental_fixture_complex - Multiple snapshots with overlapping operations
- Tests 6 different snapshot range combinations
- Verifies net change computation (e.g., data added then deleted doesn't appear)
test_incremental_scan_edge_cases - Edge cases across 7 snapshots and 3 data files
- Partial deletes from multiple files
- Cross-file operations
- Empty result sets
test_incremental_scan_builder_options - Builder API functionality
- Column projection (.select())
- Batch size configuration
- Multiple batch size scenarios
test_add_file_path_column - Unit tests for file path column addition
- Normal case with RunEndEncoded arrays
- Empty batch handling with StringArray
- Special characters in file paths

All tests passing: ✅ 4 incremental scan tests, ✅ 3 file path column tests

API Example

// Scan changes between snapshot 2 and snapshot 5
let incremental_scan = table
    .scan()
    .incremental()
    .from_snapshot_id(2)
    .to_snapshot_id(5)
    .select(vec!["id", "name"])  // Column projection
    .with_batch_size(Some(1024)) // Configure batch size
    .build()?;

let (appends_stream, deletes_stream) = incremental_scan.to_unzipped_arrow()?;

// Process appended rows
while let Some(batch) = appends_stream.next().await {
    let batch = batch?;
    // batch contains appended data
}

// Process deleted positions
while let Some(batch) = deletes_stream.next().await {
    let batch = batch?;
    // batch contains (pos, _file) tuples for deleted positions
}

github-actions

license-eye has checked 363 files.

Valid	Invalid	Ignored	Fixed
296	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 366 files.

Valid	Invalid	Ignored	Fixed
299	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 366 files.

Valid	Invalid	Ignored	Fixed
299	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 366 files.

Valid	Invalid	Ignored	Fixed
299	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 366 files.

Valid	Invalid	Ignored	Fixed
299	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 366 files.

Valid	Invalid	Ignored	Fixed
299	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 366 files.

Valid	Invalid	Ignored	Fixed
299	2	65	0

Click to see the invalid file list

crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

github-actions

license-eye has checked 367 files.

Valid	Invalid	Ignored	Fixed
299	3	65	0

Click to see the invalid file list

crates/iceberg/src/scan/incremental/tests.rs
crates/playground/Cargo.toml
crates/playground/src/main.rs

Use this command to fix any missing license headers

```bash

docker run -it --rm -v $(pwd):/github/workspace apache/skywalking-eyes header fix

</details>

…eberg-rust into gb/incremental-bootstrap

vustef

Thanks Gerald, looking good to me, just a few last follow ups.

crates/iceberg/src/arrow/reader.rs

crates/iceberg/src/scan/incremental/mod.rs

vustef · 2025-10-30T11:59:50Z

crates/iceberg/src/scan/incremental/mod.rs

+            })?
+            .clone();
+
+        // TODO: What properties do we need to verify about the snapshots? What about


Let's remove this comment and track it separately elsewhere (Jira or google doc with open questions)

crates/iceberg/src/arrow/reader.rs

crates/iceberg/src/scan/incremental/context.rs

crates/iceberg/src/scan/incremental/mod.rs

crates/iceberg/src/arrow/reader.rs

Co-authored-by: Vukasin Stefanovic <[email protected]>

…eberg-rust into gb/incremental-bootstrap

crates/iceberg/src/scan/incremental/mod.rs

* WIP, initial draft of incremental scan * . * . * cargo fmt * Implement unzipped stream * Remove printlns * Add API method for unzipped stream * . * Remove comment * Rename var * Add import * Measure time * Fix typo * Undo some changes * Change type name * Add comment header * Fail when encountering equality deletes * Add comments * Add some preliminary tests * Format * Remove playground * Add more tests * Clippy * . * . * Adapt tests * . * Add test * Add tests * Add tests * Format * Add test * Format * . * Rm newline * Rename trait function * Reuse schema * . * remove clone * Add test for adding file_path column * Make `from_snapshot` mandatory * Error out if incremental scan encounters neither Append nor Delete * . * Add materialized variant of add_file_path_column * . * Allow dead code * Some PR comments * . * More PR comments * . * Add comments * Avoid cloning * Add reference to PR * Some PR comments * . * format * Allow overwrite operation for now * Fix file_path column * Add overwrite test * Unwrap delete vector * . * Add assertion * Avoid cloning the mutex guard * Abort when encountering a deleted delete file * Adjust comment * Update crates/iceberg/src/arrow/reader.rs Co-authored-by: Vukasin Stefanovic <[email protected]> * Add check * Update crates/iceberg/src/scan/incremental/mod.rs --------- Co-authored-by: Vukasin Stefanovic <[email protected]>

FFI and Julia Bindings for incremental scan changes that were introduced with RelationalAI/iceberg-rust#3. I refactored the file structures a bit, for both Rust and Julia code. Rust code reuses some common parts through macros, for Julia I didn't bother to do that (mostly afraid of macros and ccall interaction being a rabbit hole with little benefit). Note that I had to use struct instead of const for `ScanRef`, since now with additional type, we have method overloads, which if we use `ScanRef` const aliases actually use same type, and then become overwrites instead of overloads. There's also a new test data, and new test that exercises positional delete and inserts. --------- Co-authored-by: Gerald Berger <[email protected]>

gbrgr added 17 commits October 23, 2025 09:58

WIP, initial draft of incremental scan

fb5c737

Merge branch 'main' into gb/incremental-bootstrap

0fec28d

.

f255405

.

be932c8

cargo fmt

a378abe

Implement unzipped stream

317b5c6

Remove printlns

06694a8

Add API method for unzipped stream

4514695

.

f56e068

Remove comment

4125328

Rename var

9fa6360

Add import

14dd4e9

Measure time

837947a

Fix typo

5a7a223

Undo some changes

065c754

Change type name

f9ea05c

Add comment header

06cba66

github-actions bot reviewed Oct 24, 2025

View reviewed changes

Merge branch 'main' into gb/incremental-bootstrap

d3d57e3

github-actions bot reviewed Oct 24, 2025

View reviewed changes

Fail when encountering equality deletes

2ddde41

github-actions bot reviewed Oct 24, 2025

View reviewed changes

Add comments

facbeeb

github-actions bot reviewed Oct 24, 2025

View reviewed changes

gbrgr added 2 commits October 25, 2025 12:22

Add some preliminary tests

ae76630

Format

2780634

github-actions bot reviewed Oct 25, 2025

View reviewed changes

gbrgr and others added 11 commits October 29, 2025 13:08

Some PR comments

e31f76e

.

02f9a81

format

ec24817

Allow overwrite operation for now

efc57e8

Fix file_path column

f78fcca

Add overwrite test

97efb4c

Merge branch 'main' into gb/incremental-bootstrap

055b9b0

Unwrap delete vector

3aa40cb

.

584a684

Merge branch 'gb/incremental-bootstrap' of github.com:RelationalAI/ic…

e666dfd

…eberg-rust into gb/incremental-bootstrap

Add assertion

ff8a38c

vustef approved these changes Oct 30, 2025

View reviewed changes

Avoid cloning the mutex guard

f31bb64

vustef reviewed Oct 30, 2025

View reviewed changes

crates/iceberg/src/scan/incremental/context.rs Show resolved Hide resolved

gbrgr added 2 commits October 30, 2025 13:57

Abort when encountering a deleted delete file

c6f5ab6

Adjust comment

c0989b0

vustef reviewed Oct 30, 2025

View reviewed changes

crates/iceberg/src/scan/incremental/mod.rs Outdated Show resolved Hide resolved

vustef reviewed Oct 30, 2025

View reviewed changes

crates/iceberg/src/arrow/reader.rs Outdated Show resolved Hide resolved

gbrgr and others added 3 commits October 30, 2025 14:44

Update crates/iceberg/src/arrow/reader.rs

b95414c

Co-authored-by: Vukasin Stefanovic <[email protected]>

Add check

87da883

Merge branch 'gb/incremental-bootstrap' of github.com:RelationalAI/ic…

4982ee2

…eberg-rust into gb/incremental-bootstrap

vustef reviewed Oct 30, 2025

View reviewed changes

crates/iceberg/src/scan/incremental/mod.rs Outdated Show resolved Hide resolved

Update crates/iceberg/src/scan/incremental/mod.rs

6c60b90

gbrgr enabled auto-merge (squash) October 30, 2025 14:01

vustef mentioned this pull request Oct 30, 2025

Support (incremental) changelog scan for Change Data Capture use-cases apache/iceberg-rust#1636

Open

gbrgr merged commit 2a5a365 into main Oct 30, 2025
16 checks passed

gbrgr deleted the gb/incremental-bootstrap branch October 30, 2025 14:08

vustef mentioned this pull request Oct 31, 2025

FFI and Julia Bindings for incremental scan changes RelationalAI/RustyIceberg.jl#6

Merged

feat(core): Add incremental scan for appends and positional deletes #3

feat(core): Add incremental scan for appends and positional deletes #3

Uh oh!

Conversation

gbrgr commented Oct 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Incremental Scan Implementation

Summary

Key Features

Implementation Details

Core Components

Restrictions

Testing

API Example

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

vustef left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

vustef Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

gbrgr commented Oct 23, 2025 •

edited

Loading