Conversation
Update iceberg and iceberg-catalog-rest to rev 418213731e91544f5eb31a3efa459e88f599030e, which includes the fix for incremental scans with from=None silently dropping EXISTING manifest entries from expired snapshots. Also pulls in arrow/parquet v58.1.0 via the updated lock file. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
iceberg-rust rev 418213731e91 depends on arrow/parquet v58.1.0; keeping the FFI crate on 57.x caused two incompatible versions of RecordBatch to be linked, failing to compile. Align all arrow-* and parquet pins to "58.1". Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Accepts Vector{String} with optional validity BitVector, handling pointer
extraction and GC preservation internally. The low-level ptr/len overload
is retained for performance-sensitive callers with pre-allocated buffers.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Contributor
Author
|
I am seeing good throughput improvements locally: vs. (some results are noisy due to local execution, but all workloads consistently show improvements. More so for numeric types due to batched gathering on the Rust side). |
…diate Vec Build StringArray directly with OffsetBuffer + values Buffer in a single pass, replacing Vec<Option<&str>> + StringArray::from. Use new_unchecked to skip Arrow's UTF-8 re-validation — Julia strings are guaranteed valid UTF-8. For 20M x 32-byte strings this eliminates ~320 MB of intermediate Vec<Option<&str>> storage and ~640 MB of redundant UTF-8 validation reads per column. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
robertbuessow
approved these changes
Apr 30, 2026
robertbuessow
left a comment
robertbuessow
left a comment
There was a problem hiding this comment.
Not sure I understood everything but I think it's good enough. Some nits + it would be good to improve error handling between rust and julia.
- Extract encode worker loop body into encode_worker_loop() - Retain panic message: downcast Box<dyn Any> to &str / String before formatting - Add iceberg_take_gather_error() FFI + thread-local to surface gather errors immediately in Julia exceptions rather than deferring to writer close - Clarify lengths_ptr doc: array of byte lengths per string - Merge identical sequential/scattered validity branches in build_null_buffer_scattered - Add explanatory comments: bitvector merging, re-alignment, all-valid bit-set Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #90 +/- ##
==========================================
- Coverage 84.42% 82.81% -1.61%
==========================================
Files 9 9
Lines 873 966 +93
==========================================
+ Hits 737 800 +63
- Misses 136 166 +30 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Export improvements
This PR reworks the Parquet/Iceberg write path for throughput, cleanliness, and configurability.
New write APIs
WriterConfigstruct with configurable Parquet properties: compression codec, dictionary encoding, plain encoding, row group size, page size, write batch size, and statisticsColumnBatch/write_columns— zero-copy write path for flat column buffers (bypasses Arrow IPC serialization)GatheredBatch/write_columns— gathered-column write path that assembles columns from scattered slices (selection vectors + validity bitmaps) directly in the calling threadset_encode_workers!to configure the encode thread pool size before first useGlobal encode worker pool
A single pool of N OS threads (default:
Sys.CPU_THREADS) is shared across all writers. Eachwrite_columns/writecall gathers or serializes data in the calling thread, submits aRecordBatchto the pool, and returns immediately — encode and Parquet I/O run on pool threads. Per-writer ordering is preserved via aMutex<ConcreteDataFileWriter>in the sharedWriterState: only one pool thread encodes a given writer at a time.close_writerwaits for all in-flight tasks to drain before finalising the file.This design lets Julia pipeline the gather/serialize step on the main thread with encode work happening concurrently on pool threads, rather than blocking end-to-end on each write.
Tests
Added tests for
ColumnBatch,GatheredBatch(including scattered slices, nullable columns, string columns), decimal types (Int32/Int64/Int128 backing), and allWriterConfigParquet properties.