Conversation
Includes fix for FixedSizeBinary LEFT JOIN bug - apache/arrow-rs#8981 Cherry-picked test and API updates from - apache#19355
…ons (apache#19078)" This reverts commit ada0923.
…e#19904) ## Which issue does this PR close? <!-- We generally require a GitHub issue to be filed for all bug fixes and enhancements and this helps us generate change logs for our releases. You can link an issue to this PR using the GitHub syntax. For example `Closes apache#123` indicates that this PR will close issue apache#123. --> - Part of apache#19798 ## Rationale for this change <!-- Why are you proposing this change? If this is already explained clearly in the issue then this section is not needed. Explaining clearly why changes are proposed helps reviewers understand your changes and offer better suggestions for fixes. --> This is an edge case, would prefer to fix upstream in arrow-rs instead of having handling code here, so just disable test for now. - arrow-rs issue: apache/arrow-rs#9227 ## What changes are included in this PR? <!-- There is no need to duplicate the description in the issue here but it is sometimes worth providing a summary of the individual changes in this PR. --> Disable edge-case array_union SLT ## Are these changes tested? <!-- We typically require tests for all PRs in order to: 1. Prevent the code from being accidentally broken by subsequent changes 2. Serve as another way to document the expected behavior of the code If tests are not included in your PR, please explain why (for example, are they covered by existing tests)? --> Test related change ## Are there any user-facing changes? <!-- If there are user-facing changes then we may require documentation to be updated before approving the PR. --> No. <!-- If there are any breaking changes to public APIs, please add the `api change` label. -->
Hot-fix for security_audit CI on upgrade-df-ver5250-a. Addresses 5 RUSTSEC advisories that landed between 2026-04-14 and 2026-04-27, after upstream DataFusion 52.5.0 was tagged: - RUSTSEC-2026-0098 — rustls-webpki: name constraints for URI names - RUSTSEC-2026-0099 — rustls-webpki: name constraints accepted for wildcard certs - RUSTSEC-2026-0104 — rustls-webpki: reachable panic in CRL parsing - RUSTSEC-2026-0112 — astral-tokio-tar: PAX header desync - RUSTSEC-2026-0113 — astral-tokio-tar: unpack_in chmod via symlink Bumps Cargo.lock only (no Cargo.toml dep range changes): - astral-tokio-tar 0.6.0 -> 0.6.1 - rustls-webpki 0.103.10 -> 0.103.13 When upstream backports these (or the next 52.x release ships them), this commit will auto-drop on rebase.
Cherry-pick of apache#21032. Adapted for 52.5.0: - Two unused imports in datafusion-expr's test module (`IntervalUnit`, `logical_int64`) compile-fail under -D warnings; dropped. - Two sql-integration tests and several sqllogictest expected-error patterns assumed the friendlier Display impl from apache#20605 was already on 52.5.0 (it isn't); updated those patterns to the pre-apache#20605 verbose `TypeSignatureClass::Native(LogicalType(...))` shape and to the post-apache#21032 `failed with: ...` (no quoted inner message) shape. - Also includes the author's own follow-up CI fix (was a separate commit upstream, folded here). Signed-off-by: yaommen <myanstu@163.com>
Patches RUSTSEC-2026-0145 (PAX header desynchronization, GHSA-3cv2-h65g-fgmm). Transitive dev-dep of `datafusion-cli` via `testcontainers` / `testcontainers-modules`; Cargo.lock-only update.
…23706) ## Which issue does this PR close? - Closes apache#23705. ## Rationale for this change Arrow arrays containing no null values commonly omit the null bitmap. The whole-partition evaluation path for `LEAD` and `LAG` with `IGNORE NULLS` unconditionally unwrapped that optional bitmap, causing a panic for valid non-null input arrays. ## What changes are included in this PR? - Fall back to the existing regular shift implementation when the input has no null bitmap. - Add a regression test covering both `LEAD` and `LAG` with an array that explicitly has no null bitmap. ## Are these changes tested? Yes. ## Are there any user-facing changes? `LEAD` and `LAG` with `IGNORE NULLS` no longer panic when the input Arrow array has no null bitmap. There are no public API changes. (cherry picked from commit a8d1af6)
UnionExec::try_new recomputes the union schema via union_schema() on every construction, including with_new_children rebuilds performed by optimizer passes. For wide unions whose children share one schema (generated UNION ALL, unions of per-partition scans) this makes physical planning O(n^2) in child count. Two fast paths: - union_schema(): if every input schema is pointer- or content-equal to the first, return the first schema (merging N identical schemas is the identity operation). - UnionExec::with_new_children(): when child count and per-position child schemas are unchanged, reuse the existing schema. Plan properties are always recomputed, since they can legitimately change when schemas do not. (DataFusion 53 has an analogous properties optimization upstream, apache#19792; this variant targets the 51 line.) Adds a union_schema benchmark covering shared-Arc, content-equal, and adversarial (last child differs) shapes. All existing union unit tests pass. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
github-actions
Bot
added
documentation
…e per output chunk sort_batch_stream reserves the sum of get_record_batch_memory_size over the chunks from sort_batch_chunked. take() shares the dictionary values across every chunk, so the dictionary is counted once per chunk. A 2.5 MB batch with a 50k-value dictionary asks for 121.8 MB. With the disk manager disabled there is no spill fallback and the sort fails.
…once sort_batch_stream reserved the sum of per-chunk sizes for the output of sort_batch_chunked. Dictionary values are shared by every chunk after take(), so a batch with a large dictionary reserved chunks x dictionary and failed with ResourcesExhausted or starved the pool. Size the sorted output with one pass over all chunks so a shared buffer is counted once. ReservationStream releases per batch, and per-batch sizes can now add up to more than was reserved, so it releases at most what remains; the rest is freed when the stream ends.
MinIO withdrew minio/minio from Docker Hub on 2026-09-11, so the storage integration tests fail to start their container. Point the image pinned by testcontainers-modules at quay.io, which serves the same tags. Minimal backport of apache#25216.
…d batch Clamping the per-batch release kept the reservation from underflowing but freed shared buffers as soon as the first batch was emitted, while the queued batches still referenced them. get_record_batches_release_sizes charges each buffer to the last batch that references it, so the sum is the shared-buffer total and the stream releases exactly what each batch stops holding. ReservationStream had no other user and is removed.
get_record_batch_memory_size and get_record_batches_memory_size count into a scalar again; only sort_batch_stream builds the release-size vector. Add a test that drops a partially consumed sorted stream and checks the pool returns to zero, and share the test scaffolding.
datafusion/common/src/utils/memory.rs as of apache/datafusion main 9951f0b (2026-09-24): apache#24319 introduced the counter and apache#24364, apache#24837, apache#25335 refined it. Counts each buffer once across a sequence of batches; replaces the fork-local helpers.
…unter Walk the chunks in reverse so each buffer is charged to the last chunk that references it; the counter total is the reservation and the per-chunk counts are what each emitted chunk releases.
…servation fix(physical-plan): count buffers shared across sorted output chunks once
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
7 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.
Revision c plus two forward-ported fixes that landed on the DF 51 line (
upgrade-df-ver5100-a) on 2026-08-14, after revision c was cut. Both are customer-reported bugs thatmainalready carries via influxdata/influxdb_pro#5221 — without forward-porting them, merging the 52.5 upgrade would silently drop both.mainNew commits on top of revision c
fix: prevent LEAD/LAG IGNORE NULLS panic without null bitmap (#23706)— cherry-pick of fix: prevent LEAD/LAG IGNORE NULLS panic without null bitmap (#23706) #102. Upstream fix (fix: prevent LEAD/LAG IGNORE NULLS panic without null bitmap apache/datafusion#23706, authored by xudong.w), cherry-picked to the 51 line by @egillespie. Fixes a panic reported by a Cloud Dedicated customer in influxdata/EAR#7049. Applies cleanly; not present in the 52.5.0 base.fix: fast paths for UnionExec schema recomputation— cherry-pick of fix: fast paths for UnionExec schema recomputation (O(n²) planning on wide unions) #100 (@reidkaufmann). Two fast paths inUnionExec/union_schemathat eliminate the O(n²)-in-child-count schema recomputation during physical planning of wide unions. Root-caused via influxdata/EAR#7010 (Vitol): a dedup-heavyLIMIT 1plan over ~4,177 per-partition union branches took ~50s to plan and ~0.06s to execute; with the fix, planning drops to ~1.7s. Behaviour-identical — same plan. Applies cleanly; not present in the 52.5.0 base.One adaptation was needed: the accompanying
benches/union_schema.rsis reformatted for this branch's rustfmt. The 52.x line uses edition-2024 import sorting, socriterion_group/criterion_mainandExecutionPlanorder differently than on the 51 line. Two lines, folded into the cherry-pick.Unchanged — the 10 patches described in #99, including the
to_timestamptimezone revert (influxdata/EAR#6765) and theOneOfsignature diagnostics cherry-pick.Per convention this PR is a review surface only — do not merge. Consumers pin
e9e2e6d36bf63e39f746f82ee9b33c47feb65361by rev