Skip to content

[SkipRecovery] Re-enable mixed-delete tests on Iceberg 1.10.2+ [reduced-it] - #15964

Open
wjxiz1992 wants to merge 2 commits into
NVIDIA:mainfrom
wjxiz1992:codex/skip12885-iceberg111
Open

[SkipRecovery] Re-enable mixed-delete tests on Iceberg 1.10.2+ [reduced-it]#15964
wjxiz1992 wants to merge 2 commits into
NVIDIA:mainfrom
wjxiz1992:codex/skip12885-iceberg111

Conversation

@wjxiz1992

@wjxiz1992 wjxiz1992 commented Sep 11, 2026

Copy link
Copy Markdown
Collaborator

JaCoCo production line coverage: not fully measurable locally — no compatible Scala 2.13 / shim 411 nightly baseline is published.

Contributes to #12885.

Description

This test-only change restores enforced CPU/GPU result checks for mixed Iceberg position and equality deletes on Iceberg 1.10.2 and newer. Both format-v2 and format-v3 tests currently allow failures and permanently pin their data seed, even when running with the upstream equality-delete cache fix. Production behavior is unchanged. The historical mismatch remains quarantined on older or unknown Iceberg versions; no count or row mismatch was observed on the tested 1.11.0 runtime.

Apache Iceberg #15514, released in 1.11.0 and backported to 1.10.2, fixes equality-delete records being cached in a query-dependent field order. This PR:

  • Narrows both the expected-failure and permanent-seed markers to older/unknown versions, recovering six cases across the three reader modes.
  • Keeps every count, full-row, and GPU-scan assertion. For the v3 count query only, disables adaptive query execution so the plan assertion sees the scan rather than an unfinished adaptive plan; full-row reads retain their existing configuration.
  • Resets the session configuration before fixture cleanup so SHOW TABLES does not inherit the scan-only assertion.

Older supported dependencies are not fixed by this PR, so #12885 should remain open.

Validation

Review follow-up: 44 actual decorator/seed-policy combinations passed, including the 1.10.1/1.10.2 boundary, later releases, and an unknown runtime. Re-ran all six mixed-delete cases plus runtime/shim detection on Spark 4.1.1 / Iceberg 1.11.0 with seed 1749483297: 7 passed (v3 includes OOM injection). Iceberg 1.10.2 was policy-checked only; it was not GPU-runtime-tested locally.

Local Spark 4.1.1 / Scala 2.13 / Iceberg 1.11.0, Hadoop catalog, Python 3.10.18, UTC, RTX 5880 Ada:

  • Complete merge-on-read file plus runtime/shim detection: 52 passed, no xfails or skips (seed 1749239638).
  • Six recovered cases plus runtime detection, original seed 1749483297 and forced OOM injection: 7 passed. Each reader returned CPU=GPU counts of 626 (v2) and 549 (v3), with full-row equality also passing.
  • Same focused scope, fresh seed 20260911: 7 passed.
  • Shim-411 mvn package for dist,integration_tests: BUILD SUCCESS. Validation used production base 31db6361829e8cbd258be94eda2842b667aef048; the later PCBS-only merge does not overlap this patch.

Reduced premerge collection retains all six recovered cases: each changed test has only the reader-type parametrization. Databricks is unaffected because the existing test harness explicitly skips Iceberg tests there. No production performance change.

JaCoCo execution data was captured during the forced-OOM run. Measured module deltas: none; sql-plugin, iceberg, delta-lake, shuffle-plugin, and udf-compiler deltas are N/A without a matching baseline. Published shim-350/401/404 reports cannot establish a shim-411 delta; no partial total is claimed.

AI assistance: The change and PR description were prepared with Codex assistance.

Checklists

Documentation

  • Updated for new or modified user-facing features or behaviors
  • No user-facing change

Testing

  • Added or modified tests to cover new code paths
  • Covered by existing tests
    (Please provide the names of the existing tests in the PR description.)
  • Not required

Performance

  • Tests ran and results are added in the PR description
  • Issue filed with a link in the PR description
  • Not required

Signed-off-by: Allen Xu <allxu@nvidia.com>
Copilot AI lite review requested due to automatic review settings September 11, 2026 04:00
@wjxiz1992 wjxiz1992 added the test Only impacts tests label Sep 11, 2026
@wjxiz1992 wjxiz1992 self-assigned this Sep 11, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟢 Approval recommended

No unresolved blocking issues were identified.

Pull request overview

Re-enables mixed Iceberg delete validation for Iceberg 1.11+ while preserving quarantine behavior for older or unknown versions.

Changes:

  • Gates xfail and fixed-seed markers by Iceberg version.
  • Disables AQE for the v3 count-plan assertion.
  • Resets Spark configuration before cleanup.
File summaries
File Description
integration_tests/src/main/python/iceberg/iceberg_merge_on_read_test.py Updates mixed-delete test gating and session cleanup.
Review details
  • Files reviewed: 1/1 changed files
  • Comments generated: 0
  • Review effort level: Lite

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

@greptile-apps

greptile-apps Bot commented Sep 11, 2026

Copy link
Copy Markdown
Contributor

RetriggerConfidence Score: 5/5

The PR appears safe to merge; no actionable correctness, security, or repository-rule violations remain.

Summary

  • Uses a patch-level version threshold so older and unknown runtimes remain quarantined.
  • Preserves deterministic seeds for quarantined runtimes.
  • Disables AQE for the v3 count plan assertion and resets session configuration before fixture cleanup.

Reviews (2) · Last reviewed commit: "Include Iceberg 1.10.2 in mixed-delete t..."

@wjxiz1992

Copy link
Copy Markdown
Collaborator Author

build


pytestmark = iceberg_unsupported_mark

# Iceberg 1.11.0 fixes cached equality-delete records being interpreted in the wrong field order:

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can the quarantine gate account for Iceberg 1.10.2 as a fixed runtime? Apache backported the equality-delete schema-ordering fix to the 1.10.x branch and released it in 1.10.2.

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated.

Signed-off-by: Allen Xu <allxu@nvidia.com>
@wjxiz1992 wjxiz1992 changed the title [SkipRecovery] Re-enable mixed-delete tests on Iceberg 1.11+ [reduced-it] [SkipRecovery] Re-enable mixed-delete tests on Iceberg 1.10.2+ [reduced-it] Sep 11, 2026
@wjxiz1992

Copy link
Copy Markdown
Collaborator Author

build

@wjxiz1992

Copy link
Copy Markdown
Collaborator Author

build

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

test Only impacts tests

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants