Skip to content

chore(deps): consolidate Dependabot dependency updates - #179

Merged
schenksj merged 3 commits into
mainfrom
deps/consolidated-dependabot-2026-06
Jun 19, 2026
Merged

schenksj merged 3 commits into
mainfrom
deps/consolidated-dependabot-2026-06

Conversation

@schenksj

@schenksj schenksj commented Jun 14, 2026 •

Copy link
Copy Markdown
Collaborator

Consolidates the open Dependabot PRs into a single change after triage. Each update was applied to the appropriate manifest and the result was verified to compile/run.

Applied

Maven (pom.xml)

Cargo (native/)

GitHub Actions

Held back (incompatible — not applied)

Verification

  • cargo check — clean
  • mvn test-compile — BUILD SUCCESS
  • PythonParityTest + BooleanFieldTest pass under junit 5.12.2 / surefire 3.5.6 (verified on JDK 11)

The three held-back PRs (#168, #171, #172) and the scala portion of #173 should remain open for a future coordinated upgrade; the others (#162–#167, #169, #170, #173, #175, #177, #178) can be closed in favor of this PR.

🤖 Generated with Claude Code

Triage and apply the open Dependabot dependency PRs in a single change.

Maven (pom.xml):
- jackson-databind 2.15.2 -> 2.22.0 (#178)
- maven-surefire-plugin 3.1.2 -> 3.5.6 (#177)
- exec-maven-plugin 3.1.0 -> 3.6.3 (#175)
- test-dependencies group (#173): junit 5.10.0 -> 6.0.3,
  s3 2.21.29 -> 2.43.0, azure-storage-blob 12.25.0 -> 12.33.3,
  azure-identity 1.12.0 -> 1.18.2, azure-core 1.48.0 -> 1.57.1,
  testcontainers 1.19.0 -> 2.0.5, google-cloud-storage 2.29.1 -> 2.67.0,
  google-cloud-nio 0.127.12 -> 0.131.0, mockito 5.7.0 -> 5.23.0,
  iceberg 1.7.1 -> 1.10.1, and maven-{compiler,resources,jar,source,
  javadoc,gpg}-plugin + central-publishing-maven-plugin bumps.

Cargo (native):
- futures 0.3.31 -> 0.3.32 (#164)
- tokio 1.45 -> 1.50 (#166)
- lru 0.12 -> 0.16 (#167)
- sha2 0.10 -> 0.11 (#163)
- thiserror 1 -> 2 (#170)
- libc / moka lockfile bumps (#169, #165)

GitHub Actions:
- actions/checkout v4 -> v6 (#162)

Held back (incompatible, not applied):
- scala-library 2.13.12 -> 3.8.3 (#173 sub-update): Scala 2->3 major;
  s3mock_2.13 requires Scala 2.13. Kept at 2.13.12.
- arrow-array / arrow-buffer 57 -> 58 (#168, #172): arrow, arrow-schema
  and parquet remain at 57 (pinned transitively by delta-kernel 0.19 and
  the quickwit fork). Mixing arrow 57/58 fails to compile.
- object_store 0.12 -> 0.13 (#171): breaking API changes and coupling
  with delta-kernel 0.19 / quickwit which pin 0.12.

Verified: `cargo check` clean; `mvn test-compile` clean;
PythonParityTest + BooleanFieldTest pass under junit 6 / surefire 3.5.6.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
JUnit 6.0.x raised its baseline to Java 17 (junit-jupiter-api class files
are bytecode 61). This project targets Java 11, and CI compiles with
JDK 11, so the test sources failed with "class file has wrong version
61.0, should be 55.0". It only passed locally because the dev machine
runs JDK 19.

Bump to 5.12.2 instead — the latest Java 8/11-compatible 5.x line (still
an upgrade from the original 5.10.0). Verified compiling and running
tests under JDK 11.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@schenksj

Copy link
Copy Markdown
Collaborator Author

CI on macOS ARM64 surfaced that junit-jupiter 6.0.3 requires Java 17 (class files are bytecode 61), while this project targets Java 11 — the test sources failed to compile with class file has wrong version 61.0, should be 55.0. (It passed local pre-push checks only because the dev machine runs JDK 19.)

Held junit back to 5.12.2, the latest Java 8/11-compatible 5.x line (still an upgrade from the original 5.10.0). Reproduced the failure and verified the fix locally under JDK 11 (mvn test-compile + smoke tests pass). Verified testcontainers 2.0.5 and the other major bumps are bytecode 52/55, so junit was the only Java-17 offender.

The Linux X64 leg failed at "Set up job" (runner/infra), unrelated to this change.

CI surfaced flaky failures in txlog::cache tests (e.g.
test_clear_all_caches_removes_all_entries, test_global_cache_stats).
Root cause is test cross-contamination, not a production bug: several
tests exercise the process-wide cache registry / global manifest cache,
and `clear_all_caches()` wipes that shared state. Under cargo's parallel
runner one test's clear races with another's get/put/stats.

Reproduced deterministically on the unmodified code at
`--test-threads=16`: 13/40 runs failed across test_global_cache_stats,
test_get_or_create_cache_reuse, test_global_manifest_cache and the
clear_all_caches tests. Serializing them on a shared mutex: 0/40.

The moka 0.12.13 -> 0.12.15 patch in this PR only shifted scheduling
enough to make CI hit a pre-existing latent race; production code is
sound (get_or_create_cache uses double-checked locking, the cache is
best-effort with TTL/LRU eviction so a flush racing a read is at worst a
miss, and the sole production caller is an explicit invalidate-all JNI
API).

Add a poison-tolerant mutex (lock_global_state) and guard the six tests
that touch global cache state. Local-instance tests still run in
parallel.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@schenksj

Copy link
Copy Markdown
Collaborator Author

Second CI failure was the Rust tests step: test_clear_all_caches_removes_all_entries and test_global_cache_stats (txlog::cache).

I checked whether this was masking a real bug — it isn't. It's pre-existing test cross-contamination: several tests exercise the process-wide cache registry / global manifest cache, and clear_all_caches() wipes that shared state, so under cargo's parallel runner one test's clear races with another's get/put/stats.

Proof, not hypothesis: on the unmodified code at --test-threads=16, 13/40 runs failed (across test_global_cache_stats, test_get_or_create_cache_reuse, test_global_manifest_cache, and the clear tests). After serializing the global-state tests on a shared mutex: 0/40. Full lib suite: 950/950.

Production code is sound — get_or_create_cache uses double-checked locking; the cache is best-effort (TTL+LRU eviction) so a flush racing a read is at worst a miss → fallback; the only production caller is the explicit nativeInvalidateCacheGlobal JNI API. The moka 0.12.13→0.12.15 patch in this PR just shifted scheduling enough for CI to hit a latent race that pre-exists on main.

Fix guards the six global-state tests with a poison-tolerant mutex; local-instance tests still run in parallel.

@schenksj
schenksj merged commit c2cae48 into main Jun 19, 2026
1 of 3 checks passed
@schenksj
schenksj deleted the deps/consolidated-dependabot-2026-06 branch June 19, 2026 20:51
schenksj added a commit that referenced this pull request Jul 5, 2026
iceberg-core/iceberg-api 1.11.0 ships Java 17 bytecode (class version
61.0), incompatible with this project's Java 11 target. Held at 1.10.1
(the version already verified working in #179).

actions/checkout@v7 requires a Node 24 runner action environment, which
this repo's self-hosted runners don't yet support (max Node 20).
Reverted to v6 until runners are upgraded.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

This branch was previously deployed

1 inactive deployment
test-runners — 396f854e Deployed Jun 19, 2026 by schenksj via Build & Test (macOS ARM64) #85
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants