[scottf]

Build and CI work only. The batch-publish EOB / fast-ingest code is deliberately not here — it follows in a separate PR branched from this one.

Dependencies

• jnats-server-runner 3.1.1 in every project that uses it.
• chaos-runner ported off 2.0.2, which was two breaking releases behind. 3.x moved the cluster helpers out of NatsRunnerUtils into ClusterUtils/ClusterDefaults, removed getTemporaryJetStreamStoreDirBase(), renamed connectCheckTries(0) to skipConnectValidate(), and moved setDefaultOutputLevel. It asks for LocalHost.ip.host explicitly, because the 3.x default is localhost on a Java 8 JVM and 127.0.0.1 elsewhere, and chaos-runner used to be 127.0.0.1 everywhere.

Known issue, unchanged by this PR and recorded for later: on 3.1.x a cluster's JetStream store dir is not honored — NatsServerRunner re-parses the jetstream block from configInserts through JsConfig(List), which substitutes a fresh temp dir. Measured on Linux. 4.0.1 fixes it.

Gradle

• The six projects still on 6.8.3 moved to 8.14, with wrapper artifacts regenerated by gradle wrapper --gradle-version 8.14 — jar, properties, gradlew, gradlew.bat, byte-identical to the projects already on 8.14. pcgroups-cli stays on 9.0.0.
• Those six aligned to the batch-publish template: coveralls plugin dropped (no workflow referenced it), bnd 7.1.0, test-retry 1.6.4, nexus-publish 2.0.0, junit 5.14.1 plus junit-platform-launcher, equalsverifier 4.2.3, jacoco 0.8.12, xml.enabled → xml.required (removed in Gradle 8), and the javadoc doLast favicon block removed.
• jacoco 0.8.6 could not instrument JDK 21 class files — it logged Unsupported class file major version 65 and still reported success, producing a degraded report.

Manifests

Automatic-Module-Name now equals Bundle-Name on all ten published bundles; four were missing it entirely, which left their module name derived from the jar filename. batch-publish's bnd Bundle-Description and Bundle-DocURL pointed at counters and are corrected.

Build file cleanup

• TARGET_COMPATIBILITY machinery removed from all eleven build files. Nothing set it — not this repo's workflows, not the shared ones — so only the Java 8 artifact was ever published. targetCompatibility is now a fixed JavaVersion.VERSION_1_8; jarAndArtifactName stays as the artifact id, replacing the archivesBaseName that Gradle 9 removes.
• isMerge removed from the ten files that declared it. It was referenced in exactly one place — js-publish-extensions wrapped nexusPublishing in if (isMerge || isRelease), a no-op guard. All eleven now configure it identically.

Workflows

21 files across seven projects migrated to synadia-io/workflows, 623 lines removed. Each file keeps its name: and on: block byte-for-byte; only the jobs: block changed, to the same five-line caller counters already used. This retires actions/checkout@v3, actions/setup-java@v3, the adopt distribution, the second nats-server install method, and the signArchives reference that Gradle 9 breaks. pcgroups-cli keeps its own workflows since it publishes a distribution, not a library.

The JDK question that previously blocked this is resolved: the shared workflow hardcodes JDK 21, and Gradle 6.8.3 cannot run on 21 at all — 8.14 can.

Verification

Everything below was run against the exact contents of this branch, checked out into a separate worktree, using nats-server v2.15.0-dev:

• clean test jar javadoc succeeds in all eleven standalone projects, plus :pcgroups-cli:compileJava through its parent.
• 131 tests, 0 failures — pcgroups 67, encoded-kv 19, request-many 18, schedule-message 13, counters 6, direct-batch 4, js-publish-extensions 2, retrier 2.
• chaos-runner has no test suite, so it was run for real: a three-server JetStream cluster with the chaos loop cycling servers down and up, jnats clients reconnecting and rediscovering, healthz returning to ok after each restart. Single-server mode likewise.
• batch-publish has no test sources on this branch, which is why its 44 tests are absent here; they arrive with the feature PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_01V5QAt1CFwPXNN8nSAqWdvX

[synadia-claude-reviewer]

Claude finished @scottf's task in 3m 56s —— View job

---

PR Review: Build and CI Cleanup

• Gather context and understand the request
• Read PR diff and changed source files
• Review chaos-runner API migration (ChaosRunner.java, ChaosArguments.java)
• Review build file changes (Gradle upgrades, plugin versions)
• Review workflow consolidations
• Review test file changes
• Post inline comments on specific issues
• Post final review summary

---

Overall Assessment

This is a well-scoped, mechanically thorough cleanup PR. The changes are large in volume but narrow in purpose — no logic rewrites, just infrastructure alignment. Two items worth tracking:

---

Findings

1. batch-publish uses a SNAPSHOT runtime dependency — batch-publish/build.gradle:39
(inline comment posted)

io.nats:jnats:2.26.3-SNAPSHOT makes builds non-deterministic and will block Maven Central publication. The PR description is clear that the feature PR handles the real batch-publish work, so this is presumably intentional as a placeholder. Just confirm the downstream PR bumps this to a GA release before releasing batch-publish.

2. Pre-existing NPE in ChaosRunner.start() — ChaosRunner.java:301

INSTANCE_ARGUMENTS is declared but never assigned anywhere in the class. If start() is called a second time with INSTANCE != null, the guard INSTANCE_ARGUMENTS.equals(a) throws NullPointerException instead of the intended Exception("Instance already started with different arguments."). This is present on main and not introduced by this PR, but the chaos-runner port is a good opportunity to fix it — the fix is a one-liner (INSTANCE_ARGUMENTS = a; after INSTANCE = new ChaosRunner(...)).

---

What looks correct

• chaos-runner API migration is complete and consistent: connectCheckTries(0) → skipConnectValidate(), NatsServerRunner.setDefaultOutputLevel → NatsRunnerUtils.setDefaultOutputLevel, ClusterDefaults builder replacing the old positional createNodes overload, and explicit LocalHost.ip.host for deterministic 127.0.0.1 behavior across JVM versions. The monitorStart(monitor) call with the comment that ≤ 0 disables monitoring is consistent with the old monitor < 1 ? null : monitor guard.
• Test migration (nats.io.* → io.nats.*, getURI() → getNatsLocalhostUri()) is uniform across all six test files.
• Automatic-Module-Name additions are correct; module names match Bundle-Name on all ten bundles.
• jacoco 0.8.6 → 0.8.12 correctly handles JDK 21 class files (0.8.6 silently produced degraded reports, as noted in the PR description).
• xml.required / html.required is the correct Gradle 8 syntax; xml.enabled was removed.
• isMerge removal is safe — the PR description's analysis that it guarded a no-op is accurate.
• Workflow consolidation — 21 files to 5-line callers — is clean, retires deprecated action versions (checkout@v3, setup-java@v3, adopt distribution), and removes the signArchives reference that Gradle 9 breaks.
• .gitattributes correctly enforces LF for gradlew/.sh and CRLF for .bat/.cmd.

[scottf]

The findings about NPE in ChaosRunner.start(), while true are not relevant to this PR and should/will be addressed separately