fix: race condition when attaching futures to a terminating stream

[v0y4g3r]

Summary

• Log table name, request ID, row count, and timeout when BulkStreamWriter writes time out.
• Deduplicate caller-side and internal deadline timeout diagnostics.
• Publish the project as 0.15.1-SNAPSHOT with Central plugin 0.11.0 and a guarded manual workflow.
• Fix AsyncPutListener terminal-stream/attach race (closes AsyncPutListener can miss terminal stream errors for newly attached batches #105): introduce a lock-protected terminal state machine so attach() and onError()/onCompleted() cannot race. Synchronous putNext() failures now release the unsent Arrow metadata buffer and return the write-limiter permit immediately.

Test Plan

• mvn test
• mvn spotless:check
• JAVA_HOME=... mvn -P release -DskipTests -Dgpg.skip=true clean verify
• Verify manual publishing rejects non-SNAPSHOT versions
• 11 regression tests for terminal race, callback ordering, send failure, buffer ownership, and failure precedence

[Copilot]

Pull request overview

This pull request improves bulk write timeout diagnostics by logging request context (table, request ID, row count, timeout) and deduplicating timeout reporting between caller-side waits and internal deadline enforcement. It also bumps the project version to the next snapshot and tightens publishing workflow behavior for manual runs.

Changes:

• Add timeout-context logging and deduplication in the bulk write pipeline (BulkWriteService, BulkWriteClient).
• Add tests validating caller-timeout logging propagation and “log once” behavior.
• Bump Maven versions to 0.15.1-SNAPSHOT, update Central publishing plugin, and add guarded manual publishing workflow dispatch.

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
pom.xml	Bumps project version and updates Central publishing plugin version.
ingester-rpc/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-protocol/src/main/java/io/greptime/BulkWriteClient.java	Wraps bulk write futures to log caller-side timeouts with request context.
ingester-protocol/src/test/java/io/greptime/BulkWriteClientTest.java	Adds unit test ensuring timed get(...) reports timeout to the put stage.
ingester-protocol/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-prometheus-metrics/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-integration-tests/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-grpc/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-example/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-common/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-bulk-protocol/src/main/java/io/greptime/BulkWriteService.java	Adds table-name-aware timeout diagnostics and deduplicates deadline vs caller timeout logs.
ingester-bulk-protocol/src/main/java/io/greptime/BulkWriteManager.java	Passes table name into BulkWriteService for diagnostics.
ingester-bulk-protocol/src/test/java/io/greptime/BulkWriteServiceTest.java	Adds test ensuring request-context timeout log is emitted only once.
ingester-bulk-protocol/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
ingester-all/pom.xml	Aligns module parent version to 0.15.1-SNAPSHOT.
.github/workflows/mvn_publish.yml	Adds manual workflow dispatch and rejects manual publishes for non--SNAPSHOT versions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[fengjiachun]

Almost LGTM

[fengjiachun]

Reviewed the concurrency fix carefully — the core design holds up well: terminate() snapshots and clears futuresInFlight under the lock but completes futures outside it (no callback-under-lock deadlock risk); the re-entrancy path is safe since a failed future's callback hits the terminal check in onError() and returns false; and failing pending futures on normal completion instead of silently clearing them is the right call — that silent clear was part of the original leak. The AbstractLimiter permit-leak fix and the blockingWriteLimitTest rewrite (the old in-thread assertion could never actually fail the test) are both good catches. I ran all 18 new/changed tests locally and they pass.

Two issues that can't be anchored to the diff:

1. The new BulkWriteServiceTest never runs in CI. build.yml only has test jobs for ingester-common, ingester-grpc, ingester-protocol, and integration tests — there's no job for ingester-bulk-protocol, so the main regression suite of this PR won't run anywhere automated. Also, these tests fail on JDK 9+ without --add-opens=java.base/java.nio=ALL-UNNAMED (reproduced on Java 21: 4 errors from MemoryUtil initialization; all 11 pass with the flag). Suggest adding a test_ingester_bulk_protocol job, plus a JDK-9+-activated profile setting surefire argLine so local runs on modern JDKs work too.

2. The release workflow guard is one-directional. Manual dispatch requires a -SNAPSHOT version, but the release: created path has no inverse check — creating a GitHub release while main sits at 0.15.1-SNAPSHOT would publish a snapshot through the release pipeline. A symmetric "release requires non-SNAPSHOT" step would close that hole.

Remaining line-specific notes are inline. Overall LGTM once #1 is addressed — as written, the regression tests protecting this fix won't run in CI.

[v0y4g3r]

@fengjiachun Addressed the two unanchored review findings from this review:

• e784f05 adds the missing ingester-bulk-protocol CI job, makes integration tests depend on it, and activates the Arrow --add-opens Surefire setting on JDK 9+ while preserving Java 8 behavior.
• d711c28 adds the inverse release guard so release.created rejects -SNAPSHOT versions before deployment.

Verified with the full Java 8 reactor test suite, mvn spotless:check, and the bulk protocol tests on Java 21.