[polyglotAI-bot]

Summary

Backports #3217 from main to v0.11.0, as requested by @chernser in #3231 (comment):

@polyglotAI-bot I'm a maintainer and asking you do the PR to v0.11.0 for me I'll review

Issue: #3231 (a closing keyword has no effect on a PR into a non-default branch, so close the issue manually after merge).

Cherry-picked PR

PR	Merge commit on main	Change	Backport commit
#3217	848b3116b	ResultSet#close() / QueryResponse#close() do not fail when the server ended a partly read response early (#3203)	536f9842c (clean)

Command: git cherry-pick -x -m 1 848b3116b. The commit keeps its original author and records its source commit.

Conflicts

None. The code and test changes are identical to the diff of #3217 on main, hunk by hunk. Only the blob index hashes are different.

Release notes (changed in review)

As requested in review, the release note of #3217 is in history/0_11_0_release/CHANGELOG.md (end of ## Bug Fixes), with the same text. history/latest/3203.md is removed from this PR, because v0.11.0 has no history/latest/ directory. Commit c6d853990.

Test

Done in a devbox (JDK 17 + JDK 8 toolchain, ClickHouse server 26.9.1):

• mvn -B -pl client-v2,jdbc-v2 -am -DskipTests -Dmaven.javadoc.skip=true install: BUILD SUCCESS.
• jdbc-v2 ResultSetImplTest (integration): 20 run, 0 failures. This includes the 2 new testCloseNotFullyReadResultSet cases.
• client-v2 HttpTransportTests (integration): 96 run, 69 pass. The 3 new testClosingNotFullyReadResponse cases pass.
• The bug is present on v0.11.0: with only HttpAPIClientHelper.java reverted to v0.11.0, the 3 new testClosingNotFullyReadResponse cases fail.
• The other 27 HttpTransportTests failures are the same with and without this change. They are environment failures: the tests connect to the hardcoded localhost:8123 / localhost:8443, which are not reachable in this environment (the server has a different host name), and testServerErrorHandling expects error code 636 where this server returns 400. CI runs them against a local server.

This PR adds no tests, docs or version bump other than those in #3217, and the changelog entry above.

Checklist

• Issue linked: [0.11.0] Do not fail closing a partly read response that the server ended early #3217 #3231
• Only the changes of [client-v2, jdbc-v2] Do not fail closing a partly read response that the server ended early #3217 (release note moved to the 0.11.0 changelog); no conflicts

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes HTTP response teardown for failed mid-stream queries; behavior is narrowed (swallow close errors only when a body stream was opened) but affects connection reuse and error surfacing for all HTTP users.

Overview
Backports the fix for closing partly read HTTP query results when ClickHouse ends the response body early (e.g. query fails after rows were already streamed). Closing used to trigger Apache HttpClient’s body drain and could surface ConnectionClosedException / “Premature end of chunk coded message body” on QueryResponse#close(), ResultSet#close(), or reader close even though the real failure should appear while reading.

HTTP transport wraps the response body in ResponseBodyInputStream and treats ConnectionClosedException on close as DEBUG-only when a data stream was opened; errors while reading are unchanged. If the body was never opened (Client#execute / executeUpdate), close still propagates the failure so commands without a result set are not silently ignored.

Adds integration tests for client-v2 (several compression modes) and jdbc-v2 (RowBinaryWithNamesAndTypes, JSONEachRow), plus a v0.11.0 changelog entry for #3203.

^{Reviewed by Cursor Bugbot for commit c6d8539. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chernser]

@polyglotAI-bot sync from v0.11.0 - there is fix for build

[chernser]

sync with v0.11.0 to fix the CI build

[polyglotAI-bot]

@chernser Done. I merged v0.11.0 (fd98bda, #3236) into this branch: merge commit d60c9af, no conflicts.

• The PR diff against v0.11.0 is unchanged: 4 files, +154/−2, same as [client-v2, jdbc-v2] Do not fail closing a partly read response that the server ended early #3217.
• mvn validate on the full reactor passes locally.
• CI runs again on the new head.

[github-actions]

Triage

Category: bugfix • Risk: medium

Summary
Maintainer-requested backport of #3217 from main to the v0.11.0 release branch (asked by @chernser in #3231). It fixes #3203: closing a partly read query response (QueryResponse#close(), ResultSet#close(), reader close) could throw ConnectionClosedException / "Premature end of chunk coded message body" when the server ended the response body early after a mid-stream query failure. The sole source change is in client-v2's HttpAPIClientHelper: TransportResponseImpl#close() catches ConnectionClosedException and logs it at DEBUG — but only when a data stream was created; otherwise it still propagates so Client#execute/executeUpdate failures are not silently swallowed. A new ResponseBodyInputStream (FilterInputStream) applies the same logic on stream close. Read-path errors are unchanged. The backport is reported as conflict-free and hunk-identical to #3217, with the release note moved into history/0_11_0_release/CHANGELOG.md. Integration tests are added for client-v2 (3 compression combinations) and jdbc-v2 (2 formats).

What this impacts

• client-v2 HTTP transport (HttpAPIClientHelper): response/stream close behavior for all HTTP-based users — error surfacing on close and connection reuse after partially read responses.
• jdbc-v2 indirectly via ResultSet#close() behavior (tests only in this module).
• v0.11.0 release branch changelog.

Concerns

• Medium rule fired: behavioral change in a hot-path module (client-v2) — error-on-close semantics change for every HTTP response; a swallowed ConnectionClosedException is the trade-off, mitigated by the dataStreamCreated guard and new regression tests.
• Backport targets the v0.11.0 release branch, so correctness of the cherry-pick matters for the next patch release; author reports a clean cherry-pick (-x -m 1 848b3116b) identical to the already-merged [client-v2, jdbc-v2] Do not fail closing a partly read response that the server ended early #3217.
• CI check status could not be fetched during triage; verify required checks are green before merge.

Required reviewer action

• At least one human reviewer (medium risk); confirm the diff matches merged [client-v2, jdbc-v2] Do not fail closing a partly read response that the server ended early #3217 hunk-for-hunk and that the changelog placement is correct for the 0.11.0 release.

[chernser]

Update the https://github.com/ClickHouse/clickhouse-java/blob/v0.11.0/history/0_11_0_release/CHANGELOG.md with bug fix for this PR

[polyglotAI-bot]

@chernser Done in c6d8539.

• I added the bug fix entry to history/0_11_0_release/CHANGELOG.md, at the end of ## Bug Fixes. It has the same text as the release note of [client-v2, jdbc-v2] Do not fail closing a partly read response that the server ended early #3217.
• I removed history/latest/3203.md from this PR, because the entry is now in the changelog and v0.11.0 has no history/latest/ directory. Tell me if you want to keep the file.

The code and test changes are not changed. CI runs again on the new head.