Antalya 26.8 - bump to v26.8.11.7-lts

.claude/tools/fetch_ci_report.js

@@ -574,30 +574,29 @@ async function getCIReportsFromPR(prUrl) {
 
   console.log(`Fetching CI reports for PR #${prNumber}...\n`);
 
-  // Fetch PR comments to find CI bot comment.
+  // Report URLs are the signal — do not filter by bot login.
+  // Altinity posts via github-actions[bot] with the virtual-hosted S3 URL;
+  // older comments used clickhouse-gh[bot] and path-style S3.
   // Drop GH_CONFIG_DIR before spawning gh: some agent/runner checkouts set it to a poisoned
   // config dir (no/expired auth) that makes `gh api` fail, while the default config is fine.
   // Other repo tooling (patch-release-check) does the same via `env -u GH_CONFIG_DIR gh`.
   const ghEnv = { ...process.env };
   delete ghEnv.GH_CONFIG_DIR;
   try {
-    const commentsJson = execSync(`gh api repos/Altinity/ClickHouse/issues/${prNumber}/comments --paginate --jq '.[] | select(.user.login == "clickhouse-gh[bot]") | {body, created_at}'`, {
+    const commentsJson = execSync(`gh api repos/Altinity/ClickHouse/issues/${prNumber}/comments --paginate --jq '.[] | {body, created_at}'`, {
       encoding: 'utf8',
       stdio: ['pipe', 'pipe', 'pipe'],
       env: ghEnv
     });
 
     const comments = commentsJson.trim().split('\n').filter(l => l.trim()).map(l => JSON.parse(l));
     comments.sort((a, b) => (b.created_at || '').localeCompare(a.created_at || ''));
-    if (!comments || comments.length === 0) {
-      throw new Error('No CI bot comment found');
-    }
 
-    // Search through all bot comments for CI report URLs (not just the latest). Exclude backtick and
+    // Search through all comments for CI report URLs (not just the latest). Exclude backtick and
     // quote chars so a URL quoted in markdown (e.g. inside the AI-review text) is not captured with
     // trailing junk, strip trailing punctuation, and dedupe -- otherwise the same report is fetched
-    // twice and the summary is doubled.
-    const reportUrlPattern = /https:\/\/s3\.amazonaws\.com\/altinity-build-artifacts\/json\.html\?[^\s)`'"]+/g;
+    // twice and the summary is doubled. Match both path-style and virtual-hosted S3 URLs.
+    const reportUrlPattern = /https:\/\/(?:s3\.amazonaws\.com\/altinity-build-artifacts|altinity-build-artifacts\.s3\.amazonaws\.com)\/json\.html\?[^\s)`'"]+/g;
     for (const comment of comments) {
       if (!comment.body) continue;
       let urls = comment.body.match(reportUrlPattern);
@@ -607,9 +606,9 @@ async function getCIReportsFromPR(prUrl) {
       }
     }
 
-    throw new Error('No CI report URLs found in bot comments');
+    throw new Error('No CI report URLs found in PR comments');
   } catch (error) {
-    if (error.message.includes('No CI bot comment found') || error.message.includes('No CI report URLs found')) {
+    if (error.message.includes('No CI report URLs found')) {
       throw error;
     }
     throw new Error(`Failed to fetch PR comments: ${error.message}`);

.github/actions/create_workflow_report/create_workflow_report.py

@@ -192,6 +192,9 @@ def _enrich_prs_in_release_merge_prs(df: pd.DataFrame, repo: str) -> pd.DataFram
             f"https://api.github.com/repos/{repo}/pulls/{pr_number}",
             headers=headers,
         )
+        if response.status_code == 404:
+            # NOTE (strtgbb): not in this repo — upstream PR merged from a fork
+            continue
         if response.status_code != 200:
             raise Exception(
                 f"Failed to fetch pull request info: {response.status_code} {response.text}"
@@ -207,6 +210,8 @@ def _enrich_prs_in_release_merge_prs(df: pd.DataFrame, repo: str) -> pd.DataFram
                 "pr_labels": html.escape(", ".join(sorted(label_names)), quote=True),
             }
         )
+    if not rows:
+        return pd.DataFrame(columns=["pr_number", "pr_name", "pr_labels"])
     return pd.DataFrame(rows)
 
 

.gitmodules

@@ -329,7 +329,7 @@
 	url = https://github.com/ClickHouse/double-conversion.git
 [submodule "contrib/mongo-cxx-driver"]
 	path = contrib/mongo-cxx-driver
-	url = https://github.com/mongodb/mongo-cxx-driver.git
+	url = https://github.com/ClickHouse/mongo-cxx-driver.git
 [submodule "contrib/mongo-c-driver"]
 	path = contrib/mongo-c-driver
 	url = https://github.com/ClickHouse/mongo-c-driver.git

base/poco/NetSSL_OpenSSL/src/SSLManager.cpp

@@ -340,7 +340,9 @@ void SSLManager::initDefaultContext(bool server)
 	{
 		_ptrDefaultClientContext->enableSessionCache(cacheSessions);
 	}
-	bool extendedVerification = config.getBool(prefix + CFG_EXTENDED_VERIFICATION, false);
+	/// Only an outbound connection has a requested host name to check: on an accepted socket this
+	/// would run against the client's address, which a client certificate does not name.
+	bool extendedVerification = config.getBool(prefix + CFG_EXTENDED_VERIFICATION, !server);
 	if (server)
 		_ptrDefaultServerContext->enableExtendedCertificateVerification(extendedVerification);
 	else

ci/jobs/integration_test_job.py

@@ -1329,7 +1329,7 @@ def main():
     # hard subprocess backstop). Used below to keep an empty flaky/targeted result a
     # best-effort SKIPPED only when a timeout actually exhausted the budget.
     timed_out = False
-    session_timeout_parallel = 3600 * 2
+    session_timeout_parallel = 3600 * 2.5
     session_timeout_sequential = 3600
 
     if is_llvm_coverage:

ci/jobs/scripts/check_style/experimental_settings_ignore.txt

@@ -87,3 +87,5 @@ enable_vector_similarity_index
 allow_nullable_tuple_in_extracted_subcolumns
 allow_lossy_numeric_supertype
 allow_metadata_only_named_tuple_alter
+allow_experimental_cleanup_old_data_files_compaction
+allow_experimental_iceberg_compaction

ci/jobs/scripts/check_style/various_checks.sh

@@ -210,6 +210,11 @@ tests_with_system_drop=( $(
         sort -u
 ) )
 for test_case in "${tests_with_system_drop[@]}"; do
+    # `SYSTEM DROP FILESYSTEM CACHE '<name>'` affects only the named cache, not the
+    # server-wide state, so a test which drops the cache it created itself can run in
+    # parallel. Skip a test if every `SYSTEM DROP` in it drops a cache by name.
+    grep -oiP "system\s+drop(\s+filesystem\s+cache\s+'[^']+')?" "$test_case" |
+        grep -qivP "filesystem\s+cache\s+'" || continue
     grep -qP '(--|#)\s*[Tt]ags:.*no-parallel' "$test_case" || echo "Test with SYSTEM DROP should have no-parallel tag: $test_case"
 done
 

ci/jobs/scripts/stress/stress.py

@@ -400,7 +400,13 @@ def install_thread_pool_fault_injection() -> None:
 
     logging.info("Installing thread-pool fault-injection config: %s -> %s", src, dst)
     subprocess.run(["ln", "-sf", src, dst], check=True)
-    if not call_with_retry(make_query_command("SYSTEM RELOAD CONFIG"), timeout=30, retry_count=5):
+    # NOTE (strtgbb): ARM debug + ThreadFuzzer config reload is ~18-23s; the
+    # default 15s receive_timeout loses the race and fails the job.
+    if not call_with_retry(
+        make_query_command("SYSTEM RELOAD CONFIG", receive_timeout=60),
+        timeout=90,
+        retry_count=5,
+    ):
         # Fail-close before the verify query: a stale non-zero probability left
         # over from an earlier reload would otherwise mask the reload failure.
         raise RuntimeError(
@@ -655,9 +661,9 @@ def execute_bash(full_command, timeout=120):
         raise
 
 
-def make_query_command(query: str) -> str:
+def make_query_command(query: str, receive_timeout: int = 15) -> str:
     return (
-        f'clickhouse client -q "{query}" --receive_timeout=15 --max_untracked_memory=1Gi '
+        f'clickhouse client -q "{query}" --receive_timeout={receive_timeout} --max_untracked_memory=1Gi '
         "--memory_profiler_step=1Gi --max_memory_usage_for_user=0 --max_memory_usage_in_client=1000000000 "
         "--enable-progress-table-toggle=0 "
         "--ast_fuzzer_runs=0",

ci/praktika/native_jobs.py

@@ -292,6 +292,8 @@ def _prepare_submodule_cache(workflow, workflow_config: RunConfig) -> Result:
                 no_strict=True,
             )
             Shell.check(f"rm -f {archive_path}")
+            if not created and not S3.head_object(s3_path):
+                raise RuntimeError(f"failed to upload submodule cache {s3_path}")
             info = (
                 f"cache miss, created: {cache_hash}"
                 if created
@@ -302,10 +304,12 @@ def _prepare_submodule_cache(workflow, workflow_config: RunConfig) -> Result:
         workflow_config.dump()
         status = Result.Status.OK
     except Exception as e:
-        print(f"WARNING: Submodule cache failed: {e}")
+        print(f"ERROR: Submodule cache failed: {e}")
         traceback.print_exc()
         info = f"{e}\n{traceback.format_exc()}"
-        status = Result.Status.OK  # non-fatal, jobs fall back to GitHub clone
+        # Do not continue with an empty submodule_cache_hash. Builds would
+        # skip the restore and clone the same pins themselves.
+        status = Result.Status.FAIL
 
     return Result.create_from(
         name="Submodule Cache",

ci/settings/altinity_overrides.py

@@ -59,6 +59,10 @@ class RunnerLabels:
 
 INSTALL_PYTHON_REQS_FOR_NATIVE_JOBS = ""
 
+# NOTE (strtgbb): anonymous submodule fetches of non-tip pins get refused
+# after the burst of ~150 clones; send the ambient gh token instead.
+ENABLE_SUBMODULE_CLONE_AUTH = True
+
 DISABLED_WORKFLOWS = [
     "backport_branches.py",
     "custom_build_praktika.py",

cmake/autogenerated_versions.txt

@@ -2,13 +2,13 @@
 
 # NOTE: VERSION_REVISION has nothing common with DBMS_TCP_PROTOCOL_VERSION,
 # only DBMS_TCP_PROTOCOL_VERSION should be incremented on protocol changes.
-SET(VERSION_REVISION 54518)
+SET(VERSION_REVISION 54523)
 SET(VERSION_MAJOR 26)
 SET(VERSION_MINOR 8)
-SET(VERSION_PATCH 6)
-SET(VERSION_GITHASH ec5605431dacfc812affc406cc81ca398ca68174)
-SET(VERSION_DESCRIBE v26.8.6.10001.altinitytest)
-SET(VERSION_STRING 26.8.6.10001.altinitytest)
+SET(VERSION_PATCH 11)
+SET(VERSION_GITHASH bedf2ab54b8a0c34afbf2d907eaf933324f43cd6)
+SET(VERSION_DESCRIBE v26.8.11.10001.altinitytest)
+SET(VERSION_STRING 26.8.11.10001.altinitytest)
 # end of autochange
 
 SET(VERSION_TWEAK 10001)

contrib/mongo-c-driver-cmake/CMakeLists.txt

@@ -5,13 +5,13 @@ if(NOT USE_MONGODB)
 endif()
 
 set(libbson_VERSION_MAJOR 2)
-set(libbson_VERSION_MINOR 3)
-set(libbson_VERSION_PATCH 0)
-set(libbson_VERSION 2.3.0)
+set(libbson_VERSION_MINOR 5)
+set(libbson_VERSION_PATCH 4)
+set(libbson_VERSION 2.5.4)
 set(libmongoc_VERSION_MAJOR 2)
-set(libmongoc_VERSION_MINOR 3)
-set(libmongoc_VERSION_PATCH 0)
-set(libmongoc_VERSION 2.3.0)
+set(libmongoc_VERSION_MINOR 5)
+set(libmongoc_VERSION_PATCH 4)
+set(libmongoc_VERSION 2.5.4)
 
 set(LIBBSON_SOURCES_ROOT "${ClickHouse_SOURCE_DIR}/contrib/mongo-c-driver/src")
 set(LIBBSON_SOURCE_DIR "${LIBBSON_SOURCES_ROOT}/libbson/src")
@@ -135,7 +135,7 @@ endif()
 set(LIBMONGOC_SOURCE_DIR "${LIBBSON_SOURCES_ROOT}/libmongoc/src")
 set(LIBMONGOC_BINARY_DIR "${LIBBSON_BINARY_ROOT}/libmongoc/src")
 file(GLOB_RECURSE LIBMONGOC_SOURCES "${LIBMONGOC_SOURCE_DIR}/*.c")
-set(UTF8PROC_SOURCE_DIR "${LIBBSON_SOURCES_ROOT}/utf8proc-2.8.0")
+set(UTF8PROC_SOURCE_DIR "${LIBBSON_SOURCES_ROOT}/utf8proc-2.11.3")
 set(UTF8PROC_SOURCES "${UTF8PROC_SOURCE_DIR}/utf8proc.c")
 set(UTHASH_SOURCE_DIR "${LIBBSON_SOURCES_ROOT}/uthash")
 

contrib/mongo-cxx-driver-cmake/CMakeLists.txt

@@ -54,10 +54,17 @@ set(BSONCXX_SOURCES
 
 set(BSONCXX_POLY_USE_IMPLS ON)
 
+# ClickHouse uses the `v_noabi` API, whose symbols are gated behind the "unstable ABI" switch since r4.4.0.
+set(BSONCXX_ENABLE_UNSTABLE_ABI ON)
+
 configure_file(
         ${BSONCXX_SOURCES_DIR}/lib/bsoncxx/v1/config/config.hpp.in
         ${BSONCXX_BINARY_DIR}/lib/bsoncxx/v1/config/config.hpp
 )
+configure_file(
+        ${BSONCXX_SOURCES_DIR}/lib/bsoncxx/v_noabi/bsoncxx/config/config.hpp.in
+        ${BSONCXX_BINARY_DIR}/lib/bsoncxx/v_noabi/bsoncxx/config/config.hpp
+)
 configure_file(
         ${BSONCXX_SOURCES_DIR}/lib/bsoncxx/v1/config/version.hpp.in
         ${BSONCXX_BINARY_DIR}/lib/bsoncxx/v1/config/version.hpp
@@ -152,6 +159,7 @@ set(MONGOCXX_SOURCES
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/bulk_write.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/change_stream.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/client.cpp
+        ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/client_bulk_write.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/client_encryption.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/client_session.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/collection.cpp
@@ -203,6 +211,9 @@ set(MONGOCXX_SOURCES
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/insert_one_result.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/instance.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/logger.cpp
+        ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/oidc_callback.cpp
+        ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/oidc_callback_params.cpp
+        ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/oidc_credential.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/pipeline.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/pool.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/range_options.cpp
@@ -216,6 +227,7 @@ set(MONGOCXX_SOURCES
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/search_indexes.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/server_api.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/server_error.cpp
+        ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/string_options.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/text_options.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/tls.cpp
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/transaction_options.cpp
@@ -336,10 +348,17 @@ endif()
 set(BSONCXX_STATIC 1)
 set(MONGOCXX_STATIC 1)
 
+# ClickHouse uses the `v_noabi` API, whose symbols are gated behind the "unstable ABI" switch since r4.4.0.
+set(MONGOCXX_ENABLE_UNSTABLE_ABI ON)
+
 configure_file(
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/config/config.hpp.in
         ${MONGOCXX_BINARY_DIR}/lib/mongocxx/v1/config/config.hpp
 )
+configure_file(
+        ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v_noabi/mongocxx/config/config.hpp.in
+        ${MONGOCXX_BINARY_DIR}/lib/mongocxx/v_noabi/mongocxx/config/config.hpp
+)
 configure_file(
         ${MONGOCXX_SOURCES_DIR}/lib/mongocxx/v1/config/version.hpp.in
         ${MONGOCXX_BINARY_DIR}/lib/mongocxx/v1/config/version.hpp

docs/reference/interfaces/specs/NativeFormat.mdx

@@ -196,7 +196,14 @@ The byte values differ from the compact codes: `REPLICATED` is `0x03` in this ne
 
 The `count` is the full stack length **including the leading `DEFAULT` entry** that begins every stack. The compact codes already cover every one- and two-entry stack, so a `COMBINATION` always has a `count` of at least three.
 
-**Recursive `kind_stack` for `Tuple` columns.** The `kind_stack` payload above is the byte (or `COMBINATION` sequence) for one column's own serialization info. A `Tuple` carries a `SerializationInfoTuple`, which first writes the tuple's *own* kind-stack payload and then writes one full kind-stack payload for *each* element, in order; a decoder reads back the same recursive structure. So for `Tuple(A, B, C)` the field-4 bytes are `[tuple_kind][A_kind][B_kind][C_kind]`, and each element payload is itself recursive if that element is again a composite. The `has_custom_serialization` byte (field 3) is set whenever the tuple's own info *or any element's* info is non-default, so a `Tuple` whose only special element is sparse, replicated, or detached still triggers the kind-stack payload. A decoder that reads only the single leading enum byte for a `Tuple` will stop too early and misread the remaining element-kind bytes as column data.
+**Kind stack validity.** A decoder rejects a malformed stack with `INCORRECT_DATA` rather than building the serialization it describes:
+
+- the stack starts with `DEFAULT`;
+- the stack is a subsequence of `DEFAULT`, `SPARSE`, `REPLICATED`, `DETACHED` — the order in which the kinds wrap each other. `SPARSE` can sit inside `REPLICATED` but not the other way round, and `DETACHED` is always last because a `ColumnBLOB` holds the serialized form of everything below it. This also means no kind occurs twice, which bounds `count` by the number of kinds.
+
+Note that this order is not the order of the kind byte values above. Any other stack describes a column layout no writer builds and that no materialization step unwraps back to a full column of the declared type — a `ColumnSparse` whose values column is itself sparse or replicated, for instance, stays wrapped.
+
+**Recursive `kind_stack` for `Tuple` columns.** The `kind_stack` payload above is the byte (or `COMBINATION` sequence) for one column's own serialization info. A `Tuple` carries a `SerializationInfoTuple`, which first writes the tuple's *own* kind-stack payload and then writes one full kind-stack payload for *each* element, in order; a decoder reads back the same recursive structure. So for `Tuple(A, B, C)` the field-4 bytes are `[tuple_kind][A_kind][B_kind][C_kind]`, and each element payload is itself recursive if that element is again a composite. The `has_custom_serialization` byte (field 3) is set whenever the tuple's own info *or any element's* info is non-default, so a `Tuple` whose only special element is sparse or replicated still triggers the kind-stack payload. A decoder that reads only the single leading enum byte for a `Tuple` will stop too early and misread the remaining element-kind bytes as column data.
 
 **Sparse wire format.** When `kind_stack = 0x01`, the column `data` is two streams written back-to-back in the single shared TCP stream:
 
@@ -230,6 +237,8 @@ A decoder reconstructs a dense column by selecting `elements[indexes[i]]` for ea
 
 If the wrapped column was sparse, its stack is `{DEFAULT, SPARSE, DETACHED}`, which serializes as `DETACHED_OVER_SPARSE`. A client decoding such a column reads the blob length and bytes, then decompresses the blob to recover the inner column payload (see the [`ColumnBLOB` note](#compression-negotiation) under compression).
 
+`DETACHED` is directional: it may appear only in result blocks sent by a server, and only for a whole top-level column — never as one of the element kinds of a `Tuple`. Unlike the other kinds, it makes the runtime column a `ColumnBLOB` instead of the column of the declared type, so a receiver accepts it only when its own pipeline converts the blob back. A server therefore rejects `DETACHED` in any block a client sends — external table data, scalars and insert data alike — with `INCORRECT_DATA`.
+
 ### Block variants {#block-variants}
 
 All Data-family packets share the same Block wire format. The variants differ only in their column and row counts: