Antalya-26.3 - Backport flaky-fix commits from upstream (2026-07-21)

tests/integration/test_rename_column/test.py

@@ -269,6 +269,37 @@ def rename_column_on_cluster(
             break
 
 
+def _num2_converged(nodes, table_name):
+    return all(
+        node.query(
+            "SELECT count() FROM system.columns "
+            "WHERE database = 'default' AND table IN ('{t}', '{t}_replicated') "
+            "AND name = 'num2'".format(t=table_name)
+        ).strip()
+        == "2"
+        for node in nodes
+    )
+
+
+def wait_for_rename_to_num2(nodes, table_name, attempts=12, poll_seconds=5):
+    # Best-effort cleanup renames + async ON CLUSTER replication can leave the
+    # column as foo2/foo3 on some replicas. Re-drive the idempotent rename-back
+    # until num2 is present on every node's Distributed and _replicated tables,
+    # so the strict queries below do not race schema convergence (UNKNOWN_IDENTIFIER).
+    tables = [table_name, "%s_replicated" % table_name]
+    for _ in range(attempts):
+        if _num2_converged(nodes, table_name):
+            return
+        for old_name in ("foo2", "foo3"):
+            for table in tables:
+                rename_column_on_cluster(nodes[0], table, old_name, "num2", 1, True)
+        time.sleep(poll_seconds)
+    if not _num2_converged(nodes, table_name):
+        raise Exception(
+            "columns did not converge to num2 for {} on all nodes".format(table_name)
+        )
+
+
 def alter_move(node, table_name, iterations=1, ignore_exception=False):
     i = 0
     while True:
@@ -809,14 +840,7 @@ def test_rename_distributed_parallel_insert_and_select(started_cluster):
         for task in tasks:
             task.get(timeout=240)
 
-        rename_column_on_cluster(node1, table_name, "foo2", "num2", 1, True)
-        rename_column_on_cluster(
-            node1, "%s_replicated" % table_name, "foo2", "num2", 1, True
-        )
-        rename_column_on_cluster(node1, table_name, "foo3", "num2", 1, True)
-        rename_column_on_cluster(
-            node1, "%s_replicated" % table_name, "foo3", "num2", 1, True
-        )
+        wait_for_rename_to_num2(nodes, table_name)
 
         insert(node1, table_name, 1000, col_names=["num", "num2"])
         select(node1, table_name, "num2")

tests/queries/0_stateless/01017_uniqCombined_memory_usage.sql

@@ -32,14 +32,14 @@ SELECT 'K=16';
 SELECT 'UInt32';
 SET max_memory_usage = 2000000;
 SELECT sum(u) FROM (SELECT intDiv(number, 4096) AS k, uniqCombined(16)(number % 4096) u FROM numbers(4096 * 100) GROUP BY k); -- { serverError MEMORY_LIMIT_EXCEEDED }
-SET max_memory_usage = 5230000;
+SET max_memory_usage = 6300000;
 SELECT sum(u) FROM (SELECT intDiv(number, 4096) AS k, uniqCombined(16)(number % 4096) u FROM numbers(4096 * 100) GROUP BY k);
 
 -- HashTable for UInt64 (used until (1<<11) elements), hence 2048 elements
 SELECT 'UInt64';
 SET max_memory_usage = 2000000;
 SELECT sum(u) FROM (SELECT intDiv(number, 2048) AS k, uniqCombined(16)(reinterpretAsString(number % 2048)) u FROM numbers(2048 * 100) GROUP BY k); -- { serverError MEMORY_LIMIT_EXCEEDED }
-SET max_memory_usage = 5900000;
+SET max_memory_usage = 7100000;
 SELECT sum(u) FROM (SELECT intDiv(number, 2048) AS k, uniqCombined(16)(reinterpretAsString(number % 2048)) u FROM numbers(2048 * 100) GROUP BY k);
 
 SELECT 'K=18';

tests/queries/0_stateless/02421_simple_queries_for_opentelemetry.sh

@@ -108,7 +108,15 @@ function check_tcp_attributes()
 
 function execute_query_HTTP()
 {
-    ${CLICKHOUSE_CURL} -sS "${CLICKHOUSE_URL}&database=${CLICKHOUSE_DATABASE}&query_id=$1" -d "$2"
+    # A traceparent header (with the sampled flag) forces the request to be traced, and the
+    # Referer / User-Agent headers populate the http.* attributes on the HTTPHandler span.
+    local trace_id
+    trace_id=$(${CLICKHOUSE_CLIENT} -q "SELECT lower(hex(generateUUIDv4()))")
+    ${CLICKHOUSE_CURL} -sS \
+        -H "traceparent: 00-${trace_id}-0000000000000010-01" \
+        -H "referer: some-referer" \
+        -H "user-agent: some-user-agent" \
+        "${CLICKHOUSE_URL}&database=${CLICKHOUSE_DATABASE}&query_id=$1" -d "$2"
 }
 
 function check_http_attributes()
@@ -119,32 +127,34 @@ function check_http_attributes()
   local agent="not found"
   local method="not found"
 
+  # The http.* attributes live on the HTTPHandler (SERVER) span, not on the child query span.
+  # Match it by the query_id embedded in the request URI (clickhouse.uri).
   result=$(${CLICKHOUSE_CLIENT} -q "
       SYSTEM FLUSH LOGS opentelemetry_span_log;
-      SELECT attribute['http.referer'],
-             attribute['http.user.agent'],
-             attribute['http.method']
+      SELECT attribute['http.referer']    AS referer,
+             attribute['http.user.agent'] AS user_agent,
+             attribute['http.method']     AS method
       FROM system.opentelemetry_span_log
       WHERE finish_date >= yesterday()
-      AND operation_name = 'query'
-      AND attribute['clickhouse.query_id'] = '${query_id}'
+      AND operation_name = 'HTTPHandler'
+      AND attribute['clickhouse.uri'] LIKE '%${query_id}%'
       FORMAT JSONEachRow;
     ")
 
   if [[ -z "$result" ]]; then
     echo "Error: No result returned from ClickHouse server"
     return 1
   fi
-  
-  if [[ $result == *"http.referer"* ]]; then
+
+  if [[ $result == *'"referer":"some-referer"'* ]]; then
     referer="present"
   fi
-  
-  if [[ $result == *"http.user.agent"* ]]; then
+
+  if [[ $result == *'"user_agent":"some-user-agent"'* ]]; then
     agent="present"
   fi
 
-  if [[ $result == *"http.method"* ]]; then
+  if [[ $result == *'"method":"POST"'* ]]; then
     method="present"
   fi
 
@@ -191,9 +201,8 @@ query_id=$(${CLICKHOUSE_CLIENT} -q "select generateUUIDv4()")
 execute_query $query_id 'select * from opentelemetry_test format Null'
 check_tcp_attributes $query_id
 
-# Test 7: Executes a TCP SELECT query and checks for http attributes in OpenTelemetry spans.
+# Test 7: Executes an HTTP SELECT query and checks for http attributes in OpenTelemetry spans.
 query_id=$(${CLICKHOUSE_CLIENT} -q "select generateUUIDv4()")
-execute_query "$query_id" 'set opentelemetry_start_trace_probability=1'
 execute_query_HTTP "$query_id" 'select * from opentelemetry_test FORMAT Null'
 check_http_attributes "$query_id"
 #