fix(LoaderUtilsMixin): table_enabled was allways True as this feature is enabled

Enabling/disabling tablets was changed. Now we have to check scylla yaml
or keystore schema.
For more info see
- scylladb/scylladb#21451
- scylladb/scylladb#21614

(cherry picked from commit b03a690)

Author: jsmolar

sdcm/cluster.py

@@ -5044,10 +5044,9 @@ def get_node_ip_list(verification_node):
         self.test_config.tester_obj().monitors.reconfigure_scylla_monitoring()
 
     def decommission(self, node: BaseNode, timeout: int | float = None) -> DataCenterTopologyRfControl | None:
-        with node.parent_cluster.cql_connection_patient(node) as session:
-            if tablets_enabled := is_tablets_feature_enabled(session):
-                dc_topology_rf_change = DataCenterTopologyRfControl(target_node=node)
-                dc_topology_rf_change.decrease_keyspaces_rf()
+        if tablets_enabled := is_tablets_feature_enabled(node):
+            dc_topology_rf_change = DataCenterTopologyRfControl(target_node=node)
+            dc_topology_rf_change.decrease_keyspaces_rf()
         with adaptive_timeout(operation=Operations.DECOMMISSION, node=node):
             node.run_nodetool("decommission", timeout=timeout, long_running=True, retry=0)
         self.verify_decommission(node)

sdcm/fill_db_data.py

@@ -3192,8 +3192,7 @@ def is_counter_supported(self) -> bool:
     @property
     def tablets_enabled(self) -> bool:
         """Check is tablets enabled on cluster"""
-        with self.db_cluster.cql_connection_patient(self.db_cluster.nodes[0]) as session:
-            return is_tablets_feature_enabled(session)
+        return is_tablets_feature_enabled(self.db_cluster.nodes[0])
 
     @retrying(n=3, sleep_time=20, allowed_exceptions=ProtocolException)
     def truncate_table(self, session, truncate):  # pylint: disable=no-self-use

sdcm/nemesis.py

@@ -1008,10 +1008,9 @@ def disrupt_restart_with_resharding(self):
                 "Run 'disrupt_nodetool_flush_and_reshard_on_kubernetes' instead")
 
         # If tablets in use, skipping resharding since it is not supported.
-        with self.cluster.cql_connection_patient(self.target_node) as session:
-            if is_tablets_feature_enabled(session=session):
-                if SkipPerIssues('https://github.com/scylladb/scylladb/issues/16739', params=self.tester.params):
-                    raise UnsupportedNemesis('https://github.com/scylladb/scylladb/issues/16739')
+        if is_tablets_feature_enabled(self.target_node):
+            if SkipPerIssues('https://github.com/scylladb/scylladb/issues/16739', params=self.tester.params):
+                raise UnsupportedNemesis('https://github.com/scylladb/scylladb/issues/16739')
 
         murmur3_partitioner_ignore_msb_bits = 15  # pylint: disable=invalid-name
         self.log.info(f'Restart node with resharding. New murmur3_partitioner_ignore_msb_bits value: '

sdcm/tester.py

@@ -799,9 +799,12 @@ def reliable_replication_factor(self) -> int:
         """
         n_db_nodes = str(self.params.get('n_db_nodes'))
         min_nodes_dc = min([int(nodes_num) for nodes_num in n_db_nodes.split() if int(nodes_num) > 0])
-        with self.db_cluster.cql_connection_patient(self.db_cluster.nodes[0]) as session:
-            # In case tablets are enabled, it's better to set RF smaller than dc-nodes-number, so decommission is allowed.
-            return max([min_nodes_dc - 1, 1]) if is_tablets_feature_enabled(session) else min_nodes_dc
+
+        # In case tablets are enabled, it's better to set RF smaller than dc-nodes-number, so decommission is allowed.
+        rf_candidate = max([min_nodes_dc - 1, 1]
+                           ) if is_tablets_feature_enabled(self.db_cluster.nodes[0]) else min_nodes_dc
+        # NOTE: use RF=3 at max to avoid problems on big setups
+        return min(rf_candidate, 3)
 
     @property
     def test_id(self):

sdcm/utils/features.py

@@ -16,7 +16,6 @@
 
 CONSISTENT_TOPOLOGY_CHANGES_FEATURE = "SUPPORTS_CONSISTENT_TOPOLOGY_CHANGES"
 CONSISTENT_CLUSTER_MANAGEMENT_FEATURE = "SUPPORTS_RAFT_CLUSTER_MANAGEMENT"
-TABLETS_FEATURE = "TABLETS"
 
 LOGGER = logging.getLogger(__name__)
 
@@ -75,8 +74,14 @@ def is_consistent_topology_changes_feature_enabled(session: Session) -> bool:
     return CONSISTENT_TOPOLOGY_CHANGES_FEATURE in get_enabled_features(session)
 
 
-def is_tablets_feature_enabled(session: Session) -> bool:
+def is_tablets_feature_enabled(node) -> bool:
     """ Check whether tablets enabled
-    if you need from a specific node use `patient_exclusive_cql_connection` session
     """
-    return TABLETS_FEATURE in get_enabled_features(session)
+    with node.remote_scylla_yaml() as scylla_yaml:
+        # for backward compatibility of 2024.1 and earlier
+        if "tablets" in scylla_yaml.experimental_features:
+            return True
+        if scylla_yaml.dict().get("enable_tablets"):
+            return True
+
+    return False

sdcm/utils/loader_utils.py

@@ -15,9 +15,11 @@
 import os
 import re
 import time
+from functools import cached_property
 
 from sdcm.sct_events import Severity
 from sdcm.sct_events.system import TestFrameworkEvent
+from sdcm.utils.features import is_tablets_feature_enabled
 
 DEFAULT_USER = "cassandra"
 DEFAULT_USER_PASSWORD = "cassandra"
@@ -88,6 +90,11 @@ def assemble_and_run_all_stress_cmd_by_ks_names(self, stress_queue, stress_cmd,
             }
             self._run_all_stress_cmds(stress_queue, params)
 
+    @cached_property
+    def tablets_enabled(self):
+        # is tablets feature enabled in Scylla configuration.
+        return is_tablets_feature_enabled(self.db_cluster.nodes[0])
+
     def _run_all_stress_cmds(self, stress_queue, params):
         stress_cmds = params['stress_cmd']
         if not isinstance(stress_cmds, list):

sdcm/utils/tablets/common.py

@@ -34,10 +34,9 @@ def wait_for_tablets_balanced(node):
     if not node.raft.is_enabled:
         LOGGER.info("Raft is disabled, skipping wait for balance")
         return
-    with node.parent_cluster.cql_connection_patient(node=node) as session:
-        if not is_tablets_feature_enabled(session):
-            LOGGER.info("Tablets are disabled, skipping wait for balance")
-            return
+    if not is_tablets_feature_enabled(node):
+        LOGGER.info("Tablets are disabled, skipping wait for balance")
+        return
     time.sleep(60)  # one minute gap before checking, just to give some time to the state machine
     client = RemoteCurlClient(host="127.0.0.1:10000", endpoint="", node=node)
     LOGGER.info("Waiting for tablets to be balanced")

sla_per_user_system_test.py

@@ -22,6 +22,7 @@
 from sdcm.sct_events import Severity
 from sdcm.sct_events.workload_prioritisation import WorkloadPrioritisationEvent
 from test_lib.sla import ServiceLevel, Role, User
+from sdcm.utils.features import is_tablets_feature_enabled
 
 
 # pylint: disable=too-many-public-methods
@@ -371,6 +372,14 @@ def test_read_throughput_1to5_ratio(self):
 
     def _two_users_load_througput_workload(self, shares, load, expected_shares_ratio=None):
         session = self.prepare_schema()
+
+        if is_tablets_feature_enabled(self.db_cluster.nodes[0]):
+            self.run_pre_create_keyspace()
+            # after several test runs with Tablets decided to decrease by half of the percent(usually tests show about 96.8 - 97.5)
+            # due to unbalanced shards utilization with tablets(particular tablet belong to particular shard)
+            # during gauss distribution read
+            self.MIN_CPU_UTILIZATION = 96.5
+
         self.create_test_data_and_wait_no_compaction()
 
         # Define Service Levels/Roles/Users