Handle a zero-worker Databricks cluster in _init_nodes

user_tools/src/spark_rapids_pytools/cloud_api/databricks_aws.py

@@ -193,7 +193,8 @@ def _set_name_from_props(self) -> None:
     def _init_nodes(self):
         # assume that only one master node
         master_nodes_from_conf = self.props.get_value_silent('driver')
-        worker_nodes_from_conf = self.props.get_value_silent('executors')
+        # a zero-worker (single-node) cluster has no `executors` entry at all
+        worker_nodes_from_conf = self.props.get_value_silent('executors') or []
         num_workers = self.props.get_value_silent('num_workers')
         if num_workers is None and self.props.get_value_silent('autoscale') is not None:
             target_workers = self.props.get_value_silent('autoscale', 'target_workers')

user_tools/src/spark_rapids_pytools/cloud_api/databricks_azure.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2023-2025, NVIDIA CORPORATION.
+# Copyright (c) 2023-2026, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -244,7 +244,8 @@ def _set_name_from_props(self) -> None:
     def _init_nodes(self):
         # assume that only one driver node
         driver_nodes_from_conf = self.props.get_value_silent('driver')
-        worker_nodes_from_conf = self.props.get_value_silent('executors')
+        # a zero-worker (single-node) cluster has no `executors` entry at all
+        worker_nodes_from_conf = self.props.get_value_silent('executors') or []
         num_workers = self.props.get_value_silent('num_workers')
         if num_workers is None and self.props.get_value_silent('autoscale') is not None:
             target_workers = self.props.get_value_silent('autoscale', 'target_workers')

user_tools/tests/spark_rapids_tools_ut/test_cluster.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2023-2025, NVIDIA CORPORATION.
+# Copyright (c) 2023-2026, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,13 +14,22 @@
 
 """Test Identifying cluster from properties"""
 
+import json
+
 import pytest
 
+from spark_rapids_pytools.cloud_api.sp_types import CspEnv, get_platform
+from spark_rapids_pytools.common.prop_manager import JSONPropertiesContainer
 from spark_rapids_tools import CspPath
 from spark_rapids_tools.cloud import ClientCluster
 from spark_rapids_tools.exceptions import InvalidPropertiesSchema
 from .conftest import SparkRapidsToolsUT, all_cpu_cluster_props
 
+databricks_cluster_props = [
+    pytest.param(CspEnv.DATABRICKS_AWS, 'cluster/databricks/aws-cpu-00.json', id='databricks_aws'),
+    pytest.param(CspEnv.DATABRICKS_AZURE, 'cluster/databricks/azure-cpu-00.json', id='databricks_azure')
+]
+
 
 class TestClusterCSP(SparkRapidsToolsUT):  # pylint: disable=too-few-public-methods
     """
@@ -36,3 +45,32 @@ def test_cluster_invalid_path(self, get_ut_data_dir):
     def test_define_cluster_type_from_schema(self, csp, prop_path, get_ut_data_dir):
         client_cluster = ClientCluster(CspPath(f'{get_ut_data_dir}/{prop_path}'))
         assert client_cluster.platform_name == csp
+
+
+class TestDatabricksClusterWorkers(SparkRapidsToolsUT):  # pylint: disable=too-few-public-methods
+    """
+    Class testing how the Databricks platforms build the worker nodes of a cluster from its
+    `clusters get` properties when the `executors` entry is absent, which is what the API
+    returns for a zero-worker (single-node) cluster and for a terminated cluster
+    """
+    @staticmethod
+    def _load_cluster(csp_enum, prop_path, data_dir, num_workers):
+        with open(f'{data_dir}/{prop_path}', encoding='utf8') as prop_file:
+            props = json.load(prop_file)
+        props.pop('executors', None)
+        props['num_workers'] = num_workers
+        platform = get_platform(csp_enum)(ctxt_args={})
+        return platform.load_cluster_by_prop(JSONPropertiesContainer(prop_arg=props, file_load=False))
+
+    @pytest.mark.parametrize('csp_enum,prop_path', databricks_cluster_props)
+    def test_zero_worker_cluster_raises_no_workers_error(self, csp_enum, prop_path, get_ut_data_dir):
+        # a single-node cluster has num_workers 0 and no executors entry; the platform's own
+        # "no worker nodes" error is the expected answer, not a TypeError from iterating None
+        with pytest.raises(RuntimeError, match='The cluster has no worker nodes'):
+            self._load_cluster(csp_enum, prop_path, get_ut_data_dir, num_workers=0)
+
+    @pytest.mark.parametrize('csp_enum,prop_path', databricks_cluster_props)
+    def test_terminated_cluster_generates_workers(self, csp_enum, prop_path, get_ut_data_dir):
+        # a terminated multi-node cluster has no executors entry either; its workers are generated
+        cluster = self._load_cluster(csp_enum, prop_path, get_ut_data_dir, num_workers=2)
+        assert cluster.get_workers_count() == 2