[Autoscaler][Local Node Provider] Log a warning if max_workers < len(worker_ips) (#24635)

Logs a warning when a user sets max_workers for local node provider less than the number of available ips.

Also removes defaults of 0 for min_workers and max_workers from example configs to help prevent users inadvertantly setting max_workers=0 again.

python/ray/autoscaler/_private/local/config.py

@@ -100,6 +100,13 @@ def prepare_manual(config: Dict[str, Any]) -> Dict[str, Any]:
     else:
         node_type["max_workers"] = max_workers
 
+    if max_workers < num_ips:
+        cli_logger.warning(
+            f"The value of `max_workers` supplied ({max_workers}) is less"
+            f" than the number of available worker ips ({num_ips})."
+            f" At most {max_workers} Ray worker nodes will connect to the cluster."
+        )
+
     return config
 
 

python/ray/autoscaler/local/development-example.yaml

@@ -1,11 +1,7 @@
 cluster_name: default
-min_workers: 0
-max_workers: 0
 docker:
     image: ""
     container_name: ""
-upscaling_speed: 1.0
-idle_timeout_minutes: 5
 provider:
     type: local
     head_ip: YOUR_HEAD_NODE_HOSTNAME

python/ray/autoscaler/local/example-full.yaml

@@ -45,12 +45,14 @@ auth:
 # The minimum number of workers nodes to launch in addition to the head
 # node. This number should be >= 0.
 # Typically, min_workers == max_workers == len(worker_ips).
-min_workers: 0
+# This field is optional.
+min_workers: TYPICALLY_THE_NUMBER_OF_WORKER_IPS
 
 # The maximum number of workers nodes to launch in addition to the head node.
 # This takes precedence over min_workers.
 # Typically, min_workers == max_workers == len(worker_ips).
-max_workers: 0
+# This field is optional.
+max_workers: TYPICALLY_THE_NUMBER_OF_WORKER_IPS
 # The default behavior for manually managed clusters is
 # min_workers == max_workers == len(worker_ips),
 # meaning that Ray is started on all available nodes of the cluster.

python/ray/tests/test_autoscaler_yaml.py

@@ -1,5 +1,6 @@
 import jsonschema
 import logging
+import mock
 import os
 import sys
 import tempfile
@@ -262,6 +263,33 @@ class AutoscalingConfigTest(unittest.TestCase):
         # Check that worker config numbers were clipped to 3.
         assert prepared_config == expected_prepared
 
+        not_enough_workers_config = copy.deepcopy(base_config)
+
+        # Max workers is less than than the three available ips.
+        # The user is probably has probably made an error. Make sure we log a warning.
+        not_enough_workers_config["max_workers"] = 0
+        not_enough_workers_config["min_workers"] = 0
+        with mock.patch(
+            "ray.autoscaler._private.local.config.cli_logger.warning"
+        ) as warning:
+            prepared_config = prepare_config(not_enough_workers_config)
+            warning.assert_called_with(
+                "The value of `max_workers` supplied (0) is less"
+                " than the number of available worker ips (3)."
+                " At most 0 Ray worker nodes will connect to the cluster."
+            )
+        expected_prepared = yaml.safe_load(EXPECTED_LOCAL_CONFIG_STR)
+        # We logged a warning.
+        # However, prepare_config does not repair the strange config setting:
+        expected_prepared["max_workers"] = 0
+        expected_prepared["available_node_types"]["local.cluster.node"][
+            "max_workers"
+        ] = 0
+        expected_prepared["available_node_types"]["local.cluster.node"][
+            "min_workers"
+        ] = 0
+        assert prepared_config == expected_prepared
+
     def testValidateNetworkConfig(self):
         web_yaml = (
             "https://raw.githubusercontent.com/ray-project/ray/"