Revert "[KubeRay][Autoscaler][Core] Add a flag to disable ray status version check (#26584)" (#26597)

Reverts #26584

Seems it breaking test_advanced_4

python/ray/_private/gcs_utils.py

@@ -1,7 +1,6 @@
 import enum
 import logging
 import time
-import traceback
 from functools import wraps
 from typing import List, Optional
 
@@ -97,15 +96,13 @@ def create_gcs_channel(address: str, aio=False):
     return init_grpc_channel(address, options=_GRPC_OPTIONS, asynchronous=aio)
 
 
-def check_health(address: str, timeout=2, skip_version_check=False) -> bool:
+def check_health(address: str, timeout=2) -> bool:
     """Checks Ray cluster health, before / without actually connecting to the
     cluster via ray.init().
 
     Args:
         address: Ray cluster / GCS address string, e.g. ip:port.
         timeout: request timeout.
-        skip_version_check: If True, will skip comparision of GCS Ray version with local
-            Ray version. If False (default), will raise exception on mismatch.
     Returns:
         Returns True if the cluster is running and has matching Ray version.
         Returns False if no service is running.
@@ -117,14 +114,9 @@ def check_health(address: str, timeout=2, skip_version_check=False) -> bool:
         stub = gcs_service_pb2_grpc.HeartbeatInfoGcsServiceStub(channel)
         resp = stub.CheckAlive(req, timeout=timeout)
     except grpc.RpcError:
-        traceback.print_exc()
         return False
     if resp.status.code != GcsCode.OK:
         raise RuntimeError(f"GCS running at {address} is unhealthy: {resp.status}")
-
-    if skip_version_check:
-        return True
-    # Otherwise, continue to check for Ray version match.
     if resp.ray_version is None:
         resp.ray_version = "<= 1.12"
     if resp.ray_version != ray.__version__:

python/ray/autoscaler/_private/kuberay/run_autoscaler.py

@@ -21,17 +21,7 @@ def run_kuberay_autoscaler(cluster_name: str, cluster_namespace: str):
     ray_address = f"{head_ip}:6379"
     while True:
         try:
-            # Autoscaler Ray version might not exactly match GCS version, so skip the
-            # version check when checking GCS status.
-            subprocess.check_call(
-                [
-                    "ray",
-                    "health-check",
-                    "--address",
-                    ray_address,
-                    "--skip-version-check",
-                ]
-            )
+            subprocess.check_call(["ray", "health-check", "--address", ray_address])
             # Logging is not ready yet. Print to stdout for now.
             print("The Ray head is ready. Starting the autoscaler.")
             break

python/ray/scripts/scripts.py

@@ -6,7 +6,6 @@ import signal
 import subprocess
 import sys
 import time
-import traceback
 import urllib
 import urllib.parse
 from datetime import datetime
@@ -2324,13 +2323,7 @@ def kuberay_autoscaler(cluster_name: str, cluster_namespace: str) -> None:
     help="Health check for a specific component. Currently supports: "
     "[ray_client_server]",
 )
-@click.option(
-    "--skip-version-check",
-    is_flag=True,
-    default=False,
-    help="Skip comparison of GCS version with local Ray version.",
-)
-def healthcheck(address, redis_password, component, skip_version_check):
+def healthcheck(address, redis_password, component):
     """
     This is NOT a public api.
 
@@ -2341,12 +2334,9 @@ def healthcheck(address, redis_password, component, skip_version_check):
 
     if not component:
         try:
-            if ray._private.gcs_utils.check_health(
-                address, skip_version_check=skip_version_check
-            ):
+            if ray._private.gcs_utils.check_health(address):
                 sys.exit(0)
         except Exception:
-            traceback.print_exc()
             pass
         sys.exit(1)
 

python/ray/tests/test_advanced_4.py

@@ -1,4 +1,3 @@
-import mock
 import subprocess
 import sys
 
@@ -113,15 +112,6 @@ def test_check_health(shutdown_only):
     assert check_health(addr)
 
 
-def test_check_health_version_check():
-    with mock.patch("ray.__version__", "FOO-VERSION"):
-        conn = ray.init()
-        addr = conn.address_info["address"]
-        assert check_health(addr, skip_version_check=True)
-        with pytest.raises(RuntimeError):
-            check_health(addr)
-
-
 def test_back_pressure(shutdown_only_with_initialization_check):
     ray.init()