[Core] Bind runtime env agent and dashboard agent server to specified ip instead of 0.0.0.0

cpp/test_submit_cpp_job.py

@@ -21,9 +21,7 @@ def headers():
 
 @pytest.fixture(scope="module")
 def job_sdk_client(headers):
-    with _ray_start(
-        include_dashboard=True, num_cpus=1, _node_ip_address="0.0.0.0"
-    ) as ctx:
+    with _ray_start(include_dashboard=True, num_cpus=1) as ctx:
         address = ctx.address_info["webui_url"]
         assert wait_until_server_available(address)
         yield JobSubmissionClient(format_web_url(address), headers=headers)

python/ray/_private/runtime_env/agent/main.py

@@ -218,13 +218,10 @@ def parent_dead_callback(msg):
         check_raylet_task = create_check_raylet_task(
             args.log_dir, gcs_client, parent_dead_callback, loop
         )
-    runtime_env_agent_ip = (
-        "127.0.0.1" if args.node_ip_address == "127.0.0.1" else "0.0.0.0"
-    )
     try:
         web.run_app(
             app,
-            host=runtime_env_agent_ip,
+            host=args.node_ip_address,
             port=args.runtime_env_agent_port,
             loop=loop,
         )

python/ray/dashboard/agent.py

@@ -111,11 +111,14 @@ def _init_non_minimal(self):
                 ),
             )  # noqa
         )
-        grpc_ip = "127.0.0.1" if self.ip == "127.0.0.1" else "0.0.0.0"
         try:
             self.grpc_port = add_port_to_grpc_server(
-                self.server, build_address(grpc_ip, self.dashboard_agent_port)
+                self.server, build_address(self.ip, self.dashboard_agent_port)
             )
+            if self.ip != "127.0.0.1" and self.ip != "localhost":
+                self.grpc_port = add_port_to_grpc_server(
+                    self.server, f"127.0.0.1:{self.dashboard_agent_port}"
+                )
         except Exception:
             # TODO(SongGuyang): Catch the exception here because there is
             # port conflict issue which brought from static port. We should
@@ -129,7 +132,7 @@ def _init_non_minimal(self):
         else:
             logger.info(
                 "Dashboard agent grpc address: %s",
-                build_address(grpc_ip, self.grpc_port),
+                build_address(self.ip, self.grpc_port),
             )
 
         # If the agent is not minimal it should start the http server

python/ray/dashboard/http_server_agent.py

@@ -44,10 +44,17 @@ async def _start_site_with_retry(
             try:
                 site = aiohttp.web.TCPSite(
                     self.runner,
-                    "127.0.0.1" if self.ip == "127.0.0.1" else "0.0.0.0",
+                    self.ip,
                     self.listen_port,
                 )
                 await site.start()
+                # if self.ip not in ["127.0.0.1", "localhost"]:
+                #     local_site = aiohttp.web.TCPSite(
+                #         self.runner,
+                #         "127.0.0.1",
+                #         self.listen_port,
+                #     )
+                #     await local_site.start()
                 if attempt > 0:
                     logger.info(
                         f"Successfully started agent on port {self.listen_port} "

python/ray/dashboard/modules/job/tests/test_job_agent.py

@@ -25,7 +25,7 @@
     run_string_as_driver_nonblocking,
     wait_until_server_available,
 )
-from ray._common.network_utils import parse_address, build_address
+from ray._common.network_utils import build_address
 from ray.dashboard.modules.job.common import (
     JOB_ACTOR_NAME_TEMPLATE,
     SUPERVISOR_ACTOR_RAY_NAMESPACE,
@@ -77,8 +77,10 @@ def __init__(self, *args, **kwargs):
 @pytest_asyncio.fixture
 async def job_sdk_client(make_sure_dashboard_http_port_unused):
     with _ray_start(include_dashboard=True, num_cpus=1) as ctx:
-        ip, _ = parse_address(ctx.address_info["webui_url"])
-        agent_address = build_address(ip, DEFAULT_DASHBOARD_AGENT_LISTEN_PORT)
+        # Use the actual node IP address instead of parsing from webui_url
+        # which might contain localhost/127.0.0.1
+        node_ip = ctx.address_info["node_ip_address"]
+        agent_address = build_address(node_ip, DEFAULT_DASHBOARD_AGENT_LISTEN_PORT)
         assert wait_until_server_available(agent_address)
         head_address = ctx.address_info["webui_url"]
         assert wait_until_server_available(head_address)
@@ -469,8 +471,9 @@ async def test_job_log_in_multiple_node(
         dashboard_agent_listen_port=DEFAULT_DASHBOARD_AGENT_LISTEN_PORT + 2
     )
 
-    ip, _ = parse_address(cluster.webui_url)
-    agent_address = build_address(ip, DEFAULT_DASHBOARD_AGENT_LISTEN_PORT)
+    # Get the actual node IP from the cluster head node instead of parsing from webui_url
+    node_ip = cluster.head_node.node_ip_address
+    agent_address = build_address(node_ip, DEFAULT_DASHBOARD_AGENT_LISTEN_PORT)
     assert wait_until_server_available(agent_address)
     client = JobAgentSubmissionClient(format_web_url(agent_address))
 
@@ -595,18 +598,21 @@ async def test_non_default_dashboard_agent_http_port(tmp_path):
     """
     import subprocess
 
-    cmd = (
-        "ray start --head " f"--dashboard-agent-listen-port {get_current_unused_port()}"
-    )
+    dashboard_agent_port = get_current_unused_port()
+    cmd = "ray start --head " f"--dashboard-agent-listen-port {dashboard_agent_port}"
     subprocess.check_output(cmd, shell=True)
 
     try:
         # We will need to wait for the ray to be started in the subprocess.
         address_info = ray.init("auto", ignore_reinit_error=True).address_info
 
-        ip, _ = parse_address(address_info["webui_url"])
+        # Get the actual node IP address from the nodes list
+        nodes = list_nodes()
+        assert len(nodes) > 0, "No nodes found"
+        node_ip = nodes[0].node_ip
+
         dashboard_agent_listen_port = address_info["dashboard_agent_listen_port"]
-        agent_address = build_address(ip, dashboard_agent_listen_port)
+        agent_address = build_address(node_ip, dashboard_agent_listen_port)
         print("agent address = ", agent_address)
 
         agent_client = JobAgentSubmissionClient(format_web_url(agent_address))