[WIP][dashboard] remove DataSource.agents for good.

python/ray/dashboard/agent.py

@@ -202,8 +202,8 @@ async def run(self):
         http_port = -1 if not self.http_server else self.http_server.http_port
         grpc_port = -1 if not self.server else self.grpc_port
         await self.gcs_aio_client.internal_kv_put(
-            f"{dashboard_consts.DASHBOARD_AGENT_PORT_PREFIX}{self.node_id}".encode(),
-            json.dumps([http_port, grpc_port]).encode(),
+            f"{dashboard_consts.DASHBOARD_AGENT_ADDR_PREFIX}{self.node_id}".encode(),
+            json.dumps([self.ip, http_port, grpc_port]).encode(),
             True,
             namespace=ray_constants.KV_NAMESPACE_DASHBOARD,
         )

python/ray/dashboard/client/src/common/ProfilingLink.tsx

@@ -20,7 +20,7 @@ import { ClassNameProps } from "./props";
 type CpuProfilingLinkProps = PropsWithChildren<
   {
     pid: string | number | null | undefined;
-    ip: string | null | undefined;
+    nodeId: string | null | undefined;
     type: string | null;
   } & ClassNameProps
 >;
@@ -34,7 +34,7 @@ type TaskProfilingStackTraceProps = {
 type MemoryProfilingProps = PropsWithChildren<
   {
     pid: string | number | null | undefined;
-    ip: string | null | undefined;
+    nodeId: string | null | undefined;
     type?: string | null;
   } & ClassNameProps
 >;
@@ -92,15 +92,20 @@ export const TaskCpuStackTraceLink = ({
 
 export const CpuStackTraceLink = ({
   pid,
-  ip,
+  nodeId,
   type = "",
 }: CpuProfilingLinkProps) => {
-  if (!pid || !ip || typeof pid === "undefined" || typeof ip === "undefined") {
+  if (
+    !pid ||
+    !nodeId ||
+    typeof pid === "undefined" ||
+    typeof nodeId === "undefined"
+  ) {
     return <div></div>;
   }
   return (
     <Link
-      href={`worker/traceback?pid=${pid}&ip=${ip}&native=0`}
+      href={`worker/traceback?pid=${pid}&node_id=${nodeId}&native=0`}
       target="_blank"
       title="Sample the current Python stack trace for this worker."
       rel="noreferrer"
@@ -112,16 +117,16 @@ export const CpuStackTraceLink = ({
 
 export const CpuProfilingLink = ({
   pid,
-  ip,
+  nodeId,
   type = "",
 }: CpuProfilingLinkProps) => {
-  if (!pid || !ip) {
+  if (!pid || !nodeId) {
     return <div></div>;
   }
 
   return (
     <Link
-      href={`worker/cpu_profile?pid=${pid}&ip=${ip}&duration=5&native=0`}
+      href={`worker/cpu_profile?pid=${pid}&node_id=${nodeId}&duration=5&native=0`}
       target="_blank"
       title="Profile the Python worker for 5 seconds (default) and display a CPU flame graph."
       rel="noreferrer"
@@ -283,13 +288,13 @@ export const ProfilerButton = ({
 
 export const MemoryProfilingButton = ({
   pid,
-  ip,
+  nodeId,
   type = "",
 }: MemoryProfilingProps) => {
-  if (!pid || !ip) {
+  if (!pid || !nodeId) {
     return <div></div>;
   }
-  const profilerUrl = `memory_profile?pid=${pid}&ip=${ip}`;
+  const profilerUrl = `memory_profile?pid=${pid}&node_id=${nodeId}`;
 
   return <ProfilerButton profilerUrl={profilerUrl} type={type} />;
 };

python/ray/dashboard/client/src/components/ActorTable.tsx

@@ -609,19 +609,19 @@ const ActorTable = ({
                       <br />
                       <CpuProfilingLink
                         pid={pid}
-                        ip={address?.ipAddress}
+                        nodeId={address?.rayletId}
                         type=""
                       />
                       <br />
                       <CpuStackTraceLink
                         pid={pid}
-                        ip={address?.ipAddress}
+                        nodeId={address?.rayletId}
                         type=""
                       />
                       <br />
                       <MemoryProfilingButton
                         pid={pid}
-                        ip={address?.ipAddress}
+                        nodeId={address?.rayletId}
                       />
                     </React.Fragment>
                   </TableCell>

python/ray/dashboard/client/src/pages/actor/ActorDetail.tsx

@@ -188,19 +188,19 @@ const ActorDetailPage = () => {
               <div>
                 <CpuStackTraceLink
                   pid={actorDetail.pid}
-                  ip={actorDetail.address?.ipAddress}
+                  nodeId={actorDetail.address?.rayletId}
                   type=""
                 />
                 <br />
                 <CpuProfilingLink
                   pid={actorDetail.pid}
-                  ip={actorDetail.address?.ipAddress}
+                  nodeId={actorDetail.address?.rayletId}
                   type=""
                 />
                 <br />
                 <MemoryProfilingButton
                   pid={actorDetail.pid}
-                  ip={actorDetail.address?.ipAddress}
+                  nodeId={actorDetail.address?.rayletId}
                   type=""
                 />
               </div>

python/ray/dashboard/client/src/pages/job/JobDetailInfoPage.tsx

@@ -172,19 +172,19 @@ export const JobMetadataSection = ({ job }: JobMetadataSectionProps) => {
             <div>
               <CpuStackTraceLink
                 pid={job.driver_info?.pid}
-                ip={job.driver_info?.node_ip_address}
+                nodeId={job.driver_info?.node_id}
                 type="Driver"
               />
               <br />
               <CpuProfilingLink
                 pid={job.driver_info?.pid}
-                ip={job.driver_info?.node_ip_address}
+                nodeId={job.driver_info?.node_id}
                 type="Driver"
               />
               <br />
               <MemoryProfilingButton
                 pid={job.driver_info?.pid}
-                ip={job.driver_info?.node_ip_address}
+                nodeId={job.driver_info?.node_id}
                 type="Driver"
               />
             </div>

python/ray/dashboard/client/src/pages/job/JobRow.tsx

@@ -116,19 +116,19 @@ export const JobRow = ({ job }: JobRowProps) => {
         )}
         <CpuStackTraceLink
           pid={job.driver_info?.pid}
-          ip={job.driver_info?.node_ip_address}
+          nodeId={job.driver_info?.node_id}
           type="Driver"
         />
         <br />
         <CpuProfilingLink
           pid={job.driver_info?.pid}
-          ip={job.driver_info?.node_ip_address}
+          nodeId={job.driver_info?.node_id}
           type="Driver"
         />
         <br />
         <MemoryProfilingButton
           pid={job.driver_info?.pid}
-          ip={job.driver_info?.node_ip_address}
+          nodeId={job.driver_info?.node_id}
           type="Driver"
         />
       </TableCell>

python/ray/dashboard/client/src/pages/node/NodeRow.tsx

@@ -227,7 +227,6 @@ type WorkerRowProps = {
  */
 export const WorkerRow = ({ node, worker }: WorkerRowProps) => {
   const {
-    ip,
     mem,
     raylet: { nodeId },
   } = node;
@@ -278,11 +277,11 @@ export const WorkerRow = ({ node, worker }: WorkerRowProps) => {
           Log
         </Link>
         <br />
-        <CpuProfilingLink pid={pid} ip={ip} type="" />
+        <CpuProfilingLink pid={pid} nodeId={nodeId} type="" />
         <br />
-        <CpuStackTraceLink pid={pid} ip={ip} type="" />
+        <CpuStackTraceLink pid={pid} nodeId={nodeId} type="" />
         <br />
-        <MemoryProfilingButton pid={pid} ip={ip} />
+        <MemoryProfilingButton pid={pid} nodeId={nodeId} />
       </TableCell>
       <TableCell>
         <PercentageBar num={Number(cpu)} total={100}>

python/ray/dashboard/consts.py

@@ -3,7 +3,7 @@
 from ray._private.ray_constants import env_bool, env_integer
 
 DASHBOARD_LOG_FILENAME = "dashboard.log"
-DASHBOARD_AGENT_PORT_PREFIX = "DASHBOARD_AGENT_PORT_PREFIX:"
+DASHBOARD_AGENT_ADDR_PREFIX = "DASHBOARD_AGENT_ADDR_PREFIX:"
 DASHBOARD_AGENT_LOG_FILENAME = "dashboard_agent.log"
 DASHBOARD_AGENT_CHECK_PARENT_INTERVAL_S_ENV_NAME = (
     "RAY_DASHBOARD_AGENT_CHECK_PARENT_INTERVAL_S"  # noqa

python/ray/dashboard/datacenter.py

@@ -1,5 +1,5 @@
 import logging
-from typing import Any, List, Optional
+from typing import List, Optional
 
 import ray.dashboard.consts as dashboard_consts
 from ray._private.utils import (
@@ -26,9 +26,6 @@ class DataSource:
     # {actor id hex(str): actor table data(dict of ActorTableData
     # in gcs.proto)}
     actors = MutableNotificationDict()
-    # {job id hex(str): job table data(dict of JobTableData in gcs.proto)}
-    # {node id hex(str): dashboard agent [http port(int), grpc port(int)]}
-    agents = Dict()
     # {node id hex(str): gcs node info(dict of GcsNodeInfo in gcs.proto)}
     nodes = Dict()
     # {node id hex(str): worker list}
@@ -48,7 +45,6 @@ async def purge():
         # Purge data that is out of date.
         # These data sources are maintained by DashboardHead,
         # we do not needs to purge them:
-        #   * agents
         #   * nodes
         alive_nodes = {
             node_id
@@ -188,42 +184,6 @@ async def get_all_node_summary(cls):
             for node_id in DataSource.nodes.keys()
         ]
 
-    @classmethod
-    async def get_agent_infos(
-        cls, target_node_ids: Optional[List[str]] = None
-    ) -> Dict[str, Dict[str, Any]]:
-        """Fetches running Agent (like HTTP/gRPC ports, IP, etc) running on every node
-
-        :param target_node_ids: Target node ids to fetch agent info for. If omitted will
-                                fetch the info for all agents
-        """
-
-        # Return all available agent infos in case no target node-ids were provided
-        target_node_ids = target_node_ids or DataSource.agents.keys()
-
-        missing_node_ids = [
-            node_id for node_id in target_node_ids if node_id not in DataSource.agents
-        ]
-        if missing_node_ids:
-            logger.warning(
-                f"Agent info was not found for {missing_node_ids}"
-                f" (having agent infos for {list(DataSource.agents.keys())})"
-            )
-            return {}
-
-        def _create_agent_info(node_id: str):
-            (http_port, grpc_port) = DataSource.agents[node_id]
-            node_ip = DataSource.nodes[node_id]["nodeManagerAddress"]
-
-            return dict(
-                ipAddress=node_ip,
-                httpPort=int(http_port or -1),
-                grpcPort=int(grpc_port or -1),
-                httpAddress=f"{node_ip}:{http_port}",
-            )
-
-        return {node_id: _create_agent_info(node_id) for node_id in target_node_ids}
-
     @classmethod
     async def get_actor_infos(cls, actor_ids: Optional[List[str]] = None):
         target_actor_table_entries: dict[str, Optional[dict]]