Revert "[core] (cgroups 8/n) Wiring CgroupManager into the raylet. (#56297)"

This reverts commit 161dd95.

Author: israbbani

src/ray/common/cgroup2/BUILD.bazel

@@ -1,40 +1,13 @@
 load("//bazel:ray.bzl", "ray_cc_library")
 
-config_setting(
-    name = "is_linux",
-    constraint_values = ["@platforms//os:linux"],
-)
-
-# Public targets.
-ray_cc_library(
-    name = "cgroup_manager",
-    srcs = select({
-        ":is_linux": ["cgroup_manager.cc"],
-        "//conditions:default": ["noop_cgroup_manager.cc"],
-    }),
-    hdrs = ["cgroup_manager.h"],
-    visibility = ["//visibility:public"],
-    deps = [
-        ":cgroup_driver_interface",
-        ":cgroup_manager_interface",
-        "//src/ray/common:status",
-        "//src/ray/common:status_or",
-    ] + select({
-        ":is_linux": [
-            ":scoped_cgroup_operation",
-            "//src/ray/util:logging",
-            "@com_google_absl//absl/strings",
-        ],
-        "//conditions:default": [],
-    }),
-)
-
 ray_cc_library(
     name = "cgroup_driver_interface",
     hdrs = [
         "cgroup_driver_interface.h",
     ],
-    visibility = ["//visibility:public"],
+    target_compatible_with = [
+        "@platforms//os:linux",
+    ],
     deps = [
         "//src/ray/common:status",
         "//src/ray/common:status_or",
@@ -46,42 +19,51 @@ ray_cc_library(
     hdrs = [
         "cgroup_manager_interface.h",
     ],
-    visibility = ["//visibility:public"],
+    target_compatible_with = [
+        "@platforms//os:linux",
+    ],
     deps = [
         "//src/ray/common:status",
         "//src/ray/common:status_or",
     ],
 )
 
 ray_cc_library(
-    name = "sysfs_cgroup_driver",
-    srcs = ["sysfs_cgroup_driver.cc"],
+    name = "cgroup_manager",
+    srcs = ["cgroup_manager.cc"],
     hdrs = [
-        "sysfs_cgroup_driver.h",
+        "cgroup_manager.h",
+        "scoped_cgroup_operation.h",
     ],
     target_compatible_with = [
         "@platforms//os:linux",
     ],
-    visibility = ["//visibility:public"],
     deps = [
         ":cgroup_driver_interface",
+        ":cgroup_manager_interface",
         "//src/ray/common:status",
         "//src/ray/common:status_or",
         "//src/ray/util:logging",
         "@com_google_absl//absl/strings",
     ],
 )
 
-# Private Targets.
 ray_cc_library(
-    name = "scoped_cgroup_operation",
+    name = "sysfs_cgroup_driver",
+    srcs = ["sysfs_cgroup_driver.cc"],
     hdrs = [
-        "scoped_cgroup_operation.h",
+        "sysfs_cgroup_driver.h",
     ],
     target_compatible_with = [
         "@platforms//os:linux",
     ],
-    visibility = [":__subpackages__"],
+    deps = [
+        ":cgroup_driver_interface",
+        "//src/ray/common:status",
+        "//src/ray/common:status_or",
+        "//src/ray/util:logging",
+        "@com_google_absl//absl/strings",
+    ],
 )
 
 ray_cc_library(
@@ -92,7 +74,6 @@ ray_cc_library(
     target_compatible_with = [
         "@platforms//os:linux",
     ],
-    visibility = [":__subpackages__"],
     deps = [
         ":cgroup_driver_interface",
         "//src/ray/common:status",
@@ -106,7 +87,6 @@ ray_cc_library(
     target_compatible_with = [
         "@platforms//os:linux",
     ],
-    visibility = [":__subpackages__"],
     deps = [
         "//src/ray/common:id",
         "//src/ray/common:status",

src/ray/common/cgroup2/noop_cgroup_manager.cc

@@ -283,8 +283,6 @@ ray_cc_binary(
         "//src/ray/common:lease",
         "//src/ray/common:ray_config",
         "//src/ray/common:status",
-        "//src/ray/common/cgroup2:cgroup_manager",
-        "//src/ray/common/cgroup2:sysfs_cgroup_driver",
         "//src/ray/core_worker:metrics",
         "//src/ray/gcs/gcs_client:gcs_client_lib",
         "//src/ray/object_manager:ownership_object_directory",

src/ray/raylet/BUILD.bazel

@@ -24,14 +24,11 @@
 #include "gflags/gflags.h"
 #include "nlohmann/json.hpp"
 #include "ray/common/asio/instrumented_io_context.h"
-#include "ray/common/cgroup2/cgroup_manager.h"
-#include "ray/common/cgroup2/sysfs_cgroup_driver.h"
 #include "ray/common/constants.h"
 #include "ray/common/id.h"
 #include "ray/common/lease/lease.h"
 #include "ray/common/ray_config.h"
 #include "ray/common/status.h"
-#include "ray/common/status_or.h"
 #include "ray/core_worker/metrics.h"
 #include "ray/gcs/gcs_client/gcs_client.h"
 #include "ray/object_manager/ownership_object_directory.h"
@@ -114,30 +111,6 @@ DEFINE_bool(huge_pages, false, "Enable huge pages.");
 DEFINE_string(labels,
               "",
               "Define the key-value format of node labels, which is a serialized JSON.");
-DEFINE_bool(
-    enable_resource_isolation,
-    false,
-    "Enables resource isolation through cgroupv2. The raylet will create and "
-    "manage a cgroup hierarchy that separates system processes and worker processes "
-    "into separate cgroups.");
-DEFINE_string(
-    cgroup_path,
-    "",
-    "Path of the cgroup that the raylet will take ownership of to create its cgorup "
-    "hierarchy. The raylet process must have read, write, and execute permission for "
-    "this path. If enable_resource_isolation is true, then this cannot be empty.");
-DEFINE_int64(
-    system_reserved_cpu_weight,
-    -1,
-    "The amount of cores reserved for ray system processes. It will be applied "
-    "as a cpu.weight constraint to the system cgroup. 10000 - "
-    "system_reserved_cpu_weight will be applied as a constraint to the "
-    "application cgroup. If enable resource isolation is true, then this cannot be -1.");
-DEFINE_int64(system_reserved_memory_bytes,
-             -1,
-             "The amount of memory in bytes reserved for ray system processes. It will "
-             "be applied as a memory.min constraint to the sytem cgroup. If enable "
-             "resource isolation is true, then this cannot be -1");
 
 absl::flat_hash_map<std::string, std::string> parse_node_labels(
     const std::string &labels_json_str) {
@@ -245,48 +218,12 @@ int main(int argc, char *argv[]) {
   const std::string session_name = FLAGS_session_name;
   const bool is_head_node = FLAGS_head;
   const std::string labels_json_str = FLAGS_labels;
-  const bool enable_resource_isolation = FLAGS_enable_resource_isolation;
-  const std::string cgroup_path = FLAGS_cgroup_path;
-  const int64_t system_reserved_cpu_weight = FLAGS_system_reserved_cpu_weight;
-  const int64_t system_reserved_memory_bytes = FLAGS_system_reserved_memory_bytes;
 
   RAY_CHECK_NE(FLAGS_cluster_id, "") << "Expected cluster ID.";
   ray::ClusterID cluster_id = ray::ClusterID::FromHex(FLAGS_cluster_id);
   RAY_LOG(INFO) << "Setting cluster ID to: " << cluster_id;
   gflags::ShutDownCommandLineFlags();
 
-  // TODO(#54703): Link OSS documentation once it's available in the error messages.
-  if (enable_resource_isolation) {
-    RAY_CHECK(!cgroup_path.empty())
-        << "Failed to start up raylet. If enable_resource_isolation is set to true, "
-           "cgroup_path cannot be empty.";
-    RAY_CHECK_NE(system_reserved_cpu_weight, -1)
-        << "Failed to start up raylet. If enable_resource_isolation is set to true, "
-           "system_reserved_cpu_weight must be set to a value between [1,10000]";
-    RAY_CHECK_NE(system_reserved_memory_bytes, -1)
-        << "Failed to start up raylet. If enable_resource_isolation is set to true, "
-           "system_reserved_memory_byres must be set to a value > 0";
-
-    std::unique_ptr<ray::SysFsCgroupDriver> cgroup_driver;
-    ray::StatusOr<std::unique_ptr<ray::CgroupManager>> cgroup_manager =
-        ray::CgroupManager::Create(std::move(cgroup_path),
-                                   node_id,
-                                   system_reserved_cpu_weight,
-                                   system_reserved_memory_bytes,
-                                   std::move(cgroup_driver));
-
-    // TODO(#54703) - Link to OSS documentation once available.
-    RAY_CHECK(cgroup_manager.ok())
-        << "Failed to start raylet. Could not create CgroupManager because of "
-        << cgroup_manager.ToString();
-
-#ifndef __linux__
-    RAY_LOG(WARNING)
-        << "Resource isolation with cgroups is only supported in linux. Please set "
-           "enable_resource_isolation to false. This is likely a misconfiguration.";
-#endif
-  }
-
   // Configuration for the node manager.
   ray::raylet::NodeManagerConfig node_manager_config;