revert proc_id

Author: fregataa

src/ai/backend/agent/agent.py

@@ -235,7 +235,6 @@ def __init__(
         distro: str,
         local_config: Mapping[str, Any],
         computers: MutableMapping[DeviceName, ComputerContext],
-        proc_uid: int,
         restarting: bool = False,
     ) -> None:
         self.image_labels = kernel_config["image"]["labels"]
@@ -254,7 +253,6 @@ def __init__(
         self.computers = computers
         self.restarting = restarting
         self.local_config = local_config
-        self.proc_uid = proc_uid
 
     @abstractmethod
     async def get_extra_envs(self) -> Mapping[str, str]:
@@ -575,7 +573,6 @@ def mount_static_binary(filename: str, target_path: str) -> None:
                     already_injected_hooks.add(hook_path)
 
         self.additional_allowed_syscalls = sorted(list(additional_allowed_syscalls_set))
-        environ["ADDITIONAL_GIDS"] = ",".join(map(str, additional_gid_set))
         update_additional_gids(environ, additional_gids)
 
     def get_overriding_uid(self) -> Optional[int]:
@@ -619,7 +616,6 @@ class AbstractAgent(
     computers: MutableMapping[DeviceName, ComputerContext]
     images: Mapping[str, str]
     port_pool: Set[int]
-    proc_uid: int
 
     redis: Redis
 
@@ -675,7 +671,6 @@ def __init__(
                 local_config["container"]["port-range"][1] + 1,
             )
         )
-        self.proc_uid = os.geteuid()
         self.stats_monitor = stats_monitor
         self.error_monitor = error_monitor
         self._pending_creation_tasks = defaultdict(set)
@@ -1761,7 +1756,6 @@ async def init_kernel_context(
         kernel_image: ImageRef,
         kernel_config: KernelCreationConfig,
         *,
-        proc_uid: int,
         restarting: bool = False,
         cluster_ssh_port_mapping: Optional[ClusterSSHPortMapping] = None,
     ) -> AbstractKernelCreationContext:
@@ -1888,7 +1882,6 @@ async def create_kernel(
                 kernel_image,
                 kernel_config,
                 restarting=restarting,
-                proc_uid=self.proc_uid,
                 cluster_ssh_port_mapping=cluster_info.get("cluster_ssh_port_mapping"),
             )
             environ: dict[str, str] = {**kernel_config["environ"]}
@@ -1911,10 +1904,10 @@ async def create_kernel(
                 if KernelFeatures.UID_MATCH in ctx.kernel_features:
                     environ["LOCAL_GROUP_ID"] = str(kernel_gid)
 
-            update_additional_gids(environ, sgids)
             environ.update(
                 await ctx.get_extra_envs(),
             )
+            update_additional_gids(environ, sgids)
             image_labels = kernel_config["image"]["labels"]
 
             agent_architecture = get_arch_name()

src/ai/backend/agent/docker/agent.py

@@ -211,7 +211,6 @@ def __init__(
         agent_sockpath: Path,
         resource_lock: asyncio.Lock,
         network_plugin_ctx: NetworkPluginContext,
-        proc_uid: int,
         restarting: bool = False,
         cluster_ssh_port_mapping: Optional[ClusterSSHPortMapping] = None,
         gwbridge_subnet: Optional[str] = None,
@@ -226,7 +225,6 @@ def __init__(
             distro,
             local_config,
             computers,
-            proc_uid=proc_uid,
             restarting=restarting,
         )
         scratch_dir = (self.local_config["container"]["scratch-root"] / str(kernel_id)).resolve()
@@ -303,7 +301,7 @@ async def prepare_resource_spec(self) -> Tuple[KernelResourceSpec, Optional[Mapp
         return resource_spec, resource_opts
 
     def _chown(self, paths: Iterable[Path], uid: Optional[int], gid: Optional[int]) -> None:
-        if self.proc_uid == 0:  # only possible when I am root.
+        if os.geteuid() == 0:  # only possible when I am root.
             for p in paths:
                 if KernelFeatures.UID_MATCH in self.kernel_features:
                     _uid = uid if uid is not None else self.local_config["container"]["kernel-uid"]
@@ -1706,7 +1704,6 @@ async def init_kernel_context(
         kernel_image: ImageRef,
         kernel_config: KernelCreationConfig,
         *,
-        proc_uid: int,
         restarting: bool = False,
         cluster_ssh_port_mapping: Optional[ClusterSSHPortMapping] = None,
     ) -> DockerKernelCreationContext:
@@ -1725,7 +1722,6 @@ async def init_kernel_context(
             self.agent_sockpath,
             self.resource_lock,
             self.network_plugin_ctx,
-            proc_uid=proc_uid,
             restarting=restarting,
             cluster_ssh_port_mapping=cluster_ssh_port_mapping,
             gwbridge_subnet=self.gwbridge_subnet,

src/ai/backend/agent/dummy/agent.py

@@ -62,7 +62,6 @@ def __init__(
         distro: str,
         local_config: Mapping[str, Any],
         computers: MutableMapping[DeviceName, ComputerContext],
-        proc_uid: int,
         restarting: bool = False,
         *,
         dummy_config: Mapping[str, Any],
@@ -77,7 +76,6 @@ def __init__(
             distro,
             local_config,
             computers,
-            proc_uid=proc_uid,
             restarting=restarting,
         )
         self.dummy_config = dummy_config
@@ -322,7 +320,6 @@ async def init_kernel_context(
         kernel_image: ImageRef,
         kernel_config: KernelCreationConfig,
         *,
-        proc_uid: int,
         restarting: bool = False,
         cluster_ssh_port_mapping: Optional[ClusterSSHPortMapping] = None,
     ) -> DummyKernelCreationContext:
@@ -337,7 +334,6 @@ async def init_kernel_context(
             distro,
             self.local_config,
             self.computers,
-            proc_uid=proc_uid,
             restarting=restarting,
             dummy_config=self.dummy_config,
         )

src/ai/backend/agent/kubernetes/agent.py

@@ -112,7 +112,6 @@ def __init__(
         computers: MutableMapping[DeviceName, ComputerContext],
         workers: Mapping[str, Mapping[str, str]],
         static_pvc_name: str,
-        proc_uid: int,
         restarting: bool = False,
     ) -> None:
         super().__init__(
@@ -125,7 +124,6 @@ def __init__(
             distro,
             local_config,
             computers,
-            proc_uid=proc_uid,
             restarting=restarting,
         )
         scratch_dir = (self.local_config["container"]["scratch-root"] / str(kernel_id)).resolve()
@@ -1040,7 +1038,6 @@ async def init_kernel_context(
         kernel_image: ImageRef,
         kernel_config: KernelCreationConfig,
         *,
-        proc_uid: int,
         restarting: bool = False,
         cluster_ssh_port_mapping: Optional[ClusterSSHPortMapping] = None,
     ) -> KubernetesKernelCreationContext:
@@ -1058,7 +1055,6 @@ async def init_kernel_context(
             self.computers,
             self.workers,
             "backend-ai-static-pvc",
-            proc_uid=proc_uid,
             restarting=restarting,
         )
 

src/ai/backend/manager/registry.py

@@ -626,9 +626,7 @@ async def create_session(
                                 {
                                     "uid": user_row.container_uid,
                                     "main_gid": user_row.container_main_gid,
-                                    "supplementary_gids": (
-                                        user_row.container_supplementary_gids or []
-                                    ),
+                                    "supplementary_gids": (user_row.container_gids or []),
                                     "image_ref": image_ref,
                                     "cluster_role": DEFAULT_ROLE,
                                     "cluster_idx": 1,
@@ -1334,7 +1332,7 @@ async def enqueue_session(
                 ),
                 "uid": kernel["uid"],
                 "main_gid": kernel["main_gid"],
-                "supplementary_gids": kernel["supplementary_gids"],
+                "gids": kernel["supplementary_gids"],
                 "image": image_ref.canonical,
                 # "image_id": image_row.id,
                 "architecture": image_ref.architecture,
@@ -1854,7 +1852,7 @@ def get_image_conf(kernel: KernelRow) -> ImageConfig:
                             "cluster_hostname": binding.kernel.cluster_hostname,
                             "uid": binding.kernel.uid,
                             "main_gid": binding.kernel.main_gid,
-                            "supplementary_gids": binding.kernel.supplementary_gids or [],
+                            "supplementary_gids": binding.kernel.gids or [],
                             "idle_timeout": int(idle_timeout),
                             "mounts": [item.to_json() for item in scheduled_session.vfolder_mounts],
                             "environ": {