gpustack-runtime 0.2.3.post2__tar.gz → 0.2.4__tar.gz

{gpustack_runtime-0.2.3.post2 → gpustack_runtime-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: gpustack-runtime
-Version: 0.2.3.post2
+Version: 0.2.4
 Summary: GPUStack Runtime is library for detecting GPU resources and launching GPU workloads.
 Project-URL: Homepage, https://github.com/gpustack/runtime
 Project-URL: Bug Tracker, https://github.com/gpustack/gpustack/issues

{gpustack_runtime-0.2.3.post2 → gpustack_runtime-0.2.4}/gpustack_runtime/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.2.3.post2'
-__version_tuple__ = version_tuple = (0, 2, 3, 'post2')
+__version__ = version = '0.2.4'
+__version_tuple__ = version_tuple = (0, 2, 4)
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runtime-0.2.4/gpustack_runtime/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "59ea8c8"

{gpustack_runtime-0.2.3.post2 → gpustack_runtime-0.2.4}/gpustack_runtime/deployer/cdi/__utils__.py

@@ -143,6 +143,56 @@ def device_to_cdi_device_node(
     )
 
 
+def path_to_cdi_device_nodes(
+    path: str,
+    permission: str = "rw",
+    no_user: bool = False,
+) -> list[ConfigDeviceNode]:
+    """
+    Convert a device path, or a directory holding device paths, to ConfigDeviceNodes.
+
+    A generation may expose a bus as a directory of device nodes rather than as
+    a single node -- Ascend's UB does, which is why the operator enumerates it,
+    see addUBDevicesFromDir in
+    https://gitcode.com/Ascend/mind-cluster/blob/master/component/ascend-common/cdi/devnode.go.
+    Both shapes are accepted so that a path which is a plain device node on one
+    driver and a directory on another needs no caller-side branch.
+
+    Args:
+        path:
+            Path to the device, or to a directory of devices, on the host.
+        permission:
+            Permissions for the devices.
+        no_user:
+            Whether to omit user and group information.
+
+    Returns:
+        The ConfigDeviceNode objects, empty if the path holds no device.
+
+    """
+    p = Path(path)
+    if not p.is_dir():
+        cdn = device_to_cdi_device_node(
+            path=path,
+            permission=permission,
+            no_user=no_user,
+        )
+        return [cdn] if cdn else []
+
+    return [
+        cdn
+        for entry in sorted(p.iterdir())
+        if not entry.is_dir()
+        and (
+            cdn := device_to_cdi_device_node(
+                path=str(entry),
+                permission=permission,
+                no_user=no_user,
+            )
+        )
+    ]
+
+
 def path_to_cdi_mount(
     path: str,
     container_path: str | None = None,
@@ -181,3 +231,34 @@ def path_to_cdi_mount(
         container_path=container_path,
         options=options,
     )
+
+
+def glob_to_cdi_mounts(
+    pattern: str,
+    options: list[str] | None = None,
+) -> list[ConfigMount]:
+    """
+    Convert every path matching a glob pattern to ConfigMounts.
+
+    A user-space library is versioned in its file name, so the set of files to
+    mount cannot be spelled out ahead of time -- the operator's mount profile
+    lists them as patterns for the same reason.
+
+    Args:
+        pattern:
+            Path on the host whose last segment may carry a wildcard; the
+            directory part is taken literally, so a wildcard there matches
+            nothing. A pattern without a wildcard resolves to that path alone.
+        options:
+            Mount options.
+
+    Returns:
+        The ConfigMount objects, empty if nothing matches.
+
+    """
+    p = Path(pattern)
+    return [
+        cm
+        for path in sorted(p.parent.glob(p.name))
+        if (cm := path_to_cdi_mount(path=str(path), options=options))
+    ]

{gpustack_runtime-0.2.3.post2 → gpustack_runtime-0.2.4}/gpustack_runtime/deployer/cdi/ascend.py

@@ -1,11 +1,14 @@
 from __future__ import annotations as __future_annotations__
 
+import logging
+
 from ...detector import (
     Devices,
     ManufacturerEnum,
     detect_devices,
     filter_devices_by_manufacturer,
 )
+from ...detector.ascend import get_ascend_cann_variant
 from .__types__ import (
     Config,
     ConfigContainerEdits,
@@ -14,7 +17,44 @@ from .__types__ import (
     manufacturer_to_cdi_kind,
     manufacturer_to_runtime_env,
 )
-from .__utils__ import device_to_cdi_device_node, path_to_cdi_mount
+from .__utils__ import (
+    device_to_cdi_device_node,
+    glob_to_cdi_mounts,
+    path_to_cdi_device_nodes,
+    path_to_cdi_mount,
+)
+
+logger = logging.getLogger(__name__)
+
+_A5_CANN_VARIANT = "950"
+"""
+The CANN variant of the A5 generation, whose UB fabric replaces what the
+earlier generations reach through the shared memory device.
+"""
+
+_A5_UB_MOUNT_PATTERNS = [
+    "/usr/lib64/libummu*",
+    "/usr/lib64/liburma*",
+    "/usr/lib64/urma",
+    "/usr/lib64/libnl*",
+    "/usr/bin/urma_admin",
+    "/usr/bin/urma_perftest",
+    "/usr/bin/urma_ping",
+]
+"""
+The UB user-space libraries an A5 container needs, mirroring the operator's
+mount profile for the Ascend950 generation, see
+https://gitcode.com/Ascend/mind-cluster/blob/master/component/ascend-common/cdi/mount/profile.go.
+
+These are mounted for the A5 generation only. Some of them -- libnl above all
+-- are ordinary system libraries present on any host, so mounting them
+unconditionally would shadow what an earlier generation's container ships with
+its own image.
+
+The /usr/lib64 prefix is the operator's, and holds on the openEuler and CentOS
+hosts an A5 ships on; a Debian-derived host uses /usr/lib/<triplet>, where none
+of these match -- hence the log line when a pattern finds nothing.
+"""
 
 
 class AscendGenerator(Generator):
@@ -73,16 +113,14 @@ class AscendGenerator(Generator):
                 break
         for p in [
             "/dev/dvpp_cmdlist",
+            # UB exposes a directory of device nodes rather than a single one,
+            # so every entry below it has to be injected.
             "/dev/uburma",
             "/dev/ummu",
             "/dev/devmm_svm",
             "/dev/hisi_hdc",
         ]:
-            cdn = device_to_cdi_device_node(
-                path=p,
-            )
-            if cdn:
-                common_device_nodes.append(cdn)
+            common_device_nodes.extend(path_to_cdi_device_nodes(path=p))
         if not common_device_nodes:
             return None
 
@@ -102,6 +140,25 @@ class AscendGenerator(Generator):
             if cm:
                 common_mounts.append(cm)
 
+        # Device.appendix defaults to None and callers pass their own devices
+        # in, so every read goes through `or {}`.
+        if any(
+            get_ascend_cann_variant((dev.appendix or {}).get("arch_family"))
+            == _A5_CANN_VARIANT
+            for dev in devices
+            if dev
+        ):
+            for pattern in _A5_UB_MOUNT_PATTERNS:
+                ub_mounts = glob_to_cdi_mounts(pattern=pattern)
+                if not ub_mounts:
+                    # Otherwise this surfaces later as a container that cannot
+                    # reach the UB fabric.
+                    logger.debug(
+                        "No UB library matched %s, an A5 container will lack it",
+                        pattern,
+                    )
+                common_mounts.extend(ub_mounts)
+
         cdi_devices: list[ConfigDevice] = []
 
         all_device_nodes = []
@@ -119,7 +176,7 @@ class AscendGenerator(Generator):
             # addressed by the index, which would resolve to another NPU's
             # node. The detector already drops such a device; this guards the
             # devices a caller passes in.
-            cdn_number = dev.appendix.get("physical_id")
+            cdn_number = (dev.appendix or {}).get("physical_id")
             if cdn_number is None:
                 continue
             cdn_path = f"/dev/davinci{cdn_number}"