refactor: command ui

Signed-off-by: thxCode <[email protected]>

Author: thxCode

gpustack_runtime/cmds/deployer.py

@@ -111,6 +111,13 @@ def register(parser: _SubParsersAction):
             default=80,
         )
 
+        deploy_parser.add_argument(
+            "--host-network",
+            action="store_true",
+            help="use host network (default: False)",
+            default=False,
+        )
+
         deploy_parser.add_argument(
             "--namespace",
             type=str,
@@ -147,6 +154,7 @@ def __init__(self, args: Namespace):
         self.backend = args.backend
         self.device = args.device
         self.port = args.port
+        self.host_network = args.host_network
         self.service = args.service
         self.version = args.version
         self.namespace = args.namespace
@@ -209,7 +217,7 @@ def run(self):
         plan = WorkloadPlan(
             name=self.name,
             namespace=self.namespace,
-            host_network=True,
+            host_network=self.host_network,
             containers=[
                 Container(
                     image=f"gpustack/runner:{self.backend if self.backend else 'Host'}X.Y-{self.service}{self.version}",
@@ -229,7 +237,10 @@ def run(self):
         try:
             while True:
                 st = get_workload(name=self.name, namespace=self.namespace)
-                if st and st.state == WorkloadStatusStateEnum.RUNNING:
+                if st and st.state not in (
+                    WorkloadStatusStateEnum.PENDING,
+                    WorkloadStatusStateEnum.INITIALIZING,
+                ):
                     break
                 time.sleep(1)
 
@@ -289,6 +300,13 @@ def register(parser: _SubParsersAction):
             default=80,
         )
 
+        deploy_parser.add_argument(
+            "--host-network",
+            action="store_true",
+            help="use host network (default: False)",
+            default=False,
+        )
+
         deploy_parser.add_argument(
             "--namespace",
             type=str,
@@ -325,6 +343,7 @@ def __init__(self, args: Namespace):
         self.backend = args.backend
         self.device = args.device
         self.port = args.port
+        self.host_network = args.host_network
         self.namespace = args.namespace
         self.name = args.name
         self.image = args.image
@@ -386,7 +405,7 @@ def run(self):
         plan = WorkloadPlan(
             name=self.name,
             namespace=self.namespace,
-            host_network=True,
+            host_network=self.host_network,
             containers=[
                 Container(
                     image=self.image,
@@ -406,7 +425,10 @@ def run(self):
         try:
             while True:
                 st = get_workload(name=self.name, namespace=self.namespace)
-                if st and st.state == WorkloadStatusStateEnum.RUNNING:
+                if st and st.state not in (
+                    WorkloadStatusStateEnum.PENDING,
+                    WorkloadStatusStateEnum.INITIALIZING,
+                ):
                     break
                 time.sleep(1)
 
@@ -821,20 +843,15 @@ def format_workloads_table(sts: list[WorkloadStatus]) -> str:
     if not sts:
         return "No workloads found."
 
-    width = 100
-
     headers = ["Name", "State", "Created At"]
-    col_widths = [
-        len(str(getattr(st, attr.lower().replace(" ", "_"))))
-        for st in sts
-        for attr in headers
-    ]
-    col_widths = [max(w, len(h)) for w, h in zip(col_widths, headers, strict=False)]
-
-    total_width = sum(col_widths) + len(col_widths) * 3 + 1
-    if total_width > width:
-        scale = (width - len(col_widths) * 3 - 1) / sum(col_widths)
-        col_widths = [int(w * scale) for w in col_widths]
+    # Calculate max width for each column based on header and content
+    col_widths = []
+    for attr in headers:
+        attr_key = attr.lower().replace(" ", "_")
+        max_content_width = max(
+            [len(str(getattr(st, attr_key))) for st in sts] + [len(attr)],
+        )
+        col_widths.append(max_content_width)
 
     lines = []
     header_line = (
@@ -849,9 +866,9 @@ def format_workloads_table(sts: list[WorkloadStatus]) -> str:
 
     for st in sts:
         row = [
-            st.name.ljust(col_widths[0]),
-            st.state.ljust(col_widths[1]),
-            st.created_at.ljust(col_widths[2]),
+            str(st.name).ljust(col_widths[0]),
+            str(st.state).ljust(col_widths[1]),
+            str(st.created_at).ljust(col_widths[2]),
         ]
         line = "| " + " | ".join(row) + " |"
         lines.append(line)

gpustack_runtime/cmds/detector.py

@@ -80,63 +80,68 @@ def format_devices_table(devs: Devices) -> str:
     if not devs:
         return "No GPUs detected."
 
-    width = 100
+    # Column headers
+    col_headers = ["GPU", "Name", "Memory-Usage", "GPU-Util", "Temp", "CC"]
+    # Gather all rows to determine max width for each column
+    rows = []
+    for dev in devs:
+        row = [
+            str(dev.index),
+            dev.name if dev.name else "N/A",
+            f"{dev.memory_used}MiB / {dev.memory}MiB",
+            f"{dev.cores_utilization}%",
+            f"{dev.temperature}C" if dev.temperature is not None else "N/A",
+            dev.compute_capability if dev.compute_capability else "N/A",
+        ]
+        rows.append(row)
+
+    # Calculate max width for each column
+    col_widths = [len(header) for header in col_headers]
+    for row in rows:
+        for i, cell in enumerate(row):
+            col_widths[i] = max(col_widths[i], len(str(cell)))
+
+    # Add padding
+    col_widths = [w + 2 for w in col_widths]
+
+    # Calculate table width
+    width = sum(col_widths) + len(col_widths) + 1
 
     # Header section
     dev = devs[0]
     header_content = f"{dev.manufacturer.upper()} "
     header_content += (
         f"Driver Version: {dev.driver_version if dev.driver_version else 'N/A'} "
     )
-    header_content += f"Runtime Version: {dev.runtime_version if dev.runtime_version else 'N/A'}".rjust(
-        width - len(header_content) - 4,
+    runtime_version_str = (
+        f"Runtime Version: {dev.runtime_version if dev.runtime_version else 'N/A'}"
     )
-
     header_lines = [
         "+" + "-" * (width - 2) + "+",
-        f"| {header_content.ljust(width - 4)} |",
+        f"| {header_content.ljust(width - 4 - len(runtime_version_str))}{runtime_version_str} |",
         "|" + "-" * (width - 2) + "|",
     ]
 
-    # Column headers
-    col_headers = ["GPU", "Name", "Memory-Usage", "GPU-Util", "Temp", "CC"]
-    col_widths = [5, 31, 20, 10, 6, 6]
-
-    # Adjust column widths to fit total width
-    total_current_width = sum(col_widths) + len(col_widths) * 3 - 1
-    if total_current_width < width - 2:
-        # Distribute extra space to the name column
-        col_widths[1] += width - 2 - total_current_width
-
-    # Create column header line
+    # Column header line
     col_header_line = "|"
     for i, header in enumerate(col_headers):
-        col_header_line += f" {header.center(col_widths[i])} |"
+        col_header_line += f" {header.center(col_widths[i] - 2)} |"
     header_lines.append(col_header_line)
 
     # Separator line
-    separator = "|" + "|".join(["-" * (w + 2) for w in col_widths]) + "|"
+    separator = "|" + "|".join(["-" * w for w in col_widths]) + "|"
     header_lines.append(separator)
 
     # Device rows
     device_lines = []
-    for dev in devs:
-        row_data: list[str] = [
-            str(dev.index),
-            dev.name if dev.name else "N/A",
-            f"{dev.memory_used}MiB / {dev.memory}MiB",
-            f"{dev.cores_utilization}%",
-            f"{dev.temperature}C" if dev.temperature is not None else "N/A",
-            dev.compute_capability if dev.compute_capability else "N/A",
-        ]
-
+    for row in rows:
         row_line = "|"
-        for j, data in enumerate(row_data):
-            # Truncate name if too long
-            if j == 1 and len(data) > col_widths[j]:
-                data = data[: col_widths[j] - 3] + "..."  # noqa: PLW2901
-            row_line += f" {data.ljust(col_widths[j])} |"
-
+        for j, data in enumerate(row):
+            cell = str(data)
+            # Truncate if too long
+            if len(cell) > col_widths[j] - 2:
+                cell = cell[: col_widths[j] - 5] + "..."
+            row_line += f" {cell.ljust(col_widths[j] - 2)} |"
         device_lines.append(row_line)
 
     # Footer section

gpustack_runtime/deployer/__types__.py

@@ -277,6 +277,9 @@ class ContainerMountModeEnum(str, Enum):
     Read-write many mode.
     """
 
+    def __str__(self):
+        return self.value
+
 
 @dataclass
 class ContainerMount:
@@ -333,6 +336,9 @@ class ContainerPortProtocolEnum(str, Enum):
     SCTP protocol.
     """
 
+    def __str__(self):
+        return self.value
+
 
 @dataclass
 class ContainerPort:
@@ -518,6 +524,9 @@ class ContainerProfileEnum(str, Enum):
     Init profile.
     """
 
+    def __str__(self):
+        return self.value
+
 
 class ContainerRestartPolicyEnum(str, Enum):
     """
@@ -537,6 +546,9 @@ class ContainerRestartPolicyEnum(str, Enum):
     Never restart the container.
     """
 
+    def __str__(self):
+        return self.value
+
 
 @dataclass
 class Container:
@@ -910,6 +922,9 @@ class WorkloadStatusStateEnum(str, Enum):
     The workload is inactive.
     """
 
+    def __str__(self):
+        return self.value
+
 
 WorkloadOperationToken = str
 """

gpustack_runtime/deployer/docker.py

@@ -182,6 +182,9 @@ def parse_state(
                 return WorkloadStatusStateEnum.UNHEALTHY
             if cr.status != "running" and not _has_restart_policy(cr):
                 return WorkloadStatusStateEnum.PENDING
+            health = cr.attrs["State"].get("Health", {})
+            if health and health.get("Status") != "healthy":
+                return WorkloadStatusStateEnum.UNHEALTHY
 
         return WorkloadStatusStateEnum.RUNNING
 
@@ -444,6 +447,7 @@ def _create_pause_container(
         if workload.host_network:
             create_options["network_mode"] = "host"
         else:
+            create_options["hostname"] = workload.name
             port_mapping: dict[str, int] = {
                 # <internal port>/<protocol>: <external port>
                 f"{p.internal}/{p.protocol.lower()}": p.external or p.internal
@@ -732,9 +736,6 @@ def _create_containers(
                 },
             }
 
-            if not workload.host_network:
-                create_options["hostname"] = workload.name
-
             if workload.pid_shared:
                 create_options["pid_mode"] = pause_container_namespace
 

gpustack_runtime/deployer/kuberentes.py

@@ -59,6 +59,9 @@ class KubernetesWorkloadServiceType(str, Enum):
     LoadBalancer: Exposes the service externally using a cloud provider's load balancer.
     """
 
+    def __str__(self):
+        return self.value
+
 
 @dataclass_json
 @dataclass

gpustack_runtime/detector/__types__.py

@@ -50,6 +50,9 @@ class ManufacturerEnum(str, Enum):
     Unknown Manufacturer
     """
 
+    def __str__(self):
+        return self.value
+
 
 _MANUFACTURER_BACKEND_MAPPING: dict[ManufacturerEnum, str] = {
     ManufacturerEnum.AMD: "rocm",