Add SnapExporter class and DCGM

config.yaml

@@ -15,6 +15,12 @@ options:
     description: |
       Start the prometheus smartctl exporter at "smartctl-exporter-port". By default,
       it will start at port 10201.
+  dcgm-snap-channel:
+    type: string
+    default: "latest/stable"
+    description: |
+      Channel to install the DCGM snap if the hardware has NVIDIA GPU. By default, it will install
+      from latest/stable
   exporter-log-level:
     type: string
     default: "INFO"

src/charm.py

@@ -13,7 +13,7 @@
 from ops.model import ActiveStatus, BlockedStatus, MaintenanceStatus
 
 from hw_tools import HWTool, HWToolHelper, detect_available_tools
-from service import BaseExporter, ExporterError, HardwareExporter, SmartCtlExporter
+from service import BaseExporter, DCGMExporter, ExporterError, HardwareExporter, SmartCtlExporter
 
 logger = logging.getLogger(__name__)
 
@@ -37,6 +37,7 @@ def __init__(self, *args: Any) -> None:
             metrics_endpoints=[
                 {"path": "/metrics", "port": int(self.model.config["hardware-exporter-port"])},
                 {"path": "/metrics", "port": int(self.model.config["smartctl-exporter-port"])},
+                {"path": "/metrics", "port": 9400},
             ],
             # Setting scrape_timeout as collect_timeout in the `duration` format specified in
             # https://prometheus.io/docs/prometheus/latest/configuration/configuration/#duration
@@ -82,6 +83,9 @@ def exporters(self) -> List[BaseExporter]:
         if stored_tools & SmartCtlExporter.hw_tools():
             exporters.append(SmartCtlExporter(self.charm_dir, self.model.config))
 
+        if stored_tools & DCGMExporter.hw_tools():
+            exporters.append(DCGMExporter(self.model.config))
+
         return exporters
 
     def get_stored_tools(self) -> Set[HWTool]:
@@ -226,7 +230,7 @@ def _on_config_changed(self, event: EventBase) -> None:
                 self.model.unit.status = BlockedStatus(message)
                 return
             for exporter in self.exporters:
-                success = exporter.render_config()
+                success = exporter.configure()
                 if success:
                     exporter.restart()
                 else:

src/hw_tools.py

@@ -18,6 +18,7 @@
 import requests
 import urllib3
 from charms.operator_libs_linux.v0 import apt
+from charms.operator_libs_linux.v2 import snap
 from ops.model import ModelError, Resources
 
 import apt_helpers
@@ -187,6 +188,57 @@ def remove(self) -> None:
         """Remove details."""
 
 
+class SnapStrategy(StrategyABC):
+    """Snap strategy class."""
+
+    channel: str
+
+    @property
+    def snap(self) -> str:
+        """Snap name."""
+        return self._name.value
+
+    def install(self) -> None:
+        """Install the snap from a channel."""
+        try:
+            snap.add(self.snap, channel=self.channel)
+            logger.info("Installed %s from channel: %s", self.snap, self.channel)
+
+        # using the snap.SnapError will result into:
+        # TypeError: catching classes that do not inherit from BaseException is not allowed
+        except Exception as err:  # pylint: disable=broad-except
+            logger.error("Failed to install %s from channel: %s: %s", self.snap, self.channel, err)
+            raise err
+
+    def remove(self) -> None:
+        """Remove the snap."""
+        try:
+            snap.remove([self.snap])
+
+        # using the snap.SnapError will result into:
+        # TypeError: catching classes that do not inherit from BaseException is not allowed
+        except Exception as err:  # pylint: disable=broad-except
+            logger.error("Failed to remove %s: %s", self.snap, err)
+            raise err
+
+    def check(self) -> bool:
+        """Check if all services are active."""
+        return all(
+            service.get("active", False)
+            for service in snap.SnapCache()[self.snap].services.values()
+        )
+
+
+class DCGMExporterStrategy(SnapStrategy):
+    """DCGM strategy class."""
+
+    _name = HWTool.DCGM
+
+    def __init__(self, channel: str) -> None:
+        """Init."""
+        self.channel = channel
+
+
 class StorCLIStrategy(TPRStrategyABC):
     """Strategy to install storcli."""
 
@@ -689,13 +741,12 @@ def install(self, resources: Resources, hw_available: Set[HWTool]) -> Tuple[bool
             if strategy.name not in hw_available:
                 continue
             try:
-                # TPRStrategy
                 if isinstance(strategy, TPRStrategyABC):
                     path = fetch_tools.get(strategy.name)  # pylint: disable=W0212
                     if path:
                         strategy.install(path)
-                # APTStrategy
-                elif isinstance(strategy, APTStrategyABC):
+
+                elif isinstance(strategy, (APTStrategyABC, SnapStrategy)):
                     strategy.install()  # pylint: disable=E1120
                 logger.info("Strategy %s install success", strategy)
             except (
@@ -717,7 +768,7 @@ def remove(self, resources: Resources, hw_available: Set[HWTool]) -> None:
         for strategy in self.strategies:
             if strategy.name not in hw_available:
                 continue
-            if isinstance(strategy, (TPRStrategyABC, APTStrategyABC)):
+            if isinstance(strategy, (TPRStrategyABC, APTStrategyABC, SnapStrategy)):
                 strategy.remove()
             logger.info("Strategy %s remove success", strategy)