chore: change vm health

github-runner-manager/src/github_runner_manager/manager/runner_manager.py

@@ -13,13 +13,8 @@
 
 from github_runner_manager import constants
 from github_runner_manager.errors import GithubMetricsError, RunnerError
-from github_runner_manager.manager.cloud_runner_manager import (
-    CloudRunnerInstance,
-    CloudRunnerManager,
-    CloudRunnerState,
-    HealthState,
-)
 from github_runner_manager.manager.models import InstanceID, RunnerIdentity, RunnerMetadata
+from github_runner_manager.manager.vm_manager import VM, CloudRunnerManager, HealthState, VMState
 from github_runner_manager.metrics import events as metric_events
 from github_runner_manager.metrics import github as github_metrics
 from github_runner_manager.metrics import runner as runner_metrics
@@ -60,7 +55,7 @@ class FlushMode(Enum):
     FLUSH_BUSY = auto()
 
 
-@dataclass
+@dataclass(frozen=True)
 class RunnerInstance:
     """Represents an instance of runner.
 
@@ -70,20 +65,32 @@ class RunnerInstance:
         metadata: Metadata for the runner.
         health: The health state of the runner.
         platform_state: State on the platform.
+        platform_health: Health information queried from the platform provider.
         cloud_state: State on cloud.
     """
 
     name: str
     instance_id: InstanceID
     metadata: RunnerMetadata
-    health: HealthState
     platform_state: PlatformRunnerState | None
-    cloud_state: CloudRunnerState
+    platform_health: PlatformRunnerHealth | None
+    cloud_state: VMState
+
+    @property
+    def health(self) -> HealthState:
+        """Overall health state of the runner instance."""
+        if not self.platform_health:
+            return HealthState.UNKNOWN
+        if self.platform_health.deletable:
+            return HealthState.UNHEALTHY
+        if self.platform_health.online:
+            return HealthState.HEALTHY
+        return HealthState.UNHEALTHY
 
     @classmethod
     def from_cloud_and_platform_health(
         cls,
-        cloud_instance: CloudRunnerInstance,
+        cloud_instance: VM,
         platform_health_state: PlatformRunnerHealth | None,
     ) -> "RunnerInstance":
         """Construct an instance.
@@ -96,15 +103,15 @@ def from_cloud_and_platform_health(
             The RunnerInstance instantiated from cloud instance and platform state.
         """
         return cls(
-            name=cloud_instance.name,
+            name=cloud_instance.instance_id.name,
             instance_id=cloud_instance.instance_id,
             metadata=cloud_instance.metadata,
-            health=cloud_instance.health,
             platform_state=(
                 PlatformRunnerState.from_platform_health(platform_health_state)
                 if platform_health_state is not None
                 else None
             ),
+            platform_health=platform_health_state,
             cloud_state=cloud_instance.state,
         )
 
@@ -178,31 +185,18 @@ def get_runners(self) -> tuple[RunnerInstance, ...]:
             Information on the runners.
         """
         logger.debug("runner_manager::get_runners")
-        runner_instances = []
         cloud_runners = self._cloud.get_runners()
         runners_health_response = self._platform.get_runners_health(cloud_runners)
         logger.info("clouds runners response %s", cloud_runners)
         logger.info("runner health response %s", runners_health_response)
         runners_health = runners_health_response.requested_runners
         health_runners_map = {runner.identity.instance_id: runner for runner in runners_health}
-        for cloud_runner in cloud_runners:
-            if cloud_runner.instance_id not in health_runners_map:
-                runner_instance = RunnerInstance.from_cloud_and_platform_health(cloud_runner, None)
-                runner_instance.health = HealthState.UNKNOWN
-                runner_instances.append(runner_instance)
-                continue
-            health_runner = health_runners_map[cloud_runner.instance_id]
-            if health_runner.deletable:
-                cloud_runner.health = HealthState.UNHEALTHY
-            elif health_runner.online:
-                cloud_runner.health = HealthState.HEALTHY
-            else:
-                cloud_runner.health = HealthState.UNHEALTHY
-            runner_instance = RunnerInstance.from_cloud_and_platform_health(
-                cloud_runner, health_runner
+        return tuple(
+            RunnerInstance.from_cloud_and_platform_health(
+                cloud_runner, health_runners_map.get(cloud_runner.instance_id, None)
             )
-            runner_instances.append(runner_instance)
-        return cast(tuple[RunnerInstance], tuple(runner_instances))
+            for cloud_runner in cloud_runners
+        )
 
     def delete_runners(self, num: int) -> IssuedMetricEventsStats:
         """Delete runners.
@@ -317,7 +311,7 @@ def _cleanup_resources(
 
     def _delete_cloud_runners(
         self,
-        cloud_runners: Sequence[CloudRunnerInstance],
+        cloud_runners: Sequence[VM],
         runners_health: Sequence[PlatformRunnerHealth],
         delete_busy_runners: bool = False,
     ) -> Iterable[runner_metrics.RunnerMetrics]:
@@ -551,7 +545,7 @@ def _create_runner(args: _CreateRunnerArgs) -> InstanceID:
 
 
 def _filter_runner_to_delete(
-    cloud_runner: CloudRunnerInstance,
+    cloud_runner: VM,
     health: PlatformRunnerHealth | None,
     *,
     clean_idle: bool = False,
@@ -598,7 +592,7 @@ def _filter_runner_to_delete(
 
 
 def _runner_deletion_sort_key(
-    health_runners_map: dict[InstanceID, PlatformRunnerHealth], cloud_runner: CloudRunnerInstance
+    health_runners_map: dict[InstanceID, PlatformRunnerHealth], cloud_runner: VM
 ) -> int:
     """Order the runners in accordance to how inconvenient it is to delete them.
 

github-runner-manager/src/github_runner_manager/manager/runner_scaler.py

@@ -16,14 +16,14 @@
     MissingServerConfigError,
     ReconcileError,
 )
-from github_runner_manager.manager.cloud_runner_manager import HealthState
 from github_runner_manager.manager.runner_manager import (
     FlushMode,
     IssuedMetricEventsStats,
     RunnerInstance,
     RunnerManager,
     RunnerMetadata,
 )
+from github_runner_manager.manager.vm_manager import HealthState
 from github_runner_manager.metrics import events as metric_events
 from github_runner_manager.metrics.reconcile import (
     BUSY_RUNNERS_COUNT,

github-runner-manager/src/github_runner_manager/manager/cloud_runner_manager.py → github-runner-manager/src/github_runner_manager/manager/vm_manager.py

@@ -36,7 +36,7 @@ class HealthState(Enum):
     UNKNOWN = auto()
 
 
-class CloudRunnerState(str, Enum):
+class VMState(str, Enum):
     """Represent state of the instance hosting the runner.
 
     Attributes:
@@ -61,7 +61,7 @@ class CloudRunnerState(str, Enum):
     @staticmethod
     def from_openstack_server_status(  # pragma: no cover
         openstack_server_status: str,
-    ) -> "CloudRunnerState":
+    ) -> "VMState":
         """Create from openstack server status.
 
         The openstack server status are documented here:
@@ -73,72 +73,41 @@ def from_openstack_server_status(  # pragma: no cover
         Returns:
             The state of the runner.
         """
-        state = CloudRunnerState.UNEXPECTED
+        state = VMState.UNEXPECTED
         match openstack_server_status:
             case "BUILD":
-                state = CloudRunnerState.CREATED
+                state = VMState.CREATED
             case "REBUILD":
-                state = CloudRunnerState.CREATED
+                state = VMState.CREATED
             case "ACTIVE":
-                state = CloudRunnerState.ACTIVE
+                state = VMState.ACTIVE
             case "ERROR":
-                state = CloudRunnerState.ERROR
+                state = VMState.ERROR
             case "STOPPED":
-                state = CloudRunnerState.STOPPED
+                state = VMState.STOPPED
             case "DELETED":
-                state = CloudRunnerState.DELETED
+                state = VMState.DELETED
             case "UNKNOWN":
-                state = CloudRunnerState.UNKNOWN
+                state = VMState.UNKNOWN
             case _:
-                state = CloudRunnerState.UNEXPECTED
+                state = VMState.UNEXPECTED
         return state
 
 
-class CloudInitStatus(str, Enum):
-    """Represents the state of cloud-init script.
-
-    The cloud init script is used to launch ephemeral GitHub runners. If the script is being
-    initialized, GitHub runner is listening for jobs or GitHub runner is running the job, the
-    cloud-init script should report "running" status.
-
-    Refer to the official documentation on cloud-init status:
-    https://cloudinit.readthedocs.io/en/latest/howto/status.html.
-
-    Attributes:
-        NOT_STARTED: The cloud-init script has not yet been started.
-        RUNNING: The cloud-init script is running.
-        DONE: The cloud-init script has completed successfully.
-        ERROR: There was an error while running the cloud-init script.
-        DEGRADED: There was a non-critical issue while running the cloud-inits script.
-        DISABLED: Cloud init was disabled by other system configurations.
-    """
-
-    NOT_STARTED = "not started"
-    RUNNING = "running"
-    DONE = "done"
-    ERROR = "error"
-    DEGRADED = "degraded"
-    DISABLED = "disabled"
-
-
 @dataclass
-class CloudRunnerInstance:
+class VM:
     """Information on the runner on the cloud.
 
     Attributes:
-        name: Name of the instance hosting the runner.
-        instance_id: ID of the instance.
-        metadata: Metadata of the runner.
-        health: Health state of the runner.
-        state: State of the instance hosting the runner.
+        instance_id: VM instance identifier (NOT VM UUID).
+        metadata: Metadata associated with the VM.
+        state: The VM state.
         created_at: Creation time of the runner in the cloud provider.
     """
 
-    name: str
     instance_id: InstanceID
     metadata: RunnerMetadata
-    health: HealthState
-    state: CloudRunnerState
+    state: VMState
     created_at: datetime
 
     def is_older_than(self, seconds: float) -> bool:
@@ -248,7 +217,7 @@ def create_runner(
         self,
         runner_identity: RunnerIdentity,
         runner_context: RunnerContext,
-    ) -> CloudRunnerInstance:
+    ) -> VM:
         """Create a self-hosted runner.
 
         Args:
@@ -257,10 +226,11 @@ def create_runner(
         """
 
     @abc.abstractmethod
-    def get_runners(self) -> Sequence[CloudRunnerInstance]:
+    def get_runners(self) -> Sequence[VM]:
         """Get cloud self-hosted runners."""
 
     @abc.abstractmethod
+    # Abstract methods do not have a return value, ignore the docstring error DCO031
     def delete_vms(self, instance_ids: Sequence[InstanceID]) -> list[InstanceID]:
         """Delete cloud VM instances.
 
@@ -269,9 +239,10 @@ def delete_vms(self, instance_ids: Sequence[InstanceID]) -> list[InstanceID]:
 
         Returns:
             The deleted instance IDs.
-        """
+        """  # noqa: DCO031
 
     @abc.abstractmethod
+    # Abstract methods do not have a return value, ignore the docstring error DCO031
     def extract_metrics(self, instance_ids: Sequence[InstanceID]) -> list[RunnerMetrics]:
         """Extract metrics from cloud VMs.
 
@@ -280,7 +251,7 @@ def extract_metrics(self, instance_ids: Sequence[InstanceID]) -> list[RunnerMetr
 
         Returns:
             The fetched runner metrics.
-        """
+        """  # noqa: DCO031
 
     @abc.abstractmethod
     def cleanup(self) -> None:

github-runner-manager/src/github_runner_manager/metrics/events.py

@@ -9,7 +9,7 @@
 from pydantic import BaseModel, NonNegativeFloat
 
 from github_runner_manager.errors import IssueMetricEventError
-from github_runner_manager.manager.cloud_runner_manager import CodeInformation
+from github_runner_manager.manager.vm_manager import CodeInformation
 
 METRICS_LOG_PATH = Path("/var/log/github-runner-metrics.log")
 

github-runner-manager/src/github_runner_manager/metrics/runner.py

@@ -17,12 +17,8 @@
 from pydantic import ValidationError
 
 from github_runner_manager.errors import IssueMetricEventError, RunnerMetricsError, SSHError
-from github_runner_manager.manager.cloud_runner_manager import (
-    PostJobMetrics,
-    PreJobMetrics,
-    RunnerMetrics,
-)
 from github_runner_manager.manager.models import InstanceID
+from github_runner_manager.manager.vm_manager import PostJobMetrics, PreJobMetrics, RunnerMetrics
 from github_runner_manager.metrics import events as metric_events
 from github_runner_manager.metrics.type import GithubJobMetrics
 from github_runner_manager.openstack_cloud.constants import (

github-runner-manager/src/github_runner_manager/openstack_cloud/openstack_runner_manager.py

@@ -16,14 +16,8 @@
     OpenStackError,
     RunnerCreateError,
 )
-from github_runner_manager.manager.cloud_runner_manager import (
-    CloudRunnerInstance,
-    CloudRunnerManager,
-    CloudRunnerState,
-    RunnerMetrics,
-)
 from github_runner_manager.manager.models import InstanceID, RunnerContext, RunnerIdentity
-from github_runner_manager.manager.runner_manager import HealthState
+from github_runner_manager.manager.vm_manager import VM, CloudRunnerManager, RunnerMetrics, VMState
 from github_runner_manager.metrics import runner as runner_metrics
 from github_runner_manager.openstack_cloud.constants import (
     CREATE_SERVER_TIMEOUT,
@@ -94,7 +88,7 @@ def create_runner(
         self,
         runner_identity: RunnerIdentity,
         runner_context: RunnerContext,
-    ) -> CloudRunnerInstance:
+    ) -> VM:
         """Create a self-hosted runner.
 
         Args:
@@ -127,7 +121,7 @@ def create_runner(
         logger.info("Runner %s created successfully", instance.instance_id)
         return self._build_cloud_runner_instance(instance)
 
-    def get_runners(self) -> Sequence[CloudRunnerInstance]:
+    def get_runners(self) -> Sequence[VM]:
         """Get cloud self-hosted runners.
 
         Returns:
@@ -140,15 +134,13 @@ def cleanup(self) -> None:
         """Cleanup runner and resource on the cloud."""
         self._openstack_cloud.delete_expired_keys()
 
-    def _build_cloud_runner_instance(self, instance: OpenstackInstance) -> CloudRunnerInstance:
+    def _build_cloud_runner_instance(self, instance: OpenstackInstance) -> VM:
         """Build a new cloud runner instance from an openstack instance."""
         metadata = instance.metadata
-        return CloudRunnerInstance(
-            name=instance.instance_id.name,
+        return VM(
             metadata=metadata,
             instance_id=instance.instance_id,
-            health=HealthState.UNKNOWN,
-            state=CloudRunnerState.from_openstack_server_status(instance.status),
+            state=VMState.from_openstack_server_status(instance.status),
             created_at=instance.created_at,
         )
 

github-runner-manager/tests/unit/factories/metrics_factory.py

@@ -5,7 +5,7 @@
 
 import factory
 
-from github_runner_manager.manager.cloud_runner_manager import CodeInformation
+from github_runner_manager.manager.vm_manager import CodeInformation
 from github_runner_manager.metrics.events import Event, RunnerInstalled, RunnerStop