Configure sushy-tools BMC

stackbox/cli/__main__.py

@@ -2,6 +2,7 @@
 
 from pathlib import Path
 import sys
+import time
 
 import click
 
@@ -244,6 +245,13 @@ def init(
             config_gen = config.ConfigGenerator()
             config_gen.generate_ironic_conf(output_path=config_dir_path / "config" / "ironic.conf")
 
+            # Generate sushy-emulator.conf (BMC configuration)
+            click.echo("  Generating sushy-emulator.conf...")
+            config_gen.generate_sushy_conf(
+                output_path=config_dir_path / "config" / "sushy" / "sushy-emulator.conf",
+                debug=verbose,
+            )
+
             # Generate docker-compose.yml
             click.echo("  Generating docker-compose.yml...")
             compose_file = config_dir_path / "docker-compose.yml"
@@ -357,13 +365,88 @@ def init(
             click.echo(f"\n❌ Failed to create VM: {e}", err=True)
             sys.exit(1)
 
-        # TODO: Phase 7 (Issue #9): Configure BMC emulator
+        # Phase 7: Verify BMC can see VMs
+        click.echo("\n🔌 Verifying BMC configuration...")
+        try:
+            import requests
+
+            from stackbox.core import bmc
+
+            # Give sushy-tools a moment to discover VMs
+            time.sleep(5)
+
+            bmc_client = bmc.RedfishBMC("http://localhost:8000")
+            systems = bmc_client.get_systems()
+
+            click.echo(f"BMC discovered {len(systems)} system(s)")
+
+            # Detect TCP-specific issue: Empty systems list
+            if len(systems) == 0:
+                click.echo(
+                    "\n⚠️  BMC is running but discovered 0 systems. "
+                    "sushy-tools may not be able to connect to libvirt via TCP (qemu+tcp://libvirt:16509/system), "
+                    "or no VMs are defined in libvirt yet. This may resolve after services stabilize.",
+                    err=True,
+                )
+
+            # Verify node is visible
+            elif node_info["uuid"] in systems:
+                click.echo(f"  ✅ {node_name} ({node_info['uuid'][:8]}...)")
+            else:
+                click.echo(f"  ⚠️  {node_name} not found in BMC", err=True)
+                click.echo("     This may resolve after services stabilize")
+
+        except requests.ConnectionError as e:
+            # Scenario 1: sushy-tools container not running or unreachable
+            click.echo(
+                f"\n❌ Cannot connect to BMC service at http://localhost:8000: {e}\n"
+                "The sushy-tools container is not running, unreachable, or failed to start. "
+                "BMC must be working for node enrollment (Issue #10).",
+                err=True,
+            )
+
+        except requests.Timeout as e:
+            # Scenario 2: sushy-tools not responding
+            click.echo(
+                f"\n⚠️  BMC verification timeout: {e}\n"
+                "sushy-tools is not responding (may still be starting up). "
+                "This may resolve after services stabilize.",
+                err=True,
+            )
+
+        except requests.HTTPError as e:
+            # Scenario 3: HTTP error from sushy-tools (500, 404, etc.)
+            status_code = e.response.status_code if e.response else "unknown"
+
+            if status_code == 500:
+                click.echo(
+                    f"\n❌ BMC returned HTTP {status_code} (Internal Server Error): {e}\n"
+                    "sushy-tools encountered an internal error, possibly due to invalid configuration "
+                    "or inability to connect to libvirt via TCP. BMC must be working for node enrollment.",
+                    err=True,
+                )
+            else:
+                click.echo(
+                    f"\n❌ BMC returned HTTP {status_code}: {e}\n"
+                    "Unexpected error from sushy-tools Redfish API. BMC must be working for node enrollment.",
+                    err=True,
+                )
+
+        except Exception as e:
+            # Scenario 4: Generic error (network issues, unexpected failures)
+            click.echo(
+                f"\n⚠️  BMC verification failed: {e}\n"
+                "Unexpected error during BMC verification. This may resolve after services stabilize.",
+                err=True,
+            )
+
         # TODO: Phase 8 (Issue #10): Enroll node in Ironic
         # TODO: Phase 9 (Issue #11): Configure Tempest
 
         click.echo("\n✅ Initialization complete!")
         click.echo("   Ironic API: http://localhost:6385")
         click.echo(f"   Virtual node: {node_name}")
+        click.echo("   BMC endpoint: http://localhost:8000/redfish/v1/")
 
     def rebuild(self, service: str, no_cache: bool) -> None:
         """Rebuild a service after code changes."""

stackbox/core/bmc.py

@@ -0,0 +1,177 @@
+"""BMC (Baseboard Management Controller) utilities for Redfish."""
+
+import time
+from typing import Any
+
+import click
+import requests
+
+
+class RedfishBMC:
+    """Interact with Redfish BMC (sushy-tools)."""
+
+    def __init__(self, base_url: str = "http://localhost:8000"):
+        """Initialize Redfish client.
+
+        Args:
+            base_url: Base URL of Redfish service
+        """
+        self.base_url = base_url.rstrip("/")
+        self.session = requests.Session()
+        self.session.headers.update({"Content-Type": "application/json"})
+
+    def get_systems(self) -> list[str]:
+        """Get list of systems.
+
+        Returns:
+            List of system IDs (UUIDs)
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        resp = self.session.get(f"{self.base_url}/redfish/v1/Systems", timeout=10)
+        resp.raise_for_status()
+
+        data = resp.json()
+        members = data.get("Members", [])
+
+        return [m["@odata.id"].split("/")[-1] for m in members]
+
+    def get_system(self, system_id: str) -> dict[str, Any]:
+        """Get system information.
+
+        Args:
+            system_id: System ID (usually VM UUID)
+
+        Returns:
+            System info dict
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        resp = self.session.get(f"{self.base_url}/redfish/v1/Systems/{system_id}", timeout=10)
+        resp.raise_for_status()
+
+        return resp.json()  # type: ignore[no-any-return]
+
+    def power_on(self, system_id: str) -> None:
+        """Power on a system.
+
+        Args:
+            system_id: System ID
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        self._reset_action(system_id, "On")
+        click.echo(f"✅ Powered on system {system_id}")
+
+    def power_off(self, system_id: str, force: bool = False) -> None:
+        """Power off a system.
+
+        Args:
+            system_id: System ID
+            force: Force power off
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        action = "ForceOff" if force else "GracefulShutdown"
+        self._reset_action(system_id, action)
+        click.echo(f"✅ Powered off system {system_id}")
+
+    def _reset_action(self, system_id: str, reset_type: str) -> None:
+        """Send reset action to system.
+
+        Args:
+            system_id: System ID
+            reset_type: Reset type (On, ForceOff, GracefulShutdown, etc.)
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        resp = self.session.post(
+            f"{self.base_url}/redfish/v1/Systems/{system_id}/Actions/ComputerSystem.Reset",
+            json={"ResetType": reset_type},
+            timeout=10,
+        )
+        resp.raise_for_status()
+
+    def set_boot_device(self, system_id: str, device: str, persistent: bool = False) -> None:
+        """Set boot device.
+
+        Args:
+            system_id: System ID
+            device: Boot device (Pxe, Hdd, Cd, etc.)
+            persistent: Make setting persistent
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        enabled = "Continuous" if persistent else "Once"
+
+        resp = self.session.patch(
+            f"{self.base_url}/redfish/v1/Systems/{system_id}",
+            json={
+                "Boot": {
+                    "BootSourceOverrideTarget": device,
+                    "BootSourceOverrideEnabled": enabled,
+                }
+            },
+            timeout=10,
+        )
+        resp.raise_for_status()
+
+        click.echo(f"✅ Set boot device to {device} ({'persistent' if persistent else 'once'})")
+
+    def get_power_state(self, system_id: str) -> str:
+        """Get power state of system.
+
+        Args:
+            system_id: System ID
+
+        Returns:
+            Power state (On, Off, etc.)
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        system = self.get_system(system_id)
+        return system.get("PowerState", "Unknown")  # type: ignore[no-any-return]
+
+    def wait_for_power_state(self, system_id: str, state: str, timeout: int = 30) -> bool:
+        """Wait for system to reach power state.
+
+        Args:
+            system_id: System ID
+            state: Desired power state
+            timeout: Max time to wait in seconds
+
+        Returns:
+            True if state reached, False if timeout
+
+        Raises:
+            requests.HTTPError: If API request fails
+        """
+        start = time.time()
+
+        while time.time() - start < timeout:
+            current_state = self.get_power_state(system_id)
+            if current_state == state:
+                return True
+            time.sleep(2)
+
+        return False
+
+
+def get_bmc_address_for_node(node_info: dict[str, Any]) -> str:
+    """Get Redfish BMC address for a node.
+
+    Args:
+        node_info: Node info dict from VM creation
+
+    Returns:
+        Redfish system URL (container hostname for Ironic)
+    """
+    uuid = node_info["uuid"]
+    return f"http://sushy-tools:8000/redfish/v1/Systems/{uuid}"

stackbox/core/config.py

@@ -51,3 +51,23 @@ def generate_ironic_conf(
 
         output_path.parent.mkdir(parents=True, exist_ok=True)
         output_path.write_text(config_content)
+
+    def generate_sushy_conf(
+        self,
+        output_path: Path,
+        debug: bool = True,
+    ) -> None:
+        """Generate sushy-emulator.conf from template.
+
+        Args:
+            output_path: Where to write sushy-emulator.conf
+            debug: Enable debug logging
+        """
+        template = self.env.get_template("sushy/sushy-emulator.conf.j2")
+
+        config_content = template.render(
+            debug=debug,
+        )
+
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        output_path.write_text(config_content)

stackbox/templates/docker-compose.yml.j2

@@ -66,10 +66,13 @@ services:
   sushy-tools:
     image: quay.io/metal3-io/sushy-tools:latest
     container_name: stackbox-sushy-tools
+    command: sushy-emulator --config /etc/sushy/sushy-emulator.conf
     networks:
       - stackbox
     ports:
       - "8000:8000"
+    volumes:
+      - ./config/sushy:/etc/sushy:ro
     environment:
       - SUSHY_EMULATOR_LIBVIRT_URI=qemu+tcp://libvirt:16509/system
       - SUSHY_EMULATOR_LISTEN_IP=0.0.0.0

stackbox/templates/sushy/sushy-emulator.conf.j2

@@ -0,0 +1,32 @@
+# sushy-tools emulator configuration for containerized libvirt
+
+[DEFAULT]
+# Use libvirt backend
+SUSHY_EMULATOR_BACKEND = libvirt
+
+# Libvirt URI (TCP connection to containerized libvirt)
+# Note: Unix socket not available - libvirt runs in separate container
+SUSHY_EMULATOR_LIBVIRT_URI = qemu+tcp://libvirt:16509/system
+
+# Listen on all interfaces inside container
+SUSHY_EMULATOR_LISTEN_IP = 0.0.0.0
+SUSHY_EMULATOR_LISTEN_PORT = 8000
+
+# SSL (disabled for local dev)
+SUSHY_EMULATOR_SSL_CERT = None
+SUSHY_EMULATOR_SSL_KEY = None
+
+# Authentication (disabled for local dev - noauth in Ironic)
+SUSHY_EMULATOR_AUTH_FILE = None
+
+# Boot device persistence
+SUSHY_EMULATOR_BOOT_LOADER_MAP = {}
+
+# Ignore boot device errors
+SUSHY_EMULATOR_IGNORE_BOOT_DEVICE = False
+
+# Virtual media
+SUSHY_EMULATOR_VMEDIA_VERIFY_SSL = False
+
+# Logging
+SUSHY_EMULATOR_DEBUG = {{ debug }}