portacode 1.4.17.dev5__py3-none-any.whl → 1.4.17.dev7__py3-none-any.whl

portacode/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '1.4.17.dev5'
-__version_tuple__ = version_tuple = (1, 4, 17, 'dev5')
+__version__ = version = '1.4.17.dev7'
+__version_tuple__ = version_tuple = (1, 4, 17, 'dev7')
 
 __commit_id__ = commit_id = None

portacode/connection/handlers/proxmox_infra.py

@@ -45,16 +45,38 @@ DNS_SERVER = "1.1.1.1"
 IFACES_PATH = Path("/etc/network/interfaces")
 SYSCTL_PATH = Path("/etc/sysctl.d/99-portacode-forward.conf")
 UNIT_DIR = Path("/etc/systemd/system")
-_MANAGED_CONTAINERS_CACHE_TTL_S = 30.0
-_MANAGED_CONTAINERS_CACHE: Dict[str, Any] = {"timestamp": 0.0, "summary": None}
-_MANAGED_CONTAINERS_CACHE_LOCK = threading.Lock()
-_CAPACITY_LOCK = threading.Lock()
-_PENDING_ALLOCATIONS = {"ram_mib": 0.0, "disk_gib": 0.0, "cpu_share": 0.0}
+_MANAGED_CONTAINERS_STATE_LOCK = threading.Lock()
+_MANAGED_CONTAINERS_STATE: Dict[str, Any] = {
+    "initialized": False,
+    "base_summary": None,
+    "initial_totals": {"ram_mib": 0, "disk_gib": 0, "cpu_share": 0.0},
+    "records": {},
+    "pending": {},
+}
 TEMPLATES_REFRESH_INTERVAL_S = 300
 
 ProgressCallback = Callable[[int, int, Dict[str, Any], str, Optional[Dict[str, Any]]], None]
 
 
+def _emit_host_event(
+    handler: SyncHandler,
+    payload: Dict[str, Any],
+) -> None:
+    loop = handler.context.get("event_loop")
+    if not loop or loop.is_closed():
+        logger.debug("host event skipped (no event loop) event=%s", payload.get("event"))
+        return
+
+    future = asyncio.run_coroutine_threadsafe(handler.send_response(payload), loop)
+    future.add_done_callback(
+        lambda fut: logger.warning(
+            "Failed to emit host event %s: %s", payload.get("event"), fut.exception()
+        )
+        if fut.exception()
+        else None
+    )
+
+
 def _emit_progress_event(
     handler: SyncHandler,
     *,
@@ -422,14 +444,7 @@ def _to_mib(value: Any) -> float:
 
 
 def _pick_container_ram_mib(kind: str, cfg: Dict[str, Any], entry: Dict[str, Any]) -> float:
-    """
-    Proxmox config `memory` is already MiB for both LXC and QEMU.
-    Fall back to list fields (bytes) only when it is absent/zero.
-    """
-    mem_cfg = _safe_float(cfg.get("memory"))
-    if mem_cfg:
-        return mem_cfg
-    for candidate in (entry.get("maxmem"), entry.get("mem")):
+    for candidate in (cfg.get("memory"), entry.get("maxmem"), entry.get("mem")):
         ram = _to_mib(candidate)
         if ram:
             return ram
@@ -583,10 +598,76 @@ def _ensure_containers_dir() -> None:
     CONTAINERS_DIR.mkdir(parents=True, exist_ok=True)
 
 
-def _invalidate_managed_containers_cache() -> None:
-    with _MANAGED_CONTAINERS_CACHE_LOCK:
-        _MANAGED_CONTAINERS_CACHE["timestamp"] = 0.0
-        _MANAGED_CONTAINERS_CACHE["summary"] = None
+def _copy_summary(summary: Dict[str, Any]) -> Dict[str, Any]:
+    snapshot = summary.copy()
+    containers = summary.get("containers")
+    if isinstance(containers, list):
+        snapshot["containers"] = [entry.copy() for entry in containers]
+    unmanaged = summary.get("unmanaged_containers")
+    if isinstance(unmanaged, list):
+        snapshot["unmanaged_containers"] = [entry.copy() for entry in unmanaged]
+    return snapshot
+
+
+def _refresh_container_statuses(records: List[Dict[str, Any]], config: Dict[str, Any] | None) -> None:
+    if not records or not config:
+        return
+    try:
+        proxmox = _connect_proxmox(config)
+        node = _get_node_from_config(config)
+        statuses = {
+            str(ct.get("vmid")): (ct.get("status") or "unknown").lower()
+            for ct in proxmox.nodes(node).lxc.get()
+        }
+    except Exception as exc:  # pragma: no cover - best effort
+        logger.debug("Failed to refresh container statuses: %s", exc)
+        return
+    for record in records:
+        vmid = record.get("vmid")
+        if vmid is None:
+            continue
+        try:
+            vmid_key = str(int(vmid))
+        except (ValueError, TypeError):
+            continue
+        status = statuses.get(vmid_key)
+        if status:
+            record["status"] = status
+
+
+def _initialize_managed_containers_state(force: bool = False) -> Dict[str, Any]:
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        if _MANAGED_CONTAINERS_STATE["initialized"] and not force:
+            base = _MANAGED_CONTAINERS_STATE["base_summary"]
+            return _copy_summary(base) if base else {}
+
+    config = _load_config()
+    records = _load_managed_container_records()
+    _refresh_container_statuses(records, config)
+    base_summary = _build_full_container_summary(records, config)
+    record_map: Dict[str, Dict[str, Any]] = {}
+    for record in records:
+        vmid = record.get("vmid")
+        if vmid is None:
+            continue
+        try:
+            record_map[str(int(vmid))] = record
+        except (TypeError, ValueError):
+            continue
+
+    initial_totals = {
+        "ram_mib": int(base_summary.get("total_ram_mib") or 0),
+        "disk_gib": int(base_summary.get("total_disk_gib") or 0),
+        "cpu_share": float(base_summary.get("total_cpu_share") or 0.0),
+    }
+
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        _MANAGED_CONTAINERS_STATE["initialized"] = True
+        _MANAGED_CONTAINERS_STATE["base_summary"] = base_summary
+        _MANAGED_CONTAINERS_STATE["initial_totals"] = initial_totals
+        _MANAGED_CONTAINERS_STATE["records"] = record_map
+        _MANAGED_CONTAINERS_STATE["pending"] = {}
+    return _copy_summary(base_summary)
 
 
 def _load_managed_container_records() -> List[Dict[str, Any]]:
@@ -729,13 +810,6 @@ def _build_full_container_summary(records: List[Dict[str, Any]], config: Dict[st
                     "template": record.get("template") if record else None,
                     "created_at": record.get("created_at") if record else None,
                 }
-                # Fallback to recorded specs if live probing returned zeros (e.g., just-created CT).
-                if not merged["ram_mib"] and record:
-                    merged["ram_mib"] = record.get("ram_mib") or merged["ram_mib"]
-                if not merged["disk_gib"] and record:
-                    merged["disk_gib"] = record.get("disk_gib") or merged["disk_gib"]
-                if not merged["cpu_share"] and record:
-                    merged["cpu_share"] = record.get("cpus") or merged["cpu_share"]
                 managed_entries.append(merged)
             else:
                 unmanaged_entries.append(base_entry)
@@ -793,101 +867,131 @@ def _build_full_container_summary(records: List[Dict[str, Any]], config: Dict[st
     return summary
 
 
-def _compute_free_resources(summary: Dict[str, Any]) -> Dict[str, float]:
-    """Return free resources using the same math as the dashboard."""
-    free_ram = None
-    free_disk = None
-    free_cpu = None
-    with _CAPACITY_LOCK:
-        pending_ram = float(_PENDING_ALLOCATIONS["ram_mib"])
-        pending_disk = float(_PENDING_ALLOCATIONS["disk_gib"])
-        pending_cpu = float(_PENDING_ALLOCATIONS["cpu_share"])
-    host_ram = summary.get("host_total_ram_mib")
-    alloc_ram = summary.get("allocated_ram_mib")
-    if host_ram is not None and alloc_ram is not None:
-        free_ram = max(float(host_ram) - float(alloc_ram) - pending_ram, 0.0)
-
-    host_disk = summary.get("host_total_disk_gib")
-    alloc_disk = summary.get("allocated_disk_gib")
-    if host_disk is not None and alloc_disk is not None:
-        free_disk = max(float(host_disk) - float(alloc_disk) - pending_disk, 0.0)
-
-    host_cpu = summary.get("host_total_cpu_cores")
-    alloc_cpu = summary.get("allocated_cpu_share")
-    if host_cpu is not None and alloc_cpu is not None:
-        free_cpu = max(float(host_cpu) - float(alloc_cpu) - pending_cpu, 0.0)
-
-    return {
-        "ram_mib": free_ram,
-        "disk_gib": free_disk,
-        "cpu_share": free_cpu,
+def _pending_totals(pending: Iterable[Dict[str, Any]]) -> Tuple[int, int, float]:
+    ram_total = 0
+    disk_total = 0
+    cpu_total = 0.0
+    for entry in pending:
+        ram_total += int(entry.get("ram_mib") or 0)
+        disk_total += int(entry.get("disk_gib") or 0)
+        cpu_total += float(entry.get("cpu_share") or 0.0)
+    return ram_total, disk_total, cpu_total
+
+
+def _compose_managed_containers_summary(
+    records: Iterable[Dict[str, Any]],
+    pending: Iterable[Dict[str, Any]],
+    base_summary: Dict[str, Any],
+    initial_totals: Dict[str, Any],
+) -> Dict[str, Any]:
+    managed_summary = _build_managed_containers_summary(list(records))
+    summary = _copy_summary(base_summary)
+    base_by_vmid = {
+        str(entry.get("vmid")): entry
+        for entry in base_summary.get("containers", [])
+        if entry.get("vmid") is not None
     }
+    default_storage = summary.get("default_storage")
+    merged_containers: List[Dict[str, Any]] = []
+    for entry in managed_summary["containers"]:
+        vmid = str(entry.get("vmid")) if entry.get("vmid") is not None else None
+        base_entry = base_by_vmid.get(vmid) if vmid is not None else None
+        if base_entry:
+            merged = base_entry.copy()
+            merged.update(entry)
+        else:
+            merged = entry.copy()
+            merged.setdefault("managed", True)
+            merged.setdefault("type", "lxc")
+            if default_storage and merged.get("storage"):
+                merged["matches_default_storage"] = (
+                    str(merged["storage"]).lower() == str(default_storage).lower()
+                )
+        merged_containers.append(merged)
+    summary["updated_at"] = managed_summary["updated_at"]
+    summary["count"] = managed_summary["count"]
+    summary["total_ram_mib"] = managed_summary["total_ram_mib"]
+    summary["total_disk_gib"] = managed_summary["total_disk_gib"]
+    summary["total_cpu_share"] = managed_summary["total_cpu_share"]
+    summary["containers"] = merged_containers
+
+    pending_ram, pending_disk, pending_cpu = _pending_totals(pending)
+    delta_ram = managed_summary["total_ram_mib"] - int(initial_totals.get("ram_mib") or 0)
+    delta_disk = managed_summary["total_disk_gib"] - int(initial_totals.get("disk_gib") or 0)
+    delta_cpu = managed_summary["total_cpu_share"] - float(initial_totals.get("cpu_share") or 0.0)
+
+    if "allocated_ram_mib" in summary and summary.get("allocated_ram_mib") is not None:
+        summary["allocated_ram_mib"] = round(float(summary["allocated_ram_mib"]) + delta_ram + pending_ram, 2)
+    if "allocated_disk_gib" in summary and summary.get("allocated_disk_gib") is not None:
+        summary["allocated_disk_gib"] = round(float(summary["allocated_disk_gib"]) + delta_disk + pending_disk, 2)
+    if "allocated_cpu_share" in summary and summary.get("allocated_cpu_share") is not None:
+        summary["allocated_cpu_share"] = round(float(summary["allocated_cpu_share"]) + delta_cpu + pending_cpu, 2)
+
+    if "available_ram_mib" in summary and summary.get("available_ram_mib") is not None:
+        available_ram = float(summary["available_ram_mib"]) - delta_ram - pending_ram
+        summary["available_ram_mib"] = max(int(available_ram), 0)
+    if "available_disk_gib" in summary and summary.get("available_disk_gib") is not None:
+        available_disk = float(summary["available_disk_gib"]) - delta_disk - pending_disk
+        summary["available_disk_gib"] = max(available_disk, 0.0)
+    if "available_cpu_share" in summary and summary.get("available_cpu_share") is not None:
+        available_cpu = float(summary["available_cpu_share"]) - delta_cpu - pending_cpu
+        summary["available_cpu_share"] = max(available_cpu, 0.0)
 
+    return summary
 
-def _assert_capacity_for_payload(payload: Dict[str, Any], summary: Dict[str, Any]) -> None:
-    """Validate requested container resources against current free capacity."""
-    free = _compute_free_resources(summary)
-    shortages: List[str] = []
-
-    req_ram = float(payload.get("ram_mib", 0))
-    free_ram = free.get("ram_mib")
-    if free_ram is not None and req_ram > free_ram:
-        shortages.append(f"RAM (need {int(req_ram)} MiB, free {int(free_ram)} MiB)")
-
-    req_disk = float(payload.get("disk_gib", 0))
-    free_disk = free.get("disk_gib")
-    if free_disk is not None and req_disk > free_disk:
-        shortages.append(f"Disk (need {req_disk:.1f} GiB, free {free_disk:.1f} GiB)")
-
-    req_cpu = float(payload.get("cpus", 0))
-    free_cpu = free.get("cpu_share")
-    if free_cpu is not None and req_cpu > free_cpu:
-        shortages.append(f"CPU (need {req_cpu:.2f} vCPU, free {free_cpu:.2f} vCPU)")
 
-    if shortages:
-        raise RuntimeError(f"Insufficient resources: {', '.join(shortages)}")
+def _get_managed_containers_summary(force: bool = False) -> Dict[str, Any]:
+    _initialize_managed_containers_state(force=force)
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        base_summary = _MANAGED_CONTAINERS_STATE.get("base_summary") or {}
+        records = list(_MANAGED_CONTAINERS_STATE.get("records", {}).values())
+        pending = list(_MANAGED_CONTAINERS_STATE.get("pending", {}).values())
+        initial_totals = _MANAGED_CONTAINERS_STATE.get("initial_totals", {})
+    if not base_summary:
+        return {}
+    return _compose_managed_containers_summary(records, pending, base_summary, initial_totals)
+
+
+def _reserve_container_resources(payload: Dict[str, Any], *, device_id: str, request_id: Optional[str]) -> str:
+    _initialize_managed_containers_state()
+    ram_mib = int(payload.get("ram_mib") or 0)
+    disk_gib = int(payload.get("disk_gib") or 0)
+    cpu_share = float(payload.get("cpus") or 0.0)
+    reservation_id = secrets.token_hex(8)
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        base_summary = _MANAGED_CONTAINERS_STATE.get("base_summary") or {}
+        records = list(_MANAGED_CONTAINERS_STATE.get("records", {}).values())
+        pending = list(_MANAGED_CONTAINERS_STATE.get("pending", {}).values())
+        initial_totals = _MANAGED_CONTAINERS_STATE.get("initial_totals", {})
+        summary = _compose_managed_containers_summary(records, pending, base_summary, initial_totals)
+
+        available_ram = summary.get("available_ram_mib")
+        if available_ram is not None and ram_mib > available_ram:
+            raise RuntimeError("Not enough RAM to create this container.")
+        available_disk = summary.get("available_disk_gib")
+        if available_disk is not None and disk_gib > available_disk:
+            raise RuntimeError("Not enough disk space to create this container.")
+        available_cpu = summary.get("available_cpu_share")
+        if available_cpu is not None and cpu_share > available_cpu:
+            raise RuntimeError("Not enough CPU capacity to create this container.")
+
+        _MANAGED_CONTAINERS_STATE["pending"][reservation_id] = {
+            "reservation_id": reservation_id,
+            "device_id": device_id,
+            "request_id": request_id,
+            "ram_mib": ram_mib,
+            "disk_gib": disk_gib,
+            "cpu_share": cpu_share,
+            "created_at": datetime.utcnow().isoformat() + "Z",
+        }
+    return reservation_id
 
 
-def _get_managed_containers_summary(force: bool = False) -> Dict[str, Any]:
-    def _refresh_container_statuses(records: List[Dict[str, Any]], config: Dict[str, Any] | None) -> None:
-        if not records or not config:
-            return
-        try:
-            proxmox = _connect_proxmox(config)
-            node = _get_node_from_config(config)
-            statuses = {
-                str(ct.get("vmid")): (ct.get("status") or "unknown").lower()
-                for ct in proxmox.nodes(node).lxc.get()
-            }
-        except Exception as exc:  # pragma: no cover - best effort
-            logger.debug("Failed to refresh container statuses: %s", exc)
-            return
-        for record in records:
-            vmid = record.get("vmid")
-            if vmid is None:
-                continue
-            try:
-                vmid_key = str(int(vmid))
-            except (ValueError, TypeError):
-                continue
-            status = statuses.get(vmid_key)
-            if status:
-                record["status"] = status
-
-    now = time.monotonic()
-    with _MANAGED_CONTAINERS_CACHE_LOCK:
-        cache_ts = _MANAGED_CONTAINERS_CACHE["timestamp"]
-        cached = _MANAGED_CONTAINERS_CACHE["summary"]
-    if not force and cached and now - cache_ts < _MANAGED_CONTAINERS_CACHE_TTL_S:
-        return cached
-    config = _load_config()
-    records = _load_managed_container_records()
-    _refresh_container_statuses(records, config)
-    summary = _build_full_container_summary(records, config)
-    with _MANAGED_CONTAINERS_CACHE_LOCK:
-        _MANAGED_CONTAINERS_CACHE["timestamp"] = now
-        _MANAGED_CONTAINERS_CACHE["summary"] = summary
-    return summary
+def _release_container_reservation(reservation_id: Optional[str]) -> None:
+    if not reservation_id:
+        return
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        _MANAGED_CONTAINERS_STATE.get("pending", {}).pop(reservation_id, None)
 
 
 def _format_rootfs(storage: str, disk_gib: int, storage_type: str) -> str:
@@ -1173,31 +1277,8 @@ def _start_container(proxmox: Any, node: str, vmid: int) -> Tuple[Dict[str, Any]
         logger.info("Container %s already running (%ss)", vmid, uptime)
         return status, 0.0
 
-    # Validate capacity using the same math as the dashboard and serialize allocation.
-    summary = _get_managed_containers_summary(force=True)
-    cfg = proxmox.nodes(node).lxc(vmid).config.get()
-    payload = {
-        "ram_mib": _to_mib(cfg.get("memory")),
-        "disk_gib": _pick_container_disk_gib("lxc", cfg, {"rootfs": cfg.get("rootfs")}),
-        "cpus": _pick_container_cpu_share("lxc", cfg, {}),
-    }
-    with _CAPACITY_LOCK:
-        try:
-            _assert_capacity_for_payload(payload, summary)
-        except RuntimeError as exc:
-            raise RuntimeError(f"Not enough resources to start container {vmid}: {exc}") from exc
-        _PENDING_ALLOCATIONS["ram_mib"] += float(payload["ram_mib"] or 0)
-        _PENDING_ALLOCATIONS["disk_gib"] += float(payload["disk_gib"] or 0)
-        _PENDING_ALLOCATIONS["cpu_share"] += float(payload["cpus"] or 0)
-
-    try:
-        upid = proxmox.nodes(node).lxc(vmid).status.start.post()
-        return _wait_for_task(proxmox, node, upid)
-    finally:
-        with _CAPACITY_LOCK:
-            _PENDING_ALLOCATIONS["ram_mib"] -= float(payload["ram_mib"] or 0)
-            _PENDING_ALLOCATIONS["disk_gib"] -= float(payload["disk_gib"] or 0)
-            _PENDING_ALLOCATIONS["cpu_share"] -= float(payload["cpus"] or 0)
+    upid = proxmox.nodes(node).lxc(vmid).status.start.post()
+    return _wait_for_task(proxmox, node, upid)
 
 
 def _stop_container(proxmox: Any, node: str, vmid: int) -> Tuple[Dict[str, Any], float]:
@@ -1213,11 +1294,25 @@ def _delete_container(proxmox: Any, node: str, vmid: int) -> Tuple[Dict[str, Any
     return _wait_for_task(proxmox, node, upid)
 
 
-def _write_container_record(vmid: int, payload: Dict[str, Any]) -> None:
+def _register_container_record(vmid: int, payload: Dict[str, Any], reservation_id: Optional[str] = None) -> None:
+    _initialize_managed_containers_state()
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        if reservation_id:
+            _MANAGED_CONTAINERS_STATE["pending"].pop(reservation_id, None)
+        _MANAGED_CONTAINERS_STATE["records"][str(vmid)] = payload.copy()
+
+
+def _unregister_container_record(vmid: int) -> None:
+    _initialize_managed_containers_state()
+    with _MANAGED_CONTAINERS_STATE_LOCK:
+        _MANAGED_CONTAINERS_STATE["records"].pop(str(vmid), None)
+
+
+def _write_container_record(vmid: int, payload: Dict[str, Any], reservation_id: Optional[str] = None) -> None:
     _ensure_containers_dir()
     path = CONTAINERS_DIR / f"ct-{vmid}.json"
     path.write_text(json.dumps(payload, indent=2), encoding="utf-8")
-    _invalidate_managed_containers_cache()
+    _register_container_record(vmid, payload, reservation_id=reservation_id)
 
 
 def _read_container_record(vmid: int) -> Dict[str, Any]:
@@ -1237,7 +1332,7 @@ def _remove_container_record(vmid: int) -> None:
     path = CONTAINERS_DIR / f"ct-{vmid}.json"
     if path.exists():
         path.unlink()
-        _invalidate_managed_containers_cache()
+    _unregister_container_record(vmid)
 
 
 def _build_container_payload(message: Dict[str, Any], config: Dict[str, Any]) -> Dict[str, Any]:
@@ -1785,12 +1880,14 @@ def _allocate_vmid(proxmox: Any) -> int:
     return int(proxmox.cluster.nextid.get())
 
 
-def _instantiate_container(proxmox: Any, node: str, payload: Dict[str, Any]) -> Tuple[int, float]:
+def _instantiate_container(
+    proxmox: Any, node: str, payload: Dict[str, Any], vmid: Optional[int] = None
+) -> Tuple[int, float]:
     from proxmoxer.core import ResourceException
 
     storage_type = _get_storage_type(proxmox.nodes(node).storage.get(), payload["storage"])
     rootfs = _format_rootfs(payload["storage"], payload["disk_gib"], storage_type)
-    vmid = _allocate_vmid(proxmox)
+    vmid = vmid or _allocate_vmid(proxmox)
     if not payload.get("hostname"):
         payload["hostname"] = f"ct{vmid}"
     try:
@@ -1822,6 +1919,43 @@ def _instantiate_container(proxmox: Any, node: str, payload: Dict[str, Any]) ->
         raise RuntimeError(f"Failed to create container: {exc}") from exc
 
 
+def _cleanup_failed_container(
+    proxmox: Any, node: str, vmid: int, provisioning_id: Optional[str]
+) -> None:
+    from proxmoxer.core import ResourceException
+
+    try:
+        cfg = proxmox.nodes(node).lxc(str(vmid)).config.get()
+    except ResourceException as exc:
+        msg = str(exc).lower()
+        if "does not exist" in msg or "not found" in msg:
+            return
+        logger.warning("Failed to inspect container %s after create failure: %s", vmid, exc)
+        return
+    except Exception as exc:  # pragma: no cover - best effort
+        logger.warning("Failed to inspect container %s after create failure: %s", vmid, exc)
+        return
+
+    description = (cfg or {}).get("description") or ""
+    if provisioning_id and provisioning_id not in description:
+        logger.warning(
+            "Skipping cleanup for vmid=%s; provisioning marker mismatch", vmid
+        )
+        return
+
+    try:
+        status = proxmox.nodes(node).lxc(str(vmid)).status.current.get()
+        if status.get("status") == "running":
+            _stop_container(proxmox, node, vmid)
+    except Exception as exc:  # pragma: no cover - best effort
+        logger.warning("Failed to stop container %s after create failure: %s", vmid, exc)
+
+    try:
+        _delete_container(proxmox, node, vmid)
+    except Exception as exc:  # pragma: no cover - best effort
+        logger.warning("Failed to delete container %s after create failure: %s", vmid, exc)
+
+
 class CreateProxmoxContainerHandler(SyncHandler):
     """Provision a new managed LXC container via the Proxmox API."""
 
@@ -1924,239 +2058,266 @@ class CreateProxmoxContainerHandler(SyncHandler):
             _validate_environment,
         )
 
-def _create_container():
-    proxmox = _connect_proxmox(config)
-    node = config.get("node") or DEFAULT_NODE_NAME
-    payload = _build_container_payload(message, config)
-    payload["cpulimit"] = float(payload["cpus"])
-    payload["cores"] = int(max(math.ceil(payload["cpus"]), 1))
-    payload["memory"] = int(payload["ram_mib"])
-    payload["node"] = node
-    # Validate against current free resources (same math as dashboard charts) and place a short-lived reservation.
-    summary = _get_managed_containers_summary(force=True)
-    with _CAPACITY_LOCK:
-        try:
-            _assert_capacity_for_payload(payload, summary)
-        except RuntimeError as exc:
-            raise RuntimeError(f"Not enough resources to create the container safely: {exc}") from exc
-        _PENDING_ALLOCATIONS["ram_mib"] += float(payload["ram_mib"] or 0)
-        _PENDING_ALLOCATIONS["disk_gib"] += float(payload["disk_gib"] or 0)
-        _PENDING_ALLOCATIONS["cpu_share"] += float(payload["cpus"] or 0)
-    try:
-        vmid, _ = _instantiate_container(proxmox, node, payload)
-    finally:
-        with _CAPACITY_LOCK:
-            _PENDING_ALLOCATIONS["ram_mib"] -= float(payload["ram_mib"] or 0)
-            _PENDING_ALLOCATIONS["disk_gib"] -= float(payload["disk_gib"] or 0)
-            _PENDING_ALLOCATIONS["cpu_share"] -= float(payload["cpus"] or 0)
-        logger.debug(
-            "Provisioning container node=%s template=%s ram=%s cpu=%s storage=%s",
-            node,
-            payload["template"],
-            payload["ram_mib"],
-            payload["cpus"],
-            payload["storage"],
-        )
-        payload["vmid"] = vmid
-        payload["created_at"] = datetime.utcnow().isoformat() + "Z"
-        payload["status"] = "creating"
-        payload["device_id"] = device_id
-        _write_container_record(vmid, payload)
-        return proxmox, node, vmid, payload
-
-        proxmox, node, vmid, payload = _run_lifecycle_step(
-            "create_container",
-            "Creating container",
-            "Provisioning the LXC container…",
-            "Container created.",
-            _create_container,
-        )
-
-        def _start_container_step():
-            _start_container(proxmox, node, vmid)
-
-        _run_lifecycle_step(
-            "start_container",
-            "Starting container",
-            "Booting the container…",
-            "Container startup completed.",
-            _start_container_step,
-        )
-        _update_container_record(vmid, {"status": "running"})
+        reservation_id: Optional[str] = None
+        provisioning_id: Optional[str] = None
+        created_record = False
+        proxmox: Any = None
+        node: Optional[str] = None
+        vmid: Optional[int] = None
 
-        def _bootstrap_progress_callback(
-            step_index: int,
-            total: int,
-            step: Dict[str, Any],
-            status: str,
-            result: Optional[Dict[str, Any]],
-        ):
-            label = step.get("display_name") or _friendly_step_label(step.get("name", "bootstrap"))
-            error_summary = (result or {}).get("error_summary") or (result or {}).get("error")
-            attempt = (result or {}).get("attempt")
-            if status == "in_progress":
-                message_text = f"{label} is running…"
-            elif status == "completed":
-                message_text = f"{label} completed."
-            elif status == "retrying":
-                attempt_desc = f" (attempt {attempt})" if attempt else ""
-                message_text = f"{label} failed{attempt_desc}; retrying…"
-            else:
-                message_text = f"{label} failed"
-                if error_summary:
-                    message_text += f": {error_summary}"
-            details: Dict[str, Any] = {}
-            if attempt:
-                details["attempt"] = attempt
-            if error_summary:
-                details["error_summary"] = error_summary
-            _emit_progress_event(
-                self,
-                step_index=step_index,
-                total_steps=total,
-                step_name=step.get("name", "bootstrap"),
-                step_label=label,
-                status=status,
-                message=message_text,
-                phase="bootstrap",
-                request_id=request_id,
-                details=details or None,
-                on_behalf_of_device=device_id,
+        def _create_container():
+            nonlocal reservation_id, provisioning_id, created_record, proxmox, node, vmid
+            proxmox = _connect_proxmox(config)
+            node = config.get("node") or DEFAULT_NODE_NAME
+            payload = _build_container_payload(message, config)
+            payload["cpulimit"] = float(payload["cpus"])
+            payload["cores"] = int(max(math.ceil(payload["cpus"]), 1))
+            payload["memory"] = int(payload["ram_mib"])
+            payload["node"] = node
+            reservation_id = _reserve_container_resources(
+                payload, device_id=device_id, request_id=request_id
             )
-
-        public_key, steps = _bootstrap_portacode(
-            vmid,
-            payload["username"],
-            payload["password"],
-            payload["ssh_public_key"],
-            steps=bootstrap_steps,
-            progress_callback=_bootstrap_progress_callback,
-            start_index=current_step_index,
-            total_steps=total_steps,
-            default_public_key=device_public_key if has_device_keypair else None,
-        )
-        current_step_index += len(bootstrap_steps)
-
-        service_installed = False
-        if has_device_keypair:
-            logger.info(
-                "deploying dashboard-provided Portacode keypair (device_id=%s) into container %s",
-                device_id,
-                vmid,
+            _emit_host_event(
+                self,
+                {
+                    "event": "proxmox_container_accepted",
+                    "success": True,
+                    "message": "Provisioning accepted; resources reserved.",
+                    "device_id": device_id,
+                    "request_id": request_id,
+                },
             )
-            _deploy_device_keypair(
-                vmid,
-                payload["username"],
-                device_private_key,
-                device_public_key,
+            provisioning_id = secrets.token_hex(6)
+            payload["description"] = f"{payload.get('description', MANAGED_MARKER)};provisioning_id={provisioning_id}"
+            logger.debug(
+                "Provisioning container node=%s template=%s ram=%s cpu=%s storage=%s",
+                node,
+                payload["template"],
+                payload["ram_mib"],
+                payload["cpus"],
+                payload["storage"],
             )
-            service_installed = True
-            service_start_index = current_step_index
+            try:
+                vmid = _allocate_vmid(proxmox)
+                vmid, _ = _instantiate_container(proxmox, node, payload, vmid=vmid)
+            except Exception:
+                _release_container_reservation(reservation_id)
+                if vmid is not None:
+                    _cleanup_failed_container(proxmox, node, vmid, provisioning_id)
+                raise
+            payload["vmid"] = vmid
+            payload["created_at"] = datetime.utcnow().isoformat() + "Z"
+            payload["status"] = "creating"
+            payload["device_id"] = device_id
+            try:
+                _write_container_record(vmid, payload, reservation_id=reservation_id)
+                created_record = True
+            except Exception:
+                _release_container_reservation(reservation_id)
+                _cleanup_failed_container(proxmox, node, vmid, provisioning_id)
+                raise
+            return proxmox, node, vmid, payload
 
-            auth_step_name = "setup_device_authentication"
-            auth_label = "Setting up device authentication"
-            _emit_progress_event(
-                self,
-                step_index=service_start_index,
-                total_steps=total_steps,
-                step_name=auth_step_name,
-                step_label=auth_label,
-                status="in_progress",
-                message="Notifying the server of the new device…",
-                phase="service",
-                request_id=request_id,
-                on_behalf_of_device=device_id,
+        try:
+            proxmox, node, vmid, payload = _run_lifecycle_step(
+                "create_container",
+                "Creating container",
+                "Provisioning the LXC container…",
+                "Container created.",
+                _create_container,
             )
-            _emit_progress_event(
-                self,
-                step_index=service_start_index,
-                total_steps=total_steps,
-                step_name=auth_step_name,
-                step_label=auth_label,
-                status="completed",
-                message="Authentication metadata recorded.",
-                phase="service",
-                request_id=request_id,
-                on_behalf_of_device=device_id,
+
+            def _start_container_step():
+                _start_container(proxmox, node, vmid)
+
+            _run_lifecycle_step(
+                "start_container",
+                "Starting container",
+                "Booting the container…",
+                "Container startup completed.",
+                _start_container_step,
             )
+            _update_container_record(vmid, {"status": "running"})
+
+            def _bootstrap_progress_callback(
+                step_index: int,
+                total: int,
+                step: Dict[str, Any],
+                status: str,
+                result: Optional[Dict[str, Any]],
+            ):
+                label = step.get("display_name") or _friendly_step_label(step.get("name", "bootstrap"))
+                error_summary = (result or {}).get("error_summary") or (result or {}).get("error")
+                attempt = (result or {}).get("attempt")
+                if status == "in_progress":
+                    message_text = f"{label} is running…"
+                elif status == "completed":
+                    message_text = f"{label} completed."
+                elif status == "retrying":
+                    attempt_desc = f" (attempt {attempt})" if attempt else ""
+                    message_text = f"{label} failed{attempt_desc}; retrying…"
+                else:
+                    message_text = f"{label} failed"
+                    if error_summary:
+                        message_text += f": {error_summary}"
+                details: Dict[str, Any] = {}
+                if attempt:
+                    details["attempt"] = attempt
+                if error_summary:
+                    details["error_summary"] = error_summary
+                _emit_progress_event(
+                    self,
+                    step_index=step_index,
+                    total_steps=total,
+                    step_name=step.get("name", "bootstrap"),
+                    step_label=label,
+                    status=status,
+                    message=message_text,
+                    phase="bootstrap",
+                    request_id=request_id,
+                    details=details or None,
+                    on_behalf_of_device=device_id,
+                )
 
-            install_step = service_start_index + 1
-            install_label = "Launching Portacode service"
-            _emit_progress_event(
-                self,
-                step_index=install_step,
+            public_key, steps = _bootstrap_portacode(
+                vmid,
+                payload["username"],
+                payload["password"],
+                payload["ssh_public_key"],
+                steps=bootstrap_steps,
+                progress_callback=_bootstrap_progress_callback,
+                start_index=current_step_index,
                 total_steps=total_steps,
-                step_name="launch_portacode_service",
-                step_label=install_label,
-                status="in_progress",
-                message="Running sudo portacode service install…",
-                phase="service",
-                request_id=request_id,
-                on_behalf_of_device=device_id,
+                default_public_key=device_public_key if has_device_keypair else None,
             )
+            current_step_index += len(bootstrap_steps)
+
+            service_installed = False
+            if has_device_keypair:
+                logger.info(
+                    "deploying dashboard-provided Portacode keypair (device_id=%s) into container %s",
+                    device_id,
+                    vmid,
+                )
+                _deploy_device_keypair(
+                    vmid,
+                    payload["username"],
+                    device_private_key,
+                    device_public_key,
+                )
+                service_installed = True
+                service_start_index = current_step_index
 
-            cmd = _su_command(payload["username"], "sudo -S portacode service install")
-            res = _run_pct(vmid, cmd, input_text=payload["password"] + "\n")
+                auth_step_name = "setup_device_authentication"
+                auth_label = "Setting up device authentication"
+                _emit_progress_event(
+                    self,
+                    step_index=service_start_index,
+                    total_steps=total_steps,
+                    step_name=auth_step_name,
+                    step_label=auth_label,
+                    status="in_progress",
+                    message="Notifying the server of the new device…",
+                    phase="service",
+                    request_id=request_id,
+                    on_behalf_of_device=device_id,
+                )
+                _emit_progress_event(
+                    self,
+                    step_index=service_start_index,
+                    total_steps=total_steps,
+                    step_name=auth_step_name,
+                    step_label=auth_label,
+                    status="completed",
+                    message="Authentication metadata recorded.",
+                    phase="service",
+                    request_id=request_id,
+                    on_behalf_of_device=device_id,
+                )
 
-            if res["returncode"] != 0:
+                install_step = service_start_index + 1
+                install_label = "Launching Portacode service"
                 _emit_progress_event(
                     self,
                     step_index=install_step,
                     total_steps=total_steps,
                     step_name="launch_portacode_service",
                     step_label=install_label,
-                    status="failed",
-                    message=f"{install_label} failed: {res.get('stderr') or res.get('stdout')}",
+                    status="in_progress",
+                    message="Running sudo portacode service install…",
                     phase="service",
                     request_id=request_id,
-                    details={
-                        "stderr": res.get("stderr"),
-                        "stdout": res.get("stdout"),
-                    },
                     on_behalf_of_device=device_id,
                 )
-                raise RuntimeError(res.get("stderr") or res.get("stdout") or "Service install failed")
 
-            _emit_progress_event(
-                self,
-                step_index=install_step,
-                total_steps=total_steps,
-                step_name="launch_portacode_service",
-                step_label=install_label,
-                status="completed",
-                message="Portacode service install finished.",
-                phase="service",
-                request_id=request_id,
-                on_behalf_of_device=device_id,
-            )
+                cmd = _su_command(payload["username"], "sudo -S portacode service install")
+                res = _run_pct(vmid, cmd, input_text=payload["password"] + "\n")
+
+                if res["returncode"] != 0:
+                    _emit_progress_event(
+                        self,
+                        step_index=install_step,
+                        total_steps=total_steps,
+                        step_name="launch_portacode_service",
+                        step_label=install_label,
+                        status="failed",
+                        message=f"{install_label} failed: {res.get('stderr') or res.get('stdout')}",
+                        phase="service",
+                        request_id=request_id,
+                        details={
+                            "stderr": res.get("stderr"),
+                            "stdout": res.get("stdout"),
+                        },
+                        on_behalf_of_device=device_id,
+                    )
+                    raise RuntimeError(res.get("stderr") or res.get("stdout") or "Service install failed")
 
-            logger.info("create_proxmox_container: portacode service install completed inside ct %s", vmid)
+                _emit_progress_event(
+                    self,
+                    step_index=install_step,
+                    total_steps=total_steps,
+                    step_name="launch_portacode_service",
+                    step_label=install_label,
+                    status="completed",
+                    message="Portacode service install finished.",
+                    phase="service",
+                    request_id=request_id,
+                    on_behalf_of_device=device_id,
+                )
 
-            current_step_index += 2
+                logger.info(
+                    "create_proxmox_container: portacode service install completed inside ct %s", vmid
+                )
 
-        response = {
-            "event": "proxmox_container_created",
-            "success": True,
-            "message": f"Container {vmid} is ready and Portacode key captured.",
-            "ctid": str(vmid),
-            "public_key": public_key,
-            "container": {
-                "vmid": vmid,
-                "hostname": payload["hostname"],
-                "template": payload["template"],
-                "storage": payload["storage"],
-                "disk_gib": payload["disk_gib"],
-                "ram_mib": payload["ram_mib"],
-                "cpus": payload["cpus"],
-            },
-            "setup_steps": steps,
-            "device_id": device_id,
-            "on_behalf_of_device": device_id,
-            "service_installed": service_installed,
-        }
-        if not response:
-            raise RuntimeError("create_proxmox_container produced no response payload")
-        return response
+                current_step_index += 2
+
+            return {
+                "event": "proxmox_container_created",
+                "success": True,
+                "message": f"Container {vmid} is ready and Portacode key captured.",
+                "ctid": str(vmid),
+                "public_key": public_key,
+                "container": {
+                    "vmid": vmid,
+                    "hostname": payload["hostname"],
+                    "template": payload["template"],
+                    "storage": payload["storage"],
+                    "disk_gib": payload["disk_gib"],
+                    "ram_mib": payload["ram_mib"],
+                    "cpus": payload["cpus"],
+                },
+                "setup_steps": steps,
+                "device_id": device_id,
+                "on_behalf_of_device": device_id,
+                "service_installed": service_installed,
+            }
+        except Exception:
+            if reservation_id and not created_record:
+                _release_container_reservation(reservation_id)
+            if vmid is not None and proxmox and node:
+                _cleanup_failed_container(proxmox, node, vmid, provisioning_id)
+                _remove_container_record(vmid)
+            raise
 
 
 class StartPortacodeServiceHandler(SyncHandler):

{portacode-1.4.17.dev5.dist-info → portacode-1.4.17.dev7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: portacode
-Version: 1.4.17.dev5
+Version: 1.4.17.dev7
 Summary: Portacode CLI client and SDK
 Home-page: https://github.com/portacode/portacode
 Author: Meena Erian

{portacode-1.4.17.dev5.dist-info → portacode-1.4.17.dev7.dist-info}/RECORD

@@ -1,7 +1,7 @@
 portacode/README.md,sha256=4dKtpvR8LNgZPVz37GmkQCMWIr_u25Ao63iW56s7Ke4,775
 portacode/__init__.py,sha256=oB3sV1wXr-um-RXio73UG8E5Xx6cF2ZVJveqjNmC-vQ,1086
 portacode/__main__.py,sha256=jmHTGC1hzmo9iKJLv-SSYe9BSIbPPZ2IOpecI03PlTs,296
-portacode/_version.py,sha256=36QhMCQ9bWTAYLn9nL0QaZvzLw-wFKGVbWJ4Lf_WCk8,719
+portacode/_version.py,sha256=Inn3jAXtKmJcxP1Ht9pcGtSS3m4h3OsmOpgsOB9feYM,719
 portacode/cli.py,sha256=mGLKoZ-T2FBF7IA9wUq0zyG0X9__-A1ao7gajjcVRH8,21828
 portacode/data.py,sha256=5-s291bv8J354myaHm1Y7CQZTZyRzMU3TGe5U4hb-FA,1591
 portacode/keypair.py,sha256=0OO4vHDcF1XMxCDqce61xFTlFwlTcmqe5HyGsXFEt7s,5838
@@ -22,7 +22,7 @@ portacode/connection/handlers/diff_handlers.py,sha256=iYTIRCcpEQ03vIPKZCsMTE5aZb
 portacode/connection/handlers/file_handlers.py,sha256=nAJH8nXnX07xxD28ngLpgIUzcTuRwZBNpEGEKdRqohw,39507
 portacode/connection/handlers/project_aware_file_handlers.py,sha256=AqgMnDqX2893T2NsrvUSCwjN5VKj4Pb2TN0S_SuboOE,9803
 portacode/connection/handlers/project_state_handlers.py,sha256=v6ZefGW9i7n1aZLq2jOGumJIjYb6aHlPI4m1jkYewm8,1686
-portacode/connection/handlers/proxmox_infra.py,sha256=JK5b3oQUB21brvYpFBhjBjVQwpYes_a7bPawqkjklEg,89628
+portacode/connection/handlers/proxmox_infra.py,sha256=hsEsulBU55bLF5j5ACDLo-51hZe8rMKj5bdB5n5jv9Y,97125
 portacode/connection/handlers/registry.py,sha256=qXGE60sYEWg6ZtVQzFcZ5YI2XWR6lMgw4hAL9x5qR1I,6181
 portacode/connection/handlers/session.py,sha256=uNGfiO_1B9-_yjJKkpvmbiJhIl6b-UXlT86UTfd6WYE,42219
 portacode/connection/handlers/system_handlers.py,sha256=fr12QpOr_Z8KYGUU-AYrTQwRPAcrLK85hvj3SEq1Kw8,14757
@@ -65,7 +65,7 @@ portacode/utils/__init__.py,sha256=NgBlWTuNJESfIYJzP_3adI1yJQJR0XJLRpSdVNaBAN0,3
 portacode/utils/diff_apply.py,sha256=4Oi7ft3VUCKmiUE4VM-OeqO7Gk6H7PF3WnN4WHXtjxI,15157
 portacode/utils/diff_renderer.py,sha256=S76StnQ2DLfsz4Gg0m07UwPfRp8270PuzbNaQq-rmYk,13850
 portacode/utils/ntp_clock.py,sha256=VqCnWCTehCufE43W23oB-WUdAZGeCcLxkmIOPwInYHc,2499
-portacode-1.4.17.dev5.dist-info/licenses/LICENSE,sha256=2FGbCnUDgRYuQTkB1O1dUUpu5CVAjK1j4_p6ack9Z54,1066
+portacode-1.4.17.dev7.dist-info/licenses/LICENSE,sha256=2FGbCnUDgRYuQTkB1O1dUUpu5CVAjK1j4_p6ack9Z54,1066
 test_modules/README.md,sha256=Do_agkm9WhSzueXjRAkV_xEj6Emy5zB3N3VKY5Roce8,9274
 test_modules/__init__.py,sha256=1LcbHodIHsB0g-g4NGjSn6AMuCoGbymvXPYLOb6Z7F0,53
 test_modules/test_device_online.py,sha256=QtYq0Dq9vME8Gp2O4fGSheqVf8LUtpsSKosXXk56gGM,1654
@@ -91,8 +91,8 @@ testing_framework/core/playwright_manager.py,sha256=Tw46qwxIhOFkS48C2IWIQHHNpEe-
 testing_framework/core/runner.py,sha256=j2QwNJmAxVBmJvcbVS7DgPJUKPNzqfLmt_4NNdaKmZU,19297
 testing_framework/core/shared_cli_manager.py,sha256=BESSNtyQb7BOlaOvZmm04T8Uezjms4KCBs2MzTxvzYQ,8790
 testing_framework/core/test_discovery.py,sha256=2FZ9fJ8Dp5dloA-fkgXoJ_gCMC_nYPBnA3Hs2xlagzM,4928
-portacode-1.4.17.dev5.dist-info/METADATA,sha256=LvTHxhgWmfgvPgDYRAt2H3eOg0Fi8I4R-4dYtFLyt1E,13051
-portacode-1.4.17.dev5.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-portacode-1.4.17.dev5.dist-info/entry_points.txt,sha256=lLUUL-BM6_wwe44Xv0__5NQ1BnAz6jWjSMFvZdWW3zU,48
-portacode-1.4.17.dev5.dist-info/top_level.txt,sha256=TGhTYUxfW8SyVZc_zGgzjzc24gGT7nSw8Qf73liVRKM,41
-portacode-1.4.17.dev5.dist-info/RECORD,,
+portacode-1.4.17.dev7.dist-info/METADATA,sha256=WSr8txw3OXCRLK9nD39AW20JtMkJUq_OjJ0xWv410f0,13051
+portacode-1.4.17.dev7.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+portacode-1.4.17.dev7.dist-info/entry_points.txt,sha256=lLUUL-BM6_wwe44Xv0__5NQ1BnAz6jWjSMFvZdWW3zU,48
+portacode-1.4.17.dev7.dist-info/top_level.txt,sha256=TGhTYUxfW8SyVZc_zGgzjzc24gGT7nSw8Qf73liVRKM,41
+portacode-1.4.17.dev7.dist-info/RECORD,,