wafer-core 0.1.27__py3-none-any.whl → 0.1.29__py3-none-any.whl

wafer_core/targets/types.py

@@ -0,0 +1,141 @@
+"""Target and TargetSpec: the two core concepts for GPU resource management.
+
+TargetSpec = provisioning blueprint (TOML config, "how to get a GPU")
+Target = live running resource (from provider API, "what's actually running")
+
+TargetSpec is the existing union of provider-specific frozen dataclasses
+(RunPodTarget, DigitalOceanTarget, BaremetalTarget, etc.), re-exported here
+under the name TargetSpec for clarity.
+
+Target is always fetched from provider APIs. The spec_name field links a
+live resource back to the spec that created it (None = orphan/unbound).
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, Protocol, runtime_checkable
+
+if TYPE_CHECKING:
+    pass
+
+# TargetSpec is the existing union type, re-exported under a clearer name.
+# Each variant is a frozen dataclass with provider-specific provisioning params.
+from wafer_core.utils.kernel_utils.targets.config import (  # noqa: E402
+    TargetConfig,
+)
+
+# TargetSpec = TargetConfig (same union, better name for the new API)
+TargetSpec = TargetConfig
+
+
+@dataclass(frozen=True)
+class Target:
+    """A live running GPU resource, fetched from a provider API.
+
+    This is the runtime counterpart to TargetSpec. A TargetSpec describes
+    *how* to provision a GPU; a Target describes *what's actually running*.
+
+    The provider API is the source of truth for Target state. Local caches
+    (target_state.json) are performance hints only.
+
+    Fields:
+        resource_id: Provider's unique ID (pod_id, droplet_id, or
+            "baremetal:{host}:{port}" for SSH targets with no cloud lifecycle).
+        provider: Which cloud provider owns this resource.
+        status: Current state from provider API.
+        public_ip: SSH-reachable IP address (None if not yet assigned).
+        ssh_port: SSH port (None if not yet assigned).
+        ssh_username: SSH user (typically "root" for cloud providers).
+        gpu_type: GPU model name (e.g., "MI300X", "B200").
+        name: Provider-side resource name (e.g., "wafer-runpod-mi300x-1706000000",
+            "kernelbench-pool-0"). Used for spec_name inference.
+        created_at: ISO timestamp of resource creation (None if unknown).
+        spec_name: Name of the TargetSpec that owns this resource.
+            None means unbound (orphan) — running but no spec claims it.
+        price_per_hour: Cost in $/hr (None if unknown or baremetal).
+        labels: Software metadata not available from the provider API's
+            structured fields. Examples: {"rocm_version": "7.0.2",
+            "cuda_version": "12.4", "image": "rocm/pytorch:rocm7.0.2_..."}.
+            Populated from the container image string at provision time,
+            or from SSH probe on demand. Pool queries filter on these.
+    """
+
+    resource_id: str
+    provider: str
+    status: str
+    public_ip: str | None
+    ssh_port: int | None
+    ssh_username: str
+    gpu_type: str
+    name: str | None = None
+    created_at: str | None = None
+    spec_name: str | None = None
+    price_per_hour: float | None = None
+    labels: dict[str, str] = field(default_factory=dict)
+
+    def __post_init__(self) -> None:
+        assert self.resource_id, "resource_id cannot be empty"
+        assert self.provider, "provider cannot be empty"
+        assert self.status, "status cannot be empty"
+
+
+@dataclass(frozen=True)
+class ReconcileResult:
+    """Result of comparing TargetSpecs to live Targets.
+
+    Pure data — no side effects. The caller decides what to do:
+    - Display bound/unbound/unprovisioned in CLI
+    - Terminate unbound targets
+    - Provision from unprovisioned specs
+
+    Fields:
+        bound: Specs matched to live targets (spec, target) pairs.
+        unbound: Live targets with no matching spec (orphans).
+        unprovisioned: Specs with no live target running.
+    """
+
+    bound: list[tuple[TargetSpec, Target]]
+    unbound: list[Target]
+    unprovisioned: list[TargetSpec]
+
+
+@runtime_checkable
+class TargetProvider(Protocol):
+    """Interface for querying and managing live GPU resources from a cloud provider.
+
+    Each cloud provider (RunPod, DigitalOcean, etc.) implements this protocol.
+    Methods are async because they hit external APIs.
+
+    Baremetal is a degenerate case: list_targets returns a Target built from
+    the spec's ssh_target, provision/terminate are no-ops.
+    """
+
+    async def list_targets(self) -> list[Target]:
+        """List all running resources on the provider account.
+
+        Always hits the provider API — never reads from local cache.
+        """
+        ...
+
+    async def get_target(self, resource_id: str) -> Target | None:
+        """Get a specific resource by provider ID.
+
+        Returns None if the resource doesn't exist or is terminated.
+        """
+        ...
+
+    async def provision(self, spec: TargetSpec) -> Target:
+        """Provision a new resource from a spec.
+
+        Blocks until the resource is SSH-ready.
+        Raises on failure (no silent None returns).
+        """
+        ...
+
+    async def terminate(self, resource_id: str) -> bool:
+        """Terminate a resource by provider ID.
+
+        Returns True if terminated, False if resource not found.
+        """
+        ...

wafer_core/utils/kernel_utils/targets/config.py

@@ -346,25 +346,17 @@ class RunPodTarget:
     ncu_available: bool = False
 
     def __post_init__(self) -> None:
-        """Validate configuration."""
-        from wafer_core.auth import get_api_key
+        """Validate configuration fields.
 
+        API key availability is checked at provision/query time, not here —
+        loading a spec from TOML should not require credentials.
+        """
         assert self.name, "name cannot be empty"
         assert self.ssh_key, "ssh_key cannot be empty"
         assert self.gpu_count > 0, "gpu_count must be positive"
         assert self.provision_timeout > 0, "provision_timeout must be positive"
         assert self.eval_timeout > 0, "eval_timeout must be positive"
 
-        # Check for API key (env var or ~/.wafer/auth.json)
-        api_key = get_api_key("runpod")
-        if not api_key:
-            raise ValueError(
-                "RunPod API key not found.\n"
-                "Set WAFER_RUNPOD_API_KEY environment variable, or run:\n"
-                "  wafer auth login runpod\n"
-                "Get your API key from: https://runpod.io/console/user/settings"
-            )
-
 
 @dataclass(frozen=True)
 class LocalTarget:
@@ -468,24 +460,16 @@ class DigitalOceanTarget:
     ncu_available: bool = False
 
     def __post_init__(self) -> None:
-        """Validate configuration."""
-        from wafer_core.auth import get_api_key
+        """Validate configuration fields.
 
+        API key availability is checked at provision/query time, not here —
+        loading a spec from TOML should not require credentials.
+        """
         assert self.name, "name cannot be empty"
         assert self.ssh_key, "ssh_key cannot be empty"
         assert self.provision_timeout > 0, "provision_timeout must be positive"
         assert self.eval_timeout > 0, "eval_timeout must be positive"
 
-        # Check for API key (env var or ~/.wafer/auth.json)
-        api_key = get_api_key("digitalocean")
-        if not api_key:
-            raise ValueError(
-                "DigitalOcean API key not found.\n"
-                "Set WAFER_AMD_DIGITALOCEAN_API_KEY environment variable, or run:\n"
-                "  wafer auth login digitalocean\n"
-                "Get your API key from: https://cloud.digitalocean.com/account/api/tokens"
-            )
-
 
 # Union type for target configs
 TargetConfig = (

{wafer_core-0.1.27.dist-info → wafer_core-0.1.29.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: wafer-core
-Version: 0.1.27
+Version: 0.1.29
 Summary: Core utilities and environments for Wafer GPU kernel optimization
 Requires-Python: >=3.10
 Requires-Dist: aiohttp>=3.9.0

{wafer_core-0.1.27.dist-info → wafer_core-0.1.29.dist-info}/RECORD

@@ -320,13 +320,13 @@ wafer_core/lib/rocprofiler/systems/sample/__init__.py,sha256=31rNmLPQ7OVhvlOEEOw
 wafer_core/lib/rocprofiler/systems/sample/profiler.py,sha256=CYZPTzNXd48LoCfmY6h_5RSYEdWYccuv3-t4YncHJLE,7384
 wafer_core/lib/trace_compare/PERFORMANCE.md,sha256=jkJh7ApZi8H7NKTcz8v0LNtwSFtIUqY88e3QbL749ww,3823
 wafer_core/lib/trace_compare/__init__.py,sha256=CyUPbPQDYhVLCFFA7S_jNSilG3OgqYjmHSKfR5X11go,1377
-wafer_core/lib/trace_compare/aligner.py,sha256=6HplOHCUIb0cMXA-Lu-91T-hKVTMK4bk8Ei-v7HE1G4,13471
-wafer_core/lib/trace_compare/analyzer.py,sha256=m-waAiU5S72M9J4kUwIy9fPWUecg_oOUczri8Na6xUY,29360
+wafer_core/lib/trace_compare/aligner.py,sha256=1S8Ob3RaEsIjN0HdqEx0yGsW5uf_lMrJVSH_MnZhKok,13788
+wafer_core/lib/trace_compare/analyzer.py,sha256=YkuOPA3HFX_7mNUEhE9CMOtEMGLQd12lvUkvqqeQF14,29698
 wafer_core/lib/trace_compare/api.py,sha256=JSRTcd7eZK1Z8l18TFEiA5A8ENJS1TMz7oIiw1KBbAs,8796
 wafer_core/lib/trace_compare/architecture.py,sha256=8bqlAJQeJLBHblyXvFV-w55PIKiVQDPjDQZ8Jx4tuGg,2110
-wafer_core/lib/trace_compare/classifier.py,sha256=CDGzY9TY-I5wRuEGsu4mTCdljqVTOnLWyFLyNgmkGXI,16864
+wafer_core/lib/trace_compare/classifier.py,sha256=cYAmDW8S75N6cE3mJNZM-UKCJSX7rFP-8klVrukBvNQ,17504
 wafer_core/lib/trace_compare/formatter.py,sha256=GNrCZ45ueBN05CEXjOtTuKvTI8z-g-ZZFil-ni3sWVY,37962
-wafer_core/lib/trace_compare/fusion_analyzer.py,sha256=bD_CJ3JoVg_N6vxJJULd6G8l_-O5qnLuXKDEDItcQtg,15489
+wafer_core/lib/trace_compare/fusion_analyzer.py,sha256=ga0sfxx8OCQu9Hq7uJSAMfXhnCvBaAmzVofBN7_gdV8,19843
 wafer_core/lib/trace_compare/kernel_registry.yaml,sha256=0-knXwsF3pR1x1JdIz-aWaH-5xDgTylh53E47Kf6nHo,9808
 wafer_core/lib/trace_compare/layer_segmentation.py,sha256=kI_Y1e9nrKZfdwfcrGo4h7gpMxqXI_xkgXk46zuFen4,4642
 wafer_core/lib/trace_compare/loader.py,sha256=zBHI0r7CX_wJ2mz0_-s0lm9KGSdaVaq7OKyxUL6KIlw,23997
@@ -597,9 +597,19 @@ wafer_core/sessions/__init__.py,sha256=Ybps5QclZShAELoW9bva4w6OCNrcBf8vd9nGDjYfQ
 wafer_core/sessions/agent.py,sha256=4-Q-NG_xm07FFq7hB8mjxW38nt2_S0QpwCYkPOoGRxA,5946
 wafer_core/sessions/dtypes.py,sha256=K6nOjvL6sjCGY7GTtdEygf1IZY_18R9YkHGqFyMd8wY,589
 wafer_core/sessions/hooks.py,sha256=A-txm6ufnRGQCdtP3vwh7oEOdlLN9Tv0XsjORMihuAI,4295
-wafer_core/targets/__init__.py,sha256=sHndC7AAOaHXlrmDXFLB53a5Y8DBjuyqS6nwsO2nj-Y,1728
+wafer_core/targets/__init__.py,sha256=N4lTf9MjZ5dzAShObweZzyBfPMSzwjD5qBFWnM5lczM,2800
 wafer_core/targets/digitalocean.py,sha256=cvoYpYjtSyy5t2lQAPi7ERruuuibronah_ivOiduAHQ,16550
+wafer_core/targets/pool.py,sha256=TeNE9rpr67OsGtbxniYpr9Cb3wosnf_e3kTLBbwtDok,5434
+wafer_core/targets/probe.py,sha256=rzF8tiq5GxkMR3jhryTOW0GMcoHtrN67wmHlGJuBTv8,3038
+wafer_core/targets/reconcile.py,sha256=Hftd7LyqkcTOP0Qpa_cdYpxGW2I3bkSlkQrnYjU5lns,3091
 wafer_core/targets/runpod.py,sha256=LrVmNvA6qjzL5nbGSWvtw7CHrK6bDu7_o3vKIek00Tc,20286
+wafer_core/targets/spec_store.py,sha256=uNpMdo7ASeq7_RhgAqj8CFIK39rGEbaYtYtqt--FXO0,6455
+wafer_core/targets/state_cache.py,sha256=oji4APL_tjOty_u0CJzHaP59jJAIJWQTjYvD4pCdQ3g,4479
+wafer_core/targets/types.py,sha256=MQ7ECcBAwSoWsJfGxycJoLBeoTXSYtGeXEg5ZNxfs4c,5217
+wafer_core/targets/providers/__init__.py,sha256=u6OCCgyPRymrnZmIYPLF0hdkr6aTCF301K9gSgcFWvc,1355
+wafer_core/targets/providers/baremetal.py,sha256=L0KAiTkRH_fQvCbtaEa5wlJBqsvNaY56Zq6ovBhk2YY,2452
+wafer_core/targets/providers/digitalocean.py,sha256=_TnGi9Otzsn2T_vSv40T_3HFLT559WS_ljGsrWr7j0s,5281
+wafer_core/targets/providers/runpod.py,sha256=jCA7ENFRwbTKyToGa7fw2VS3coY61ggK1m0F17-rvng,7388
 wafer_core/tools/__init__.py,sha256=wBQD45GdSfkxcT6NHzIv0IMeXCc0enwwkpm3T_9j1X8,3341
 wafer_core/tools/bash_tool.py,sha256=daoKOVGSgL0x9X_3l8Apd6-wFH4VMXMGJwVemw2FIfc,16828
 wafer_core/tools/glob_tool.py,sha256=9X5PdOjQJj7kiVNqqCZC0-1LmnE6wHx3Zc9zfMjtXdc,3533
@@ -679,7 +689,7 @@ wafer_core/utils/kernel_utils/static_checker.py,sha256=XIQkzAOkGH5xtrOuZM4tNUqVJ
 wafer_core/utils/kernel_utils/task.py,sha256=XcmKxKUWh5It6nX3zGqj77tWgA32uPfQMqNOqyD5T48,2682
 wafer_core/utils/kernel_utils/utils.py,sha256=uDZoJDxh07hJeLNlPdKN2vgB15pqIr1LbXf0YIBHU4E,43056
 wafer_core/utils/kernel_utils/targets/__init__.py,sha256=4NwRLsuJ__S4xKAfda4Ag82C5MQ3Qio-4xA5S-mQGlU,2067
-wafer_core/utils/kernel_utils/targets/config.py,sha256=V587DYkisEFoWwkmLQUW6I0mXkMEwA52sM7ZINslkK8,20625
+wafer_core/utils/kernel_utils/targets/config.py,sha256=DJPPyV7yGmyvS7cavdDENC5PQsia1dQeQYlWCTE7iUo,19975
 wafer_core/utils/kernel_utils/targets/execution.py,sha256=bZuNXCo0sIdD6hFhetLPrtDC-zMSiIsAx_aml49VVL0,15033
 wafer_core/utils/kernel_utils/targets/selection.py,sha256=5I_RG_7cfhq7uaeR28meC2EeNNKssFsK-Tc3QFG6Ze0,3590
 wafer_core/utils/modal_execution/__init__.py,sha256=jkVqYOLzCT5K73N9Od0UIUsx-99A0m6bpDrxfyXxQZ8,945
@@ -687,6 +697,6 @@ wafer_core/utils/modal_execution/modal_app.py,sha256=VfS2cX8gHtnlPXemmMcEwDPeQdh
 wafer_core/utils/modal_execution/modal_config.py,sha256=7cGX9TGqilQ3qxI3OFGXV5orjtyRU-PEDOJ4vP2oxno,4421
 wafer_core/utils/modal_execution/modal_execution.py,sha256=gChjnV6jqA3A7IRP3DfvV5cSfm_MN0X4f7JZufXgdZE,24594
 wafer_core/utils/modal_execution/test_modal.py,sha256=_jqou_hrLs1Daf1590Pnb0a_lXMMa2rczAPpW9HpoNQ,8153
-wafer_core-0.1.27.dist-info/METADATA,sha256=NYiI9hCaVd9RCCAfd8Ys0UwTMju6wiyaT7nsk4gsB8A,1477
-wafer_core-0.1.27.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-wafer_core-0.1.27.dist-info/RECORD,,
+wafer_core-0.1.29.dist-info/METADATA,sha256=Qjyx92KhI1joutpM8lF0G1zgPou-d8CdzWI80QQqKYg,1477
+wafer_core-0.1.29.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+wafer_core-0.1.29.dist-info/RECORD,,