runpod 2.0.0rc1__tar.gz → 2.0.1.dev1__tar.gz

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/.github/workflows/cd.yml

@@ -28,7 +28,7 @@ jobs:
 
   test-publish:
     if: github.ref == 'refs/heads/staging'
-    runs-on: ubuntu-latest
+    runs-on: runpod-4vCPU-ubuntu-2404 # GitHub-hosted (trusted publishing needs that) with allow-listed static IPs
     environment: test-release
     permissions:
       contents: read
@@ -58,7 +58,7 @@ jobs:
 
   release-please:
     if: github.ref == 'refs/heads/main'
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     outputs:
       release_created: ${{ steps.release.outputs.release_created }}
       tag_name: ${{ steps.release.outputs.tag_name }}
@@ -74,7 +74,7 @@ jobs:
     name: PyPI Publish
     needs: release-please
     if: ${{ always() && (needs.release-please.outputs.release_created || (github.event_name == 'workflow_dispatch' && inputs.force_publish == 'true')) }}
-    runs-on: ubuntu-latest
+    runs-on: runpod-4vCPU-ubuntu-2404 # GitHub-hosted (trusted publishing needs that) with allow-listed static IPs
     environment:
       name: pypi-production
       url: https://pypi.org/project/runpod/
@@ -109,7 +109,7 @@ jobs:
   linear-sync:
     name: Linear Release Sync
     needs: release-please
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     permissions:
       contents: read
     steps:
@@ -128,7 +128,7 @@ jobs:
     name: Linear Release Complete
     if: ${{ needs.release-please.outputs.release_created }}
     needs: [release-please, linear-sync]
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     permissions:
       contents: read
     steps:
@@ -148,7 +148,7 @@ jobs:
   #   name: Notify workers
   #   needs: [release-please, pypi-publish]
   #   if: ${{ needs.release-please.outputs.release_created || (github.event_name == 'workflow_dispatch' && inputs.force_publish == 'true') }}
-  #   runs-on: ubuntu-latest
+  #   runs-on: blacksmith-2vcpu-ubuntu-2404
   #   strategy:
   #     matrix:
   #       repo:

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/.github/workflows/ci.yml

@@ -19,7 +19,7 @@ jobs:
 
   test:
     if: github.event_name != 'schedule'
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     strategy:
       matrix:
         python-version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
@@ -57,7 +57,7 @@ jobs:
 
   e2e:
     if: github.event_name != 'schedule' && github.repository == 'runpod/runpod-python'
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     timeout-minutes: 15
     steps:
       - uses: actions/checkout@v7
@@ -82,12 +82,15 @@ jobs:
           source .venv/bin/activate
           pytest tests/e2e/ -v -p no:xdist --timeout=600 --reruns 1 --reruns-delay 5 --log-cli-level=INFO -o "addopts="
         env:
+          # the e2e provisioner still drives runpod-flash, whose import-time
+          # sunset FutureWarning is an error under pytest.ini filterwarnings
+          RUNPOD_FLASH_SILENCE_DEPRECATION: "1"
           RUNPOD_API_KEY: ${{ secrets.RUNPOD_API_KEY }}
           RUNPOD_SDK_GIT_REF: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
 
   codeql:
     if: github.event_name == 'schedule' || github.event_name == 'pull_request'
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     permissions:
       actions: read
       contents: read
@@ -96,16 +99,16 @@ jobs:
       - uses: actions/checkout@v7
 
       - name: Initialize CodeQL
-        uses: github/codeql-action/init@v4.38.1
+        uses: github/codeql-action/init@v4.39.1
         with:
           languages: python
           queries: security-extended,security-and-quality
 
       - name: Autobuild
-        uses: github/codeql-action/autobuild@v4.38.1
+        uses: github/codeql-action/autobuild@v4.39.1
 
       - name: Perform CodeQL Analysis
-        uses: github/codeql-action/analyze@v4.38.1
+        uses: github/codeql-action/analyze@v4.39.1
         with:
           category: "/language:python"
 
@@ -114,7 +117,7 @@ jobs:
   # ──────────────────────────────────────────────
 
   validation:
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     needs: [test, e2e, codeql]
     if: always()
     steps:

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/.github/workflows/vhs.yml

@@ -8,7 +8,7 @@ on:
 
 jobs:
   vhs:
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     steps:
       - uses: actions/checkout@v7
 

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/CHANGELOG.md

@@ -14,6 +14,31 @@
 * Deployment artifacts apply project ignore rules and mandatory secret-file exclusions while preserving vendored dependency assets.
 * Synchronous calls wait across polling timeouts on Python 3.10 and propagate operation timeouts.
 
+## [2.1.0](https://github.com/runpod/runpod-python/compare/v2.0.0...v2.1.0) (2026-10-28)
+
+
+### Features
+
+* add rp catalog for gpu and cpu pricing and stock ([#637](https://github.com/runpod/runpod-python/issues/637)) ([9540b8a](https://github.com/runpod/runpod-python/commit/9540b8afff98b83e33146c4c0c03d6fa07157f62))
+* add rp endpoint command group ([#640](https://github.com/runpod/runpod-python/issues/640)) ([be3f76b](https://github.com/runpod/runpod-python/commit/be3f76b4ade25eb62fcce41ffcd0d50088f8ef31))
+* add rp pod lifecycle commands ([#639](https://github.com/runpod/runpod-python/issues/639)) ([eeffe2a](https://github.com/runpod/runpod-python/commit/eeffe2ae4651eab0fb4250580457d002882cfe7f))
+* add rp volume commands for network and global volumes ([#638](https://github.com/runpod/runpod-python/issues/638)) ([598cb88](https://github.com/runpod/runpod-python/commit/598cb88b2eb00497916a8fa6117064c898d3d884))
+
+
+### Bug Fixes
+
+* **apps:** bind volume mounts from RUNPOD_MOUNTS when the runtime did not ([#626](https://github.com/runpod/runpod-python/issues/626)) ([fc4a6c7](https://github.com/runpod/runpod-python/commit/fc4a6c73c3915625d65df63416db835d0b927b8f))
+* **apps:** keep transitive torch out of GPU artifacts ([#623](https://github.com/runpod/runpod-python/issues/623)) ([25515f0](https://github.com/runpod/runpod-python/commit/25515f02a2e3ea36035c856e564106ccb8760b7f))
+* **apps:** pin requested gpu devices instead of their whole pool ([#624](https://github.com/runpod/runpod-python/issues/624)) ([432c55b](https://github.com/runpod/runpod-python/commit/432c55b8a24b5bccbe37955392f9e1a72bad475e))
+* **cli:** delete dev endpoints when rp flash dev is stopped by a signal ([#625](https://github.com/runpod/runpod-python/issues/625)) ([f866ec3](https://github.com/runpod/runpod-python/commit/f866ec3d2117b697a2534286509674657f37511f))
+* **deps:** pin certifi to >=2026.10.19 for emSign bundle remediation (GHSA-6q48-6c7w-9gw2) ([#659](https://github.com/runpod/runpod-python/issues/659)) ([714a565](https://github.com/runpod/runpod-python/commit/714a565514bae7dd2b6ca355b52e8e76d70d1102))
+
+
+### Performance Improvements
+
+* **serverless:** range-parallel VolumeCache hydrate ([#621](https://github.com/runpod/runpod-python/issues/621)) ([29704a4](https://github.com/runpod/runpod-python/commit/29704a49ab591eaca9a83d7d2eef935d9d787eb2))
+* **serverless:** take a full batch once drained ([#646](https://github.com/runpod/runpod-python/issues/646)) ([397c37d](https://github.com/runpod/runpod-python/commit/397c37d3d2a04aaa2eb3c8a83cdc359e230bcb6e))
+
 ## [1.12.0](https://github.com/runpod/runpod-python/compare/v1.11.0...v1.12.0) (2026-08-10)
 
 

{runpod-2.0.0rc1/runpod.egg-info → runpod-2.0.1.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: runpod
-Version: 2.0.0rc1
+Version: 2.0.1.dev1
 Summary: 🐍 | Python library for Runpod API and serverless worker SDK.
 Home-page: https://runpod.io
 Author: Runpod
@@ -28,25 +28,27 @@ Classifier: Topic :: Internet :: WWW/HTTP :: Dynamic Content
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: aiohttp[speedups]>=3.14.3
+Requires-Dist: aiohttp[speedups]>=3.14.5
 Requires-Dist: aiohttp-retry>=2.9.1
 Requires-Dist: backoff>=2.2.1
-Requires-Dist: boto3>=1.43.97
+Requires-Dist: boto3>=1.43.118
+Requires-Dist: certifi>=2026.10.19
 Requires-Dist: click>=8.5.0
-Requires-Dist: cloudpickle>=3.0.0
+Requires-Dist: cloudpickle>=3.1.2
 Requires-Dist: colorama<0.4.7,>=0.4.6
-Requires-Dist: cryptography>=50.0.1
-Requires-Dist: fastapi[all]>=0.141.1
-Requires-Dist: paramiko>=5.0.0
-Requires-Dist: pathspec>=0.12.1
+Requires-Dist: cryptography>=50.0.2
+Requires-Dist: fastapi[all]>=0.162.0
+Requires-Dist: packaging>=26.3
+Requires-Dist: paramiko>=5.1.0
+Requires-Dist: pathspec>=1.1.1
 Requires-Dist: prettytable>=3.18.0
 Requires-Dist: psutil>=7.2.2
 Requires-Dist: py-cpuinfo>=9.0.0
 Requires-Dist: inquirerpy==0.3.4
 Requires-Dist: requests>=2.34.2
-Requires-Dist: rich>=13.0.0
-Requires-Dist: tomli>=2.4.1
-Requires-Dist: tomlkit>=0.15.1
+Requires-Dist: rich>=15.0.0
+Requires-Dist: tomli>=2.5.0
+Requires-Dist: tomlkit>=0.15.2
 Requires-Dist: tqdm-loggable>=0.4.1
 Requires-Dist: urllib3>=2.8.0
 Requires-Dist: watchdog>=6.0.0
@@ -70,6 +72,10 @@ Dynamic: requires-python
 
 </div>
 
+<!-- runpod-agent-setup:start -->
+**Using an AI agent?** Point it to https://docs.runpod.io/agent-setup.md to install the Runpod skills and MCP server.
+<!-- runpod-agent-setup:end -->
+
 ## Installation
 
 ```bash

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/README.md

@@ -13,6 +13,10 @@
 
 </div>
 
+<!-- runpod-agent-setup:start -->
+**Using an AI agent?** Point it to https://docs.runpod.io/agent-setup.md to install the Runpod skills and MCP server.
+<!-- runpod-agent-setup:end -->
+
 ## Installation
 
 ```bash

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/docs/serverless/volume_cache.md

@@ -59,7 +59,9 @@ vc.sync(background=False)           # persist new files back to the volume, inli
 | `namespace` | `RUNPOD_ENDPOINT_ID` | Isolation key for the on-volume mirror. Must be a single safe path component. |
 | `volume_path` | `/runpod-volume` | Network-volume mount point. |
 | `best_effort` | `True` | Swallow and log errors instead of raising. Set `False` while debugging. |
-| `max_workers` | `min(32, (os.cpu_count() or 4) * 4)` | Thread count for parallel copy of large files (I/O-bound). |
+| `max_workers` | `min(32, (os.cpu_count() or 4) * 4)` | Thread count for parallel copy of large files on sync (I/O-bound). |
+| `hydrate_workers` | `max_workers` if set, else `64` | Concurrent range reads when hydrating large files. |
+| `hydrate_chunk_size` | `8 MiB` | Bytes per hydrate range. Peak hydrate memory is about `hydrate_workers * hydrate_chunk_size` (512 MiB by default). |
 
 ## How it works
 
@@ -77,9 +79,25 @@ vc.sync(background=False)           # persist new files back to the volume, inli
   files are skipped. The `small.tar` archive is re-packed as a whole whenever
   any small file has changed, since unpacking and re-diffing many tiny files
   individually is slower than the network volume's per-file overhead.
-- **Parallel copy.** Large-file transfers run across a thread pool sized by
-  `max_workers` (default `min(32, (os.cpu_count() or 4) * 4)`), since the
-  work is I/O-bound.
+- **Parallel copy.** On sync, large files are copied whole across a thread
+  pool sized by `max_workers` (default `min(32, (os.cpu_count() or 4) * 4)`),
+  since the work is I/O-bound.
+- **Range-parallel hydrate.** On hydrate, every large file is split into
+  `hydrate_chunk_size` ranges, and one pool of `hydrate_workers` threads reads
+  the ranges of all files at once, each on its own file descriptor.
+  Object-store mounts such as Global Store serve one sequential stream at
+  roughly 140 MiB/s, and reading several whole files at once doesn't add up,
+  but independent range reads scale to the host's network (~1.1 GiB/s
+  measured). Transient `ENOMEM`/`EIO`/`EAGAIN` errors, which such mounts
+  return under load, are retried with jittered exponential backoff. A file is assembled in a temp
+  file and renamed into place only when every range has copied, so a failed
+  file leaves any existing local copy untouched and is simply not restored.
+  A mirror file whose size differs from the manifest, or whose size or mtime
+  changes between ranges because a concurrent `sync()` replaced it, is
+  likewise not restored. This is a best-effort check: an object-store mount
+  can serve a replacement that lands in the middle of a single range read
+  without it being visible. Keep `hydrate_workers * hydrate_chunk_size` near the default: past ~1 GiB in
+  flight, Global Store starts failing reads with `ENOMEM`.
 - **Packing/extraction.** Packing the small-file archive uses the `tar`
   binary when available (falling back to the stdlib `tarfile` module
   otherwise). Extraction always goes through `tarfile`, validating every

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/pytest.ini

@@ -13,6 +13,7 @@ filterwarnings =
 python_files = tests.py test_*.py *_test.py
 norecursedirs = venv *.egg-info .git build tests/e2e
 asyncio_mode = auto
+asyncio_default_fixture_loop_scope = function
 markers =
     qb: Queue-based tests (local execution, fast)
     lb: Load-balanced tests (remote provisioning, slow)

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/requirements.txt

@@ -1,23 +1,25 @@
-aiohttp[speedups] >= 3.14.3
+aiohttp[speedups] >= 3.14.5
 aiohttp-retry >= 2.9.1
 
 backoff >= 2.2.1
-boto3 >= 1.43.97
+boto3 >= 1.43.118
+certifi >= 2026.10.19
 click >= 8.5.0
-cloudpickle >= 3.0.0
+cloudpickle >= 3.1.2
 colorama >= 0.4.6, < 0.4.7
-cryptography >= 50.0.1
-fastapi[all] >= 0.141.1
-paramiko >= 5.0.0
-pathspec >= 0.12.1
+cryptography >= 50.0.2
+fastapi[all] >= 0.162.0
+packaging >= 26.3
+paramiko >= 5.1.0
+pathspec >= 1.1.1
 prettytable >= 3.18.0
 psutil >= 7.2.2
 py-cpuinfo >= 9.0.0
 inquirerpy == 0.3.4
 requests >= 2.34.2
-rich >= 13.0.0
-tomli >= 2.4.1
-tomlkit >= 0.15.1
+rich >= 15.0.0
+tomli >= 2.5.0
+tomlkit >= 0.15.2
 tqdm-loggable >= 0.4.1
 urllib3 >= 2.8.0
 watchdog >= 6.0.0

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/runpod/_health/gpu.py

@@ -28,6 +28,13 @@ FALLBACK_TIMEOUT_SECONDS = 10
 MAX_ERROR_MESSAGES = 10
 
 
+class GpuAllocationError(RuntimeError):
+    """gpu_test reached the GPUs and a device failed its allocation test.
+
+    The nvidia-smi fallback only lists GPUs, so it must not override this.
+    """
+
+
 def configure() -> None:
     """Read current fitness settings; setup errors are handled by the runner."""
     global TIMEOUT_SECONDS, MAX_ERROR_MESSAGES
@@ -161,6 +168,10 @@ async def _run_gpu_test_binary() -> dict[str, Any]:
             error_msg = "GPU memory allocation test failed"
             if result["errors"]:
                 error_msg += f": {'; '.join(result['errors'][:MAX_ERROR_MESSAGES])}"
+            # Without "Found N GPUs" the binary never reached the devices
+            # (NVML or loader failure), so nvidia-smi may still decide.
+            if result["found_gpus"]:
+                raise GpuAllocationError(error_msg)
             raise RuntimeError(error_msg)
 
         log.info(
@@ -176,9 +187,7 @@ async def _run_gpu_test_binary() -> dict[str, Any]:
         raise RuntimeError(
             f"GPU test binary timed out after {TIMEOUT_SECONDS}s"
         ) from None
-    except FileNotFoundError:
-        raise
-    except PermissionError:
+    except (FileNotFoundError, PermissionError, GpuAllocationError):
         raise
     except Exception as exc:
         raise RuntimeError(f"GPU test binary execution failed: {exc}") from exc
@@ -238,8 +247,9 @@ async def _check_gpu_health() -> None:
 
     Execution strategy:
     1. Try binary test if available
-    2. Fall back to Python check if binary fails/missing
-    3. Raise RuntimeError if all methods fail
+    2. Fail if the binary reports a GPU allocation failure
+    3. Otherwise fall back to Python check if binary fails/missing
+    4. Raise RuntimeError if all methods fail
 
     Raises:
         RuntimeError: If GPU health check fails
@@ -251,6 +261,8 @@ async def _check_gpu_health() -> None:
     try:
         await _run_gpu_test_binary()
         return  # Success!
+    except GpuAllocationError:
+        raise
     except FileNotFoundError as exc:
         log.debug(f"GPU binary not found: {exc}")
         binary_error = exc

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/runpod/_health/system.py

@@ -17,6 +17,7 @@ import asyncio
 import os
 import re
 import shutil
+import sys
 import time
 from urllib.parse import urlsplit
 
@@ -39,6 +40,47 @@ GPU_BENCHMARK_TIMEOUT = 2
 CUDA_VERSION_PROBE_TIMEOUT = 5
 
 
+GPU_FRAMEWORKS = ("torch", "cupy")
+
+
+def _loaded_module(name: str):
+    """
+    Return a module the handler has already imported, without importing it.
+
+    Importing torch costs ~1.5s on a cold worker. The torch and CuPy checks
+    reuse whatever the handler loaded; a framework it has not loaded is
+    treated as unavailable, and the gpu_test binary still verifies the GPU.
+
+    Raises:
+        ImportError: If the handler has not imported the module.
+    """
+    module = sys.modules.get(name)
+    if module is None:
+        raise ImportError(f"{name} not loaded by the handler")
+    return module
+
+
+def _handler_loaded_gpu_framework() -> bool:
+    """True if the handler has imported torch or CuPy."""
+    return any(sys.modules.get(name) is not None for name in GPU_FRAMEWORKS)
+
+
+def _torch_without_cuda_context(torch) -> bool:
+    """
+    True if the handler imported torch but has not created a CUDA context.
+
+    The torch checks must never be the first to initialize CUDA in the handler
+    process. An engine built in this process afterwards (vLLM's in-process
+    AsyncLLM/LLM) sees "CUDA is initialized" and spawns its workers instead of
+    forking, re-importing the engine and torch in every child (10.6-11.5 s per
+    cold start measured with vLLM v0.30.0). This gate runs before
+    torch.cuda.is_available(), whose default cudaGetDeviceCount path also
+    initializes the driver and poisons later forks. When the handler has
+    already created a context the checks cost nothing extra.
+    """
+    return not torch.cuda.is_initialized()
+
+
 def configure() -> None:
     """Read current fitness settings; setup errors are handled by the runner."""
     global \
@@ -347,11 +389,19 @@ async def _check_cuda_initialization() -> None:
 
     Tests actual device initialization, memory access, and device properties.
     This catches issues where CUDA appears available but fails at runtime.
-    Skips silently on CPU-only workers.
+    Skips silently on CPU-only workers and when the handler has not
+    imported PyTorch or CuPy.
 
     Raises:
         RuntimeError: If CUDA initialization or device access fails
     """
+    if not _handler_loaded_gpu_framework():
+        log.debug(
+            "Handler has not loaded PyTorch or CuPy, skipping CUDA initialization "
+            "check (the gpu_test binary covers device allocation)"
+        )
+        return
+
     # Skip on CPU-only workers
     if not gpu_available():
         log.debug("No GPU detected, skipping CUDA initialization check")
@@ -359,7 +409,15 @@ async def _check_cuda_initialization() -> None:
 
     # Try PyTorch first (most common)
     try:
-        import torch
+        torch = _loaded_module("torch")
+
+        if _torch_without_cuda_context(torch):
+            log.debug(
+                "Handler has not created a CUDA context, skipping CUDA initialization "
+                "check so the engine can still fork (the gpu_test binary, where it "
+                "runs, covers device allocation)"
+            )
+            return
 
         if not torch.cuda.is_available():
             log.debug("CUDA not available in PyTorch, skipping initialization check")
@@ -403,7 +461,7 @@ async def _check_cuda_initialization() -> None:
 
     # Fallback: try CuPy
     try:
-        import cupy as cp
+        cp = _loaded_module("cupy")
 
         # Reset CuPy state
         cp.cuda.Device().synchronize()
@@ -441,11 +499,18 @@ async def _check_gpu_compute_benchmark() -> None:
     Quick GPU compute benchmark using matrix multiplication.
 
     Tests basic tensor operations to ensure GPU is functional and responsive.
-    Skips silently on CPU-only workers.
+    Skips silently on CPU-only workers and when the handler has not
+    imported PyTorch or CuPy.
 
     Raises:
         RuntimeError: If GPU compute fails or is too slow
     """
+    if not _handler_loaded_gpu_framework():
+        log.debug(
+            "Handler has not loaded PyTorch or CuPy, skipping GPU compute benchmark"
+        )
+        return
+
     # Skip on CPU-only workers
     if not gpu_available():
         log.debug("No GPU detected, skipping GPU compute benchmark")
@@ -453,7 +518,14 @@ async def _check_gpu_compute_benchmark() -> None:
 
     # Try PyTorch first
     try:
-        import torch
+        torch = _loaded_module("torch")
+
+        if _torch_without_cuda_context(torch):
+            log.debug(
+                "Handler has not created a CUDA context, skipping GPU compute "
+                "benchmark so the engine can still fork"
+            )
+            return
 
         if not torch.cuda.is_available():
             log.debug("CUDA not available in PyTorch, skipping benchmark")
@@ -492,7 +564,7 @@ async def _check_gpu_compute_benchmark() -> None:
 
     # Fallback: try CuPy
     try:
-        import cupy as cp
+        cp = _loaded_module("cupy")
 
         size = 1024
         start_time = time.perf_counter()
@@ -536,9 +608,11 @@ def auto_register_system_checks() -> None:
     Registers memory, disk, and network checks for all workers.
     Registers CUDA version, initialization, and GPU benchmark checks only if GPU is detected.
 
-    The two checks that import torch and allocate on the device are marked
+    The two checks that use torch and allocate on the device are marked
     @defer_to_worker_start so the import-time pass cannot create a CUDA context
-    before the handler module runs.
+    before the handler module runs. They only use a framework the handler has
+    already imported, and only run when the handler has already created a CUDA
+    context, so they never add a torch import or a CUDA context to cold start.
     """
     configure()
     log.debug("Registering system resource fitness checks")

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/runpod/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '2.0.0rc1'
-__version_tuple__ = version_tuple = (2, 0, 0, 'rc1')
+__version__ = version = '2.0.1.dev1'
+__version_tuple__ = version_tuple = (2, 0, 1, 'dev1')
 
-__commit_id__ = commit_id = 'g3864d374c'
+__commit_id__ = commit_id = 'ga4ba07898'

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/runpod/api/mutations/apps.py

@@ -1,11 +1,5 @@
 """GraphQL mutations for capabilities absent from the resource REST API."""
 
-MUTATION_CREATE_GLOBAL_VOLUME = """
-mutation createGlobalVolume($input: GlobalStoreBucketCreateInput!) {
-    globalStoreBucketCreate(input: $input) { id name }
-}
-"""
-
 # endpoint saves retain flash bindings, cached models, and schedules atomically.
 MUTATION_SAVE_ENDPOINT = """
 mutation saveEndpoint($input: EndpointInput!) {

{runpod-2.0.0rc1 → runpod-2.0.1.dev1}/runpod/api/queries/apps.py

@@ -1,13 +1,5 @@
 """GraphQL queries for capabilities absent from the resource REST API."""
 
-QUERY_GLOBAL_VOLUMES = """
-query myGlobalVolumes {
-    myself {
-        globalStoreBuckets { id name }
-    }
-}
-"""
-
 # rest stock filters require a power-of-two count of at least two.
 QUERY_CPU_STOCK = """
 query CpuStock($cpuFlavorInput: CpuFlavorInput, $specificsInput: SpecificsInput) {