runpod 2.0.0rc2__tar.gz → 2.0.1.dev1__tar.gz

{runpod-2.0.0rc2 → 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-2.0.0rc2 → 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
@@ -90,7 +90,7 @@ jobs:
 
   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
@@ -99,16 +99,16 @@ jobs:
       - uses: actions/checkout@v7
 
       - name: Initialize CodeQL
-        uses: github/codeql-action/init@v4.39.0
+        uses: github/codeql-action/init@v4.39.1
         with:
           languages: python
           queries: security-extended,security-and-quality
 
       - name: Autobuild
-        uses: github/codeql-action/autobuild@v4.39.0
+        uses: github/codeql-action/autobuild@v4.39.1
 
       - name: Perform CodeQL Analysis
-        uses: github/codeql-action/analyze@v4.39.0
+        uses: github/codeql-action/analyze@v4.39.1
         with:
           category: "/language:python"
 
@@ -117,7 +117,7 @@ jobs:
   # ──────────────────────────────────────────────
 
   validation:
-    runs-on: ubuntu-latest
+    runs-on: blacksmith-2vcpu-ubuntu-2404
     needs: [test, e2e, codeql]
     if: always()
     steps:

{runpod-2.0.0rc2 → 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-2.0.0rc2 → 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-2.0.0rc2/runpod.egg-info → runpod-2.0.1.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: runpod
-Version: 2.0.0rc2
+Version: 2.0.1.dev1
 Summary: 🐍 | Python library for Runpod API and serverless worker SDK.
 Home-page: https://runpod.io
 Author: Runpod
@@ -28,16 +28,18 @@ 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>=1.43.111
+Requires-Dist: boto3>=1.43.118
+Requires-Dist: certifi>=2026.10.19
 Requires-Dist: click>=8.5.0
 Requires-Dist: cloudpickle>=3.1.2
 Requires-Dist: colorama<0.4.7,>=0.4.6
 Requires-Dist: cryptography>=50.0.2
-Requires-Dist: fastapi[all]>=0.143.0
-Requires-Dist: paramiko>=5.0.0
+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

{runpod-2.0.0rc2 → 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-2.0.0rc2 → 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-2.0.0rc2 → runpod-2.0.1.dev1}/requirements.txt

@@ -1,14 +1,16 @@
-aiohttp[speedups] >= 3.14.3
+aiohttp[speedups] >= 3.14.5
 aiohttp-retry >= 2.9.1
 
 backoff >= 2.2.1
-boto3 >= 1.43.111
+boto3 >= 1.43.118
+certifi >= 2026.10.19
 click >= 8.5.0
 cloudpickle >= 3.1.2
 colorama >= 0.4.6, < 0.4.7
 cryptography >= 50.0.2
-fastapi[all] >= 0.143.0
-paramiko >= 5.0.0
+fastapi[all] >= 0.162.0
+packaging >= 26.3
+paramiko >= 5.1.0
 pathspec >= 1.1.1
 prettytable >= 3.18.0
 psutil >= 7.2.2

{runpod-2.0.0rc2 → runpod-2.0.1.dev1}/runpod/_health/system.py

@@ -65,6 +65,22 @@ def _handler_loaded_gpu_framework() -> bool:
     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 \
@@ -395,6 +411,14 @@ async def _check_cuda_initialization() -> None:
     try:
         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")
             return
@@ -496,6 +520,13 @@ async def _check_gpu_compute_benchmark() -> None:
     try:
         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")
             return
@@ -580,7 +611,8 @@ def auto_register_system_checks() -> None:
     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. They only use a framework the handler has
-    already imported, so they never add a torch import to cold start.
+    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-2.0.0rc2 → 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 = '2.0.0rc2'
-__version_tuple__ = version_tuple = (2, 0, 0, 'rc2')
+__version__ = version = '2.0.1.dev1'
+__version_tuple__ = version_tuple = (2, 0, 1, 'dev1')
 
-__commit_id__ = commit_id = 'gaa721999b'
+__commit_id__ = commit_id = 'ga4ba07898'

{runpod-2.0.0rc2 → runpod-2.0.1.dev1}/runpod/apps/api.py

@@ -164,6 +164,11 @@ class AppsApiClient:
     def __init__(self, api_key: Optional[str] = None):
         self._api_key = api_key
 
+    async def gpu_pool_members(self) -> Optional[Dict[str, Set[str]]]:
+        from .gpu import live_pool_members_async
+
+        return await live_pool_members_async(self._api_key)
+
     async def _execute(
         self,
         query: str,

runpod-2.0.1.dev1/runpod/apps/auth.py

@@ -0,0 +1,185 @@
+"""browser-approval and email-verification login flows."""
+
+import asyncio
+import base64
+import datetime as dt
+import hashlib
+import os
+import secrets
+from typing import Any, Callable, Dict, Optional
+from urllib.parse import quote
+from uuid import uuid4
+
+import aiohttp
+
+from runpod.api.rest import _build_url
+from runpod.user_agent import USER_AGENT
+
+from .api import AppsApiClient
+from .errors import AppError
+
+CONSOLE_URL = os.environ.get("RUNPOD_CONSOLE_URL", "https://console.runpod.io")
+
+POLL_INTERVAL_SECONDS = 2.0
+DEFAULT_TIMEOUT_SECONDS = 600.0
+
+_APPROVED_STATUSES = frozenset({"APPROVED", "CONSUMED"})
+_FAILED_STATUSES = frozenset({"DENIED", "EXPIRED"})
+
+
+class LoginError(AppError):
+    pass
+
+
+class _VerificationCodeRejected(LoginError):
+    pass
+
+
+async def _post_authentication_attempt(
+    session: aiohttp.ClientSession,
+    path: str,
+    body: dict,
+    *,
+    creating: bool = False,
+) -> dict:
+    try:
+        async with session.post(
+            _build_url(path),
+            json=body,
+            headers={"Idempotency-Key": str(uuid4())} if creating else {},
+            allow_redirects=False,
+        ) as response:
+            if response.status != (201 if creating else 200):
+                if response.status == 403:
+                    raise LoginError(
+                        "email sign-in is not allowed for this account; sign in "
+                        "through the console and use runpod login --api-key"
+                    )
+                if response.status == 404:
+                    if creating:
+                        raise LoginError("email sign-in is unavailable on this API")
+                    raise LoginError(
+                        "login attempt expired, was already used, or is locked; "
+                        "start login again"
+                    )
+                if response.status == 422:
+                    if creating:
+                        raise LoginError(
+                            "email sign-in request rejected; check --email"
+                        )
+                    raise _VerificationCodeRejected("verification code rejected")
+                if response.status == 429:
+                    raise LoginError(
+                        "too many login attempts; wait before trying again"
+                    )
+                raise LoginError(f"email sign-in failed (HTTP {response.status})")
+            payload = await response.json()
+    except (aiohttp.ClientError, asyncio.TimeoutError):
+        raise LoginError(
+            "could not complete email sign-in; check your connection"
+        ) from None
+    except ValueError:
+        raise LoginError("invalid response from the email sign-in API") from None
+
+    if not isinstance(payload, dict):
+        raise LoginError("invalid response from the email sign-in API")
+    return payload
+
+
+async def agent_login(email: str, *, read_code: Callable[[], str]) -> str:
+    """exchange an emailed code for an api key bound to this login's verifier."""
+    verifier = secrets.token_urlsafe(32)
+    challenge = (
+        base64.urlsafe_b64encode(hashlib.sha256(verifier.encode("ascii")).digest())
+        .decode("ascii")
+        .rstrip("=")
+    )
+    path = "/v2/signup/authentication-attempts"
+
+    async with aiohttp.ClientSession(
+        timeout=aiohttp.ClientTimeout(total=30),
+        headers={"Accept": "application/json", "User-Agent": USER_AGENT},
+        cookie_jar=aiohttp.DummyCookieJar(),
+    ) as session:
+        attempt = await _post_authentication_attempt(
+            session,
+            path,
+            {"email": email, "keyName": "runpod-cli-agent", "codeChallenge": challenge},
+            creating=True,
+        )
+        attempt_id = attempt.get("attemptId")
+        if not isinstance(attempt_id, str) or not attempt_id:
+            raise LoginError("auth attempt failed to initialize")
+
+        while True:
+            code = read_code()
+            try:
+                result = await _post_authentication_attempt(
+                    session,
+                    f"{path}/{quote(attempt_id, safe='')}/verify",
+                    {"code": code, "codeVerifier": verifier},
+                )
+            except _VerificationCodeRejected:
+                continue
+            break
+        credential = result.get("credential")
+        api_key = credential.get("apiKey") if isinstance(credential, dict) else None
+        if not isinstance(api_key, str) or not api_key.strip():
+            raise LoginError("email sign-in response did not contain an API key")
+        return api_key
+
+
+def _parse_expires_at(value: Optional[str]) -> Optional[dt.datetime]:
+    if not value:
+        return None
+    try:
+        return dt.datetime.fromisoformat(value.replace("Z", "+00:00"))
+    except ValueError:
+        return None
+
+
+def auth_url(request_id: str) -> str:
+    return f"{CONSOLE_URL}/flash/login?request={request_id}"
+
+
+async def browser_login(
+    *,
+    timeout_seconds: float = DEFAULT_TIMEOUT_SECONDS,
+    on_url: Optional[Callable[[str], None]] = None,
+    api: Optional[AppsApiClient] = None,
+) -> str:
+    """run the browser-approval flow and return the granted api key.
+
+    on_url is called with the approval url as soon as the request is
+    created (the caller renders/opens it).
+    """
+    client = api or AppsApiClient()
+    request: Dict[str, Any] = await client.create_auth_request()
+    request_id = request.get("id")
+
+    if not request_id:
+        raise LoginError("auth request failed to initialize")
+
+    if on_url is not None:
+        on_url(auth_url(request_id))
+
+    deadline = dt.datetime.now(dt.timezone.utc) + dt.timedelta(seconds=timeout_seconds)
+    expires_at = _parse_expires_at(request.get("expiresAt"))
+    if expires_at and expires_at < deadline:
+        deadline = expires_at
+
+    while True:
+        status_payload = await client.get_auth_request_status(request_id)
+        status = status_payload.get("status")
+        api_key = status_payload.get("apiKey")
+
+        if api_key and status in _APPROVED_STATUSES:
+            return api_key
+        if status in _FAILED_STATUSES:
+            raise LoginError(f"login {status.lower()}")
+        if status == "CONSUMED":
+            raise LoginError("login request was already used")
+        if dt.datetime.now(dt.timezone.utc) >= deadline:
+            raise LoginError("login timed out")
+
+        await asyncio.sleep(POLL_INTERVAL_SECONDS)