async-batch-llm 0.7.0__tar.gz → 0.7.2__tar.gz

{async_batch_llm-0.7.0 → async_batch_llm-0.7.2}/CHANGELOG.md

@@ -7,6 +7,44 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.7.2] - 2026-04-22
+
+### Fixed
+
+- `GeminiCachedModel.generate()` no longer emits misleading "Cache expired" log
+  lines with a ~56-year age under concurrent workers. The log is now inside the
+  cache lock and after the double-check, so only the worker that actually renews
+  logs the message; losing-race workers stay silent. Age is rendered as
+  "unknown (cache not yet initialized)" when `_cache_created_at` is `None`
+  instead of being computed as `time.time() - 0`.
+
+### Security
+
+- Bumped transitive dependency `authlib` 1.6.10 → 1.7.0 to clear
+  [GHSA-jj8c-mmj3-mmgv](https://github.com/authlib/authlib/security/advisories/GHSA-jj8c-mmj3-mmgv)
+  (CSRF in Authlib's OAuth cache path, medium/CVSS 5.4). async-batch-llm does not
+  use Authlib directly — it reaches us via `pydantic-ai[fastmcp]` → `fastmcp` —
+  and the vulnerable code path is not exercised, but the bump clears the
+  Dependabot alert.
+
+## [0.7.1] - 2026-04-22
+
+### Fixed
+
+- `GeminiCachedModel.prepare()` no longer crashes with `CreateCachedContentConfig`'s
+  `extra_forbidden` ValidationError when `cache_tags` is non-empty. google-genai's
+  `CreateCachedContentConfig` has no `metadata` field in the 1.x line; tags are now
+  encoded into the cache's `display_name` with a sentinel prefix (`abl-tags:<json>`)
+  and decoded on lookup. Previously any `GeminiCachedModel` with a non-empty
+  `cache_tags=` dict failed every worker's prepare() on current google-genai versions.
+
+### Changed
+
+- `cache_tags` are persisted in `CachedContent.display_name` instead of `metadata`.
+  Tag values should stay short — Gemini's `display_name` has a 128-character limit.
+  Caches created outside async-batch-llm (no `abl-tags:` prefix on display_name) are
+  treated as untagged and won't match a `GeminiCachedModel` with `cache_tags` set.
+
 ## [0.7.0] - 2026-04-16
 
 Internal refactor release. Public API (`async_batch_llm/__init__.py`) is unchanged —

{async_batch_llm-0.7.0 → async_batch_llm-0.7.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: async-batch-llm
-Version: 0.7.0
+Version: 0.7.2
 Summary: Provider-agnostic framework for high-throughput LLM processing with async workers, automatic retries, rate limiting, and intelligent validation recovery.
 Project-URL: Homepage, https://github.com/geoff-davis/async-batch-llm
 Project-URL: Documentation, https://geoff-davis.github.io/async-batch-llm/

{async_batch_llm-0.7.0 → async_batch_llm-0.7.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "async-batch-llm"
-version = "0.7.0"
+version = "0.7.2"
 description = "Provider-agnostic framework for high-throughput LLM processing with async workers, automatic retries, rate limiting, and intelligent validation recovery."
 readme = "README.md"
 requires-python = ">=3.10"

{async_batch_llm-0.7.0 → async_batch_llm-0.7.2}/src/async_batch_llm/models.py

@@ -7,12 +7,19 @@ without knowing about provider-specific details.
 Added in v0.6.0.
 """
 
+import json
 import logging
 import time
 from typing import TYPE_CHECKING, Any
 
 from .base import LLMResponse
 
+# Sentinel prefix for encoding cache_tags into Gemini's CachedContent.display_name.
+# google-genai's CreateCachedContentConfig does not expose a metadata field, so we
+# round-trip tags through display_name, marked with this prefix so we can tell
+# async-batch-llm-tagged caches apart from caches with user-chosen display names.
+_TAG_DISPLAY_NAME_PREFIX = "abl-tags:"
+
 # Conditional imports for optional dependencies
 if TYPE_CHECKING:
     from google import genai
@@ -28,6 +35,35 @@ else:
 logger = logging.getLogger(__name__)
 
 
+def _encode_tags_to_display_name(tags: dict[str, str]) -> str:
+    """Encode cache_tags as a deterministic string for the CachedContent display_name.
+
+    Uses sorted, compact JSON so equal tag sets always produce the same display_name —
+    critical for cache lookup to match. Prefixed with a sentinel so we can tell our
+    tag encoding apart from a user-assigned display name.
+    """
+    encoded = json.dumps(tags, sort_keys=True, separators=(",", ":"))
+    return f"{_TAG_DISPLAY_NAME_PREFIX}{encoded}"
+
+
+def _decode_tags_from_display_name(display_name: str | None) -> dict[str, str] | None:
+    """Decode cache_tags from a CachedContent display_name.
+
+    Returns None when the display_name is absent or was not produced by
+    _encode_tags_to_display_name. Callers should treat None as "this cache has no
+    tag metadata we can match against".
+    """
+    if not display_name or not display_name.startswith(_TAG_DISPLAY_NAME_PREFIX):
+        return None
+    try:
+        decoded = json.loads(display_name[len(_TAG_DISPLAY_NAME_PREFIX) :])
+    except ValueError:
+        return None
+    if not isinstance(decoded, dict):
+        return None
+    return decoded
+
+
 def _extract_metadata(response: Any) -> dict[str, Any] | None:
     """Extract safety ratings and finish reason from a Gemini response."""
     metadata: dict[str, Any] = {}
@@ -40,9 +76,7 @@ def _extract_metadata(response: Any) -> dict[str, Any] | None:
             if hasattr(candidate, "safety_ratings") and candidate.safety_ratings:
                 ratings: dict[str, str] = {}
                 for rating in candidate.safety_ratings:
-                    category = (
-                        str(rating.category) if hasattr(rating, "category") else "UNKNOWN"
-                    )
+                    category = str(rating.category) if hasattr(rating, "category") else "UNKNOWN"
                     probability = (
                         str(rating.probability) if hasattr(rating, "probability") else "UNKNOWN"
                     )
@@ -242,7 +276,10 @@ class GeminiCachedModel:
             cache_renewal_buffer_seconds: Renew this many seconds before expiry
                 (default: 300 = 5 minutes).
             auto_renew: Auto-renew expired caches in generate() (default: True).
-            cache_tags: Tags for precise cache matching.
+            cache_tags: Tags for precise cache matching. Encoded into the cache's
+                ``display_name`` at creation (google-genai ``CreateCachedContentConfig``
+                has no ``metadata`` field) and decoded on lookup. Keep tag values
+                short — Gemini's ``display_name`` has a 128-character limit.
             safety_settings: Default safety settings for all calls.
         """
         if genai is None:
@@ -293,7 +330,6 @@ class GeminiCachedModel:
         self._cache_lock: Any = None
         self._prepared = False
 
-
     @property
     def cache_name(self) -> str | None:
         """The name of the active cache, or None."""
@@ -396,12 +432,6 @@ class GeminiCachedModel:
 
         # Auto-renew if expired
         if self._auto_renew and self._is_cache_expired():
-            logger.info(
-                "Cache expired or about to expire, renewing before API call "
-                f"(age: {time.time() - (self._cache_created_at or 0):.0f}s, "
-                f"renewal buffer: {self._cache_renewal_buffer_seconds}s)"
-            )
-
             import asyncio
 
             if self._cache_lock is None:
@@ -409,6 +439,16 @@ class GeminiCachedModel:
 
             async with self._cache_lock:
                 if self._is_cache_expired():
+                    age_str = (
+                        f"{time.time() - self._cache_created_at:.0f}s"
+                        if self._cache_created_at is not None
+                        else "unknown (cache not yet initialized)"
+                    )
+                    logger.info(
+                        "Cache expired or about to expire, renewing before API call "
+                        f"(age: {age_str}, "
+                        f"renewal buffer: {self._cache_renewal_buffer_seconds}s)"
+                    )
                     self._cache = None
                     self._cache_created_at = None
                     self._prepared = False
@@ -476,14 +516,20 @@ class GeminiCachedModel:
                     continue
 
                 if self._cache_tags:
-                    cache_metadata = getattr(cache, "metadata", {}) or {}
-                    tags_match = all(
-                        cache_metadata.get(k) == v for k, v in self._cache_tags.items()
+                    cache_tags = _decode_tags_from_display_name(
+                        getattr(cache, "display_name", None)
                     )
+                    if cache_tags is None:
+                        logger.debug(
+                            f"Skipping cache {cache.name}: no abl-tags display_name "
+                            f"(want {self._cache_tags})"
+                        )
+                        continue
+                    tags_match = all(cache_tags.get(k) == v for k, v in self._cache_tags.items())
                     if not tags_match:
                         logger.debug(
                             f"Skipping cache {cache.name}: tags don't match "
-                            f"(want {self._cache_tags}, has {cache_metadata})"
+                            f"(want {self._cache_tags}, has {cache_tags})"
                         )
                         continue
 
@@ -496,8 +542,7 @@ class GeminiCachedModel:
                 tag_info = f" with tags {self._cache_tags}" if self._cache_tags else ""
                 age = time.time() - self._cache_created_at
                 logger.info(
-                    f"Reusing existing Gemini cache: {self._cache.name}{tag_info} "
-                    f"(age: {age:.0f}s)"
+                    f"Reusing existing Gemini cache: {self._cache.name}{tag_info} (age: {age:.0f}s)"
                 )
                 return
         except Exception as e:
@@ -513,7 +558,9 @@ class GeminiCachedModel:
             "ttl": f"{self._cache_ttl_seconds}s",
         }
         if self._cache_tags:
-            config_kwargs["metadata"] = self._cache_tags
+            # google-genai's CreateCachedContentConfig has no `metadata` field —
+            # round-trip tags through `display_name` with a sentinel prefix.
+            config_kwargs["display_name"] = _encode_tags_to_display_name(self._cache_tags)
 
         self._cache = await self._client.aio.caches.create(
             model=self._model,

{async_batch_llm-0.7.0 → async_batch_llm-0.7.2}/uv.lock

@@ -239,7 +239,7 @@ wheels = [
 
 [[package]]
 name = "async-batch-llm"
-version = "0.7.0"
+version = "0.7.2"
 source = { editable = "." }
 dependencies = [
     { name = "aiolimiter" },
@@ -319,14 +319,15 @@ wheels = [
 
 [[package]]
 name = "authlib"
-version = "1.6.10"
+version = "1.7.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "cryptography" },
+    { name = "joserfc" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/aa/e2/2cd626412bfc3c78b17ca5e5ea8d489f8cae31d40b061f4da0a89068d8a3/authlib-1.6.10.tar.gz", hash = "sha256:856a4f54d6ef3361ca6bb6d14a27e8b88f8097cca795fb428ffe13720e2ecde6", size = 165333, upload-time = "2026-04-13T13:30:34.718Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/d9/82/4d0603f30c1b4629b1f091bb266b0d7986434891d6940a8c87f8098db24e/authlib-1.7.0.tar.gz", hash = "sha256:b3e326c9aa9cc3ea95fe7d89fd880722d3608da4d00e8a27e061e64b48d801d5", size = 175890, upload-time = "2026-04-18T11:00:28.559Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/7d/f6/9093f1ed17b6e2f4ac50d214543d4ec5268902a70e2158a752a06423b5ef/authlib-1.6.10-py2.py3-none-any.whl", hash = "sha256:aa639b43292554539924a3b4aaa9e81cd67ab64d3e28b22428c61f1200240287", size = 244351, upload-time = "2026-04-13T13:30:33.34Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/48/c954218b2a250e23f178f10167c4173fecb5a75d2c206f0a67ba58006c26/authlib-1.7.0-py2.py3-none-any.whl", hash = "sha256:e36817afb02f6f0b6bf55f150782499ddd6ddf44b402bb055d3263cc65ac9ae0", size = 258779, upload-time = "2026-04-18T11:00:26.64Z" },
 ]
 
 [[package]]
@@ -1661,6 +1662,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/14/2f/967ba146e6d58cf6a652da73885f52fc68001525b4197effc174321d70b4/jmespath-1.1.0-py3-none-any.whl", hash = "sha256:a5663118de4908c91729bea0acadca56526eb2698e83de10cd116ae0f4e97c64", size = 20419, upload-time = "2026-01-22T16:35:24.919Z" },
 ]
 
+[[package]]
+name = "joserfc"
+version = "1.6.4"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cryptography" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/de/c6/de8fdbdfa75c8ca04fead38a82d573df8a82906e984c349d58665f459558/joserfc-1.6.4.tar.gz", hash = "sha256:34ce5f499bfcc5e9ad4cc75077f9278ab3227b71da9aaf28f9ab705f8a560d3c", size = 231866, upload-time = "2026-04-13T13:15:40.632Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b6/f7/210b27752e972edb36d239315b08d3eb6b14824cc4a590da2337d195260b/joserfc-1.6.4-py3-none-any.whl", hash = "sha256:3e4a22b509b41908989237a045e25c8308d5fd47ab96bdae2dd8057c6451003a", size = 70464, upload-time = "2026-04-13T13:15:39.259Z" },
+]
+
 [[package]]
 name = "jsonpath-python"
 version = "1.1.5"