supertable 2.3.8__tar.gz → 2.3.9__tar.gz

{supertable-2.3.8/supertable.egg-info → supertable-2.3.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: supertable
-Version: 2.3.8
+Version: 2.3.9
 Summary: SuperTable — versioned data lake library for SQL analytics on Parquet + Redis.
 Author: Levente Kupas
 Author-email: Levente Kupas <lkupas@kladnasoft.com>

{supertable-2.3.8 → supertable-2.3.9}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "supertable"
-version = "2.3.8"
+version = "2.3.9"
 description = "SuperTable — versioned data lake library for SQL analytics on Parquet + Redis."
 readme = "README.md"
 requires-python = ">=3.10"

{supertable-2.3.8 → supertable-2.3.9}/setup.py

@@ -19,7 +19,7 @@ long_description = readme.read_text(encoding="utf-8") if readme.exists() else ""
 
 setup(
     name="supertable",
-    version="2.3.8",
+    version="2.3.9",
     description="SuperTable — versioned data lake library for SQL analytics on Parquet + Redis.",
     long_description=long_description,
     long_description_content_type="text/markdown",

{supertable-2.3.8 → supertable-2.3.9}/supertable/__init__.py

@@ -25,7 +25,7 @@ See the ``supertable.demo`` package for runnable end-to-end demos and the
 project documentation for the full API surface.
 """
 
-__version__ = "2.3.8"
+__version__ = "2.3.9"
 
 # Re-export the core public surface so users can do ``from supertable import …``
 # instead of remembering submodule paths.

{supertable-2.3.8 → supertable-2.3.9}/supertable/config/settings.py

@@ -157,6 +157,14 @@ class Settings:
     SUPERTABLE_DUCKDB_MATERIALIZE: str = "view"    # SUPERTABLE_DUCKDB_MATERIALIZE
     SUPERTABLE_DUCKDB_PRESIGNED: bool = False      # SUPERTABLE_DUCKDB_PRESIGNED
     SUPERTABLE_DUCKDB_USE_HTTPFS: bool = False     # SUPERTABLE_DUCKDB_USE_HTTPFS
+    # Allow DuckDB to DOWNLOAD a missing extension (httpfs) from
+    # extensions.duckdb.org at query time.  Default OFF: httpfs is baked into
+    # the image and seeded into the extension dir, and an implicit network
+    # install on an offline/firewalled node can hang for minutes instead of
+    # failing.  When OFF, a genuinely missing extension raises immediately with
+    # an actionable message; set ON only on a networked box that must
+    # self-install.
+    SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD: bool = False  # SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD
     # Write-path overwrite/delete resolution via the DuckDB pushdown probe.
     # Disabled by default: the polars fallback reads only the projected key
     # columns through the storage SDK and needs no httpfs extension, so it works
@@ -444,6 +452,7 @@ def _build_settings() -> Settings:
         SUPERTABLE_DUCKDB_MATERIALIZE=_env_str("SUPERTABLE_DUCKDB_MATERIALIZE", "view"),
         SUPERTABLE_DUCKDB_PRESIGNED=_env_bool("SUPERTABLE_DUCKDB_PRESIGNED", False),
         SUPERTABLE_DUCKDB_USE_HTTPFS=_env_bool("SUPERTABLE_DUCKDB_USE_HTTPFS", False),
+        SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD=_env_bool("SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD", False),
         SUPERTABLE_DUCKDB_WRITE_PROBE=_env_bool("SUPERTABLE_DUCKDB_WRITE_PROBE", False),
         SUPERTABLE_DUCKDB_TOMBSTONE_CACHE_MAX_PER_TABLE=_env_int("SUPERTABLE_DUCKDB_TOMBSTONE_CACHE_MAX_PER_TABLE", 8),
         SUPERTABLE_DUCKDB_TOMBSTONE_CACHE_TTL_SEC=_env_int("SUPERTABLE_DUCKDB_TOMBSTONE_CACHE_TTL_SEC", 300),

{supertable-2.3.8 → supertable-2.3.9}/supertable/engine/engine_common.py

@@ -215,12 +215,43 @@ def configure_httpfs_and_s3(
     if not for_paths:
         return
 
-    # Load httpfs; install only when not already available.
+    # Load httpfs.  It is baked into the image and seeded into the DuckDB
+    # extension dir (see the container entrypoint), so LOAD normally succeeds
+    # with no network access.
+    #
+    # Why this is NOT a blind ``INSTALL`` fallback: ``INSTALL httpfs`` performs
+    # an HTTP GET to extensions.duckdb.org.  On an offline / firewalled node
+    # that socket can stall for minutes — or hang indefinitely on a blackholed
+    # route — turning a should-be-instant failure into an unbounded query hang.
+    # So we fail fast instead:
+    #   * SET autoinstall_known_extensions=false makes LOAD raise immediately
+    #     when the extension is absent, rather than silently downloading it;
+    #   * a network INSTALL is attempted ONLY when explicitly opted in via
+    #     SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD;
+    #   * otherwise we raise a clear, actionable error the caller returns.
     try:
-        con.execute("LOAD httpfs;")
+        con.execute("SET autoinstall_known_extensions=false;")
     except Exception:
-        con.execute("INSTALL httpfs;")
+        pass
+
+    try:
         con.execute("LOAD httpfs;")
+    except Exception as load_err:
+        if settings.SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD:
+            # Operator explicitly allowed reaching the network for a one-off
+            # install (e.g. an online dev box without a baked extension).
+            con.execute("INSTALL httpfs;")
+            con.execute("LOAD httpfs;")
+        else:
+            raise RuntimeError(
+                "DuckDB 'httpfs' extension is not available locally and network "
+                "auto-download is disabled, so this query cannot run. Bake/seed "
+                "httpfs into "
+                f"'{get_app_home()}/.duckdb/extensions/v<duckdb_version>/<platform>/' "
+                "(the container entrypoint restores it from /opt/duckdb-extensions), "
+                "or set SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD=true to permit a "
+                f"one-time online install. Underlying DuckDB error: {load_err}"
+            ) from load_err
 
     any_s3 = any(str(p).lower().startswith("s3://") for p in for_paths)
     any_http = any(str(p).lower().startswith(("http://", "https://")) for p in for_paths)
@@ -634,6 +665,17 @@ def init_connection(
     except Exception as e:
         logger.warning(f"[duckdb.init] home_directory pin failed: {e}")
 
+    # Never let DuckDB auto-DOWNLOAD an extension.  Everything we need (httpfs)
+    # is baked/seeded into the local extension dir; an implicit network install
+    # would reach out to extensions.duckdb.org and can hang for minutes on an
+    # offline/firewalled node — turning a should-be-instant error into an
+    # unbounded query hang.  configure_httpfs_and_s3() owns the explicit,
+    # opt-in install path (SUPERTABLE_DUCKDB_ALLOW_EXTENSION_DOWNLOAD).
+    try:
+        con.execute("SET autoinstall_known_extensions=false;")
+    except Exception as e:
+        logger.debug(f"[duckdb.init] disabling extension auto-install failed: {e}")
+
     # Resolve memory limit.
     # Single env var SUPERTABLE_DUCKDB_MEMORY_LIMIT controls both executors.
     # The `memory_limit` argument is the caller's fallback when the env var is absent.

{supertable-2.3.8 → supertable-2.3.9/supertable.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: supertable
-Version: 2.3.8
+Version: 2.3.9
 Summary: SuperTable — versioned data lake library for SQL analytics on Parquet + Redis.
 Author: Levente Kupas
 Author-email: Levente Kupas <lkupas@kladnasoft.com>