sibi-dst 2025.8.6__tar.gz → 2025.8.7__tar.gz

{sibi_dst-2025.8.6 → sibi_dst-2025.8.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sibi-dst
-Version: 2025.8.6
+Version: 2025.8.7
 Summary: Data Science Toolkit
 Author: Luis Valverde
 Author-email: lvalverdeb@gmail.com

{sibi_dst-2025.8.6 → sibi_dst-2025.8.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sibi-dst"
-version = "2025.8.6"
+version = "2025.8.7"
 description = "Data Science Toolkit"
 authors = ["Luis Valverde <lvalverdeb@gmail.com>"]
 readme = "README.md"

{sibi_dst-2025.8.6 → sibi_dst-2025.8.7}/sibi_dst/df_helper/_df_helper.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import asyncio
 from typing import Any, Dict, Optional, TypeVar, Union
 
 import dask.dataframe as dd
@@ -104,7 +105,6 @@ class HttpBackend(BaseBackend):
         return self.total_records, result
 
 
-# ---- Main DfHelper ----
 class DfHelper(ManagedResource):
     _BACKEND_STRATEGIES = {
         "sqlalchemy": SqlAlchemyBackend,
@@ -198,6 +198,37 @@ class DfHelper(ManagedResource):
         df = df.persist() if persist else df
         return df.compute() if as_pandas else df
 
+    async def load_async(
+            self,
+            *,
+            persist: bool = False,
+            as_pandas: bool = False,
+            prefer_native: bool = False,
+            **options,
+    ):
+        """
+        Async load that prefers native async backends when available,
+        otherwise runs the sync `load()` in a worker thread via asyncio.to_thread.
+
+        Args:
+            persist: same as `load`
+            as_pandas: same as `load`
+            prefer_native: if True and the backend overrides `aload`, use it.
+                           otherwise force thread offload of `load()`.
+            **options: forwarded to `load` / `aload`
+        """
+        # If the backend provided an override for `aload`, use it
+        if prefer_native and type(self.backend_strategy).aload is not BaseBackend.aload:
+            return await self.aload(persist=persist, as_pandas=as_pandas, **options)
+
+        # Fall back to offloading the sync path to a thread
+        return await asyncio.to_thread(
+            self.load,
+            persist=persist,
+            as_pandas=as_pandas,
+            **options,
+        )
+
     # ---------- dataframe post-processing ----------
     def _post_process_df(self, df: dd.DataFrame) -> dd.DataFrame:
         self.logger.debug("Post-processing DataFrame.")
@@ -240,9 +271,12 @@ class DfHelper(ManagedResource):
         return df
 
     # ---------- sinks ----------
-    def save_to_parquet(self, df: dd.DataFrame, parquet_filename: str, **kwargs):
+    def save_to_parquet(self, df: dd.DataFrame, **kwargs):
         fs: AbstractFileSystem = kwargs.pop("fs", self.fs)
-        path: str = kwargs.pop("parquet_storage_path")
+        path: str = kwargs.pop("parquet_storage_path", self.backend_parquet.parquet_storage_path if self.backend_parquet else None)
+        parquet_filename = kwargs.pop("parquet_filename" or self._backend_params.parquet_filename if self.backend_parquet else None)
+        if not parquet_filename:
+            raise ValueError("A 'parquet_filename' keyword argument must be provided.")
         if not fs:
             raise ValueError("A filesystem (fs) must be provided to save the parquet file.")
         if not path:
@@ -272,7 +306,7 @@ class DfHelper(ManagedResource):
             writer.save_to_clickhouse(df)
             self.logger.debug("Save to ClickHouse completed.")
 
-    # ---------- convenience period loaders ----------
+    # ---------- period loaders ----------
     def load_period(self, dt_field: str, start: str, end: str, **kwargs):
         final_kwargs = self._prepare_period_filters(dt_field, start, end, **kwargs)
         return self.load(**final_kwargs)

sibi_dst-2025.8.7/sibi_dst/utils/async_utils.py

@@ -0,0 +1,12 @@
+import asyncio
+import dask.dataframe as dd
+
+
+def is_dask_dataframe(df):
+    """Check if the given object is a Dask DataFrame."""
+    return isinstance(df, dd.DataFrame)
+
+async def to_thread(func, *args, **kwargs):
+    """Explicit helper to keep code clear where we hop off the event loop."""
+    return await asyncio.to_thread(func, *args, **kwargs)
+

sibi_dst-2025.8.7/sibi_dst/utils/storage_hive.py

@@ -0,0 +1,195 @@
+from __future__ import annotations
+import pandas as pd
+import dask.dataframe as dd
+from typing import Iterable, Optional, List, Tuple, Union
+import fsspec
+
+DNFFilter = List[List[Tuple[str, str, Union[str, int]]]]
+
+
+class HiveDatePartitionedStore:
+    """
+    Dask-only Parquet store with Hive-style yyyy=…/mm=…/dd=… partitions.
+
+    - `write(...)` safely "overwrites" S3 prefixes via per-object deletes (no bulk DeleteObjects).
+    - `read_range(...)` builds DNF filters and auto-matches partition types (string vs int).
+    """
+
+    def __init__(
+        self,
+        path: str,
+        *,
+        filesystem=None,                # fsspec filesystem or None to infer from path
+        date_col: str = "tracking_dt",
+        compression: str = "zstd",
+        partition_values_as_strings: bool = True,  # keep mm=07, dd=01 folder names
+        logger=None,
+    ) -> None:
+        self.path = path
+        self.fs = filesystem or fsspec.open(path).fs
+        self.date_col = date_col
+        self.compression = compression
+        self.partition_values_as_strings = partition_values_as_strings
+        self.log = logger
+
+    # ----------------- public API -----------------
+
+    def write(
+        self,
+        df: dd.DataFrame,
+        *,
+        repartition: Optional[int] = None,
+        overwrite: bool = False,
+    ) -> None:
+        """Write Dask DataFrame to Hive-style yyyy/mm/dd partitions."""
+        self._require_col(df, self.date_col)
+        ser = dd.to_datetime(df[self.date_col], errors="coerce")
+
+        if self.partition_values_as_strings:
+            parts = {
+                "yyyy": ser.dt.strftime("%Y"),
+                "mm":   ser.dt.strftime("%m"),
+                "dd":   ser.dt.strftime("%d"),
+            }
+        else:
+            parts = {
+                "yyyy": ser.dt.year.astype("int32"),
+                "mm":   ser.dt.month.astype("int8"),
+                "dd":   ser.dt.day.astype("int8"),
+            }
+
+        df = df.assign(**{self.date_col: ser}, **parts)
+
+        if repartition:
+            df = df.repartition(npartitions=repartition)
+
+        if overwrite:
+            self._safe_rm_prefix(self.path)
+
+        if self.log:
+            self.log.info(f"Writing parquet to {self.path} (hive yyyy/mm/dd)…")
+
+        df.to_parquet(
+            self.path,
+            engine="pyarrow",
+            write_index=False,
+            filesystem=self.fs,
+            partition_on=["yyyy", "mm", "dd"],
+            compression=self.compression,
+            overwrite=False,  # we pre-cleaned if overwrite=True
+        )
+
+    def read_range(
+        self,
+        start: Union[str, pd.Timestamp],
+        end: Union[str, pd.Timestamp],
+        *,
+        columns: Optional[Iterable[str]] = None,
+    ) -> dd.DataFrame:
+        """
+        Read a date window with partition pruning. Tries string filters first,
+        falls back to integer filters if Arrow infers partition types as ints.
+        """
+        str_filters = self._dnf_filters_for_range_str(start, end)
+        try:
+            return dd.read_parquet(
+                self.path,
+                engine="pyarrow",
+                filesystem=self.fs,
+                columns=list(columns) if columns else None,
+                filters=str_filters,
+            )
+        except Exception:
+            int_filters = self._dnf_filters_for_range_int(start, end)
+            return dd.read_parquet(
+                self.path,
+                engine="pyarrow",
+                filesystem=self.fs,
+                columns=list(columns) if columns else None,
+                filters=int_filters,
+            )
+
+    # Convenience: full month / single day
+    def read_month(self, year: int, month: int, *, columns=None) -> dd.DataFrame:
+        start = pd.Timestamp(year=year, month=month, day=1)
+        end = (start + pd.offsets.MonthEnd(0))
+        return self.read_range(start, end, columns=columns)
+
+    def read_day(self, year: int, month: int, day: int, *, columns=None) -> dd.DataFrame:
+        ts = pd.Timestamp(year=year, month=month, day=day)
+        return self.read_range(ts, ts, columns=columns)
+
+    # ----------------- internals -----------------
+
+    @staticmethod
+    def _pad2(n: int) -> str:
+        return f"{n:02d}"
+
+    def _safe_rm_prefix(self, path: str) -> None:
+        """Per-object delete to avoid S3 bulk DeleteObjects (and Content-MD5 issues)."""
+        if not self.fs.exists(path):
+            return
+        if self.log:
+            self.log.info(f"Cleaning prefix (safe delete): {path}")
+        for k in self.fs.find(path):
+            try:
+                (self.fs.rm_file(k) if hasattr(self.fs, "rm_file") else self.fs.rm(k, recursive=False))
+            except Exception as e:
+                if self.log:
+                    self.log.warning(f"Could not delete {k}: {e}")
+
+    @staticmethod
+    def _require_col(df: dd.DataFrame, col: str) -> None:
+        if col not in df.columns:
+            raise KeyError(f"'{col}' not in DataFrame")
+
+    # ---- DNF builders (string vs int) ----
+    def _dnf_filters_for_range_str(self, start, end) -> DNFFilter:
+        s, e = pd.Timestamp(start), pd.Timestamp(end)
+        if s > e:
+            raise ValueError("start > end")
+        sY, sM, sD = s.year, s.month, s.day
+        eY, eM, eD = e.year, e.month, e.day
+        p2 = self._pad2
+        if sY == eY and sM == eM:
+            return [[("yyyy","==",str(sY)),("mm","==",p2(sM)),("dd",">=",p2(sD)),("dd","<=",p2(eD))]]
+        clauses: DNFFilter = [
+            [("yyyy","==",str(sY)),("mm","==",p2(sM)),("dd",">=",p2(sD))],
+            [("yyyy","==",str(eY)),("mm","==",p2(eM)),("dd","<=",p2(eD))]
+        ]
+        if sY == eY:
+            for m in range(sM+1, eM):
+                clauses.append([("yyyy","==",str(sY)),("mm","==",p2(m))])
+            return clauses
+        for m in range(sM+1, 13):
+            clauses.append([("yyyy","==",str(sY)),("mm","==",p2(m))])
+        for y in range(sY+1, eY):
+            clauses.append([("yyyy","==",str(y))])
+        for m in range(1, eM):
+            clauses.append([("yyyy","==",str(eY)),("mm","==",p2(m))])
+        return clauses
+
+    @staticmethod
+    def _dnf_filters_for_range_int(start, end) -> DNFFilter:
+        s, e = pd.Timestamp(start), pd.Timestamp(end)
+        if s > e:
+            raise ValueError("start > end")
+        sY, sM, sD = s.year, s.month, s.day
+        eY, eM, eD = e.year, e.month, e.day
+        if sY == eY and sM == eM:
+            return [[("yyyy","==",sY),("mm","==",sM),("dd",">=",sD),("dd","<=",eD)]]
+        clauses: DNFFilter = [
+            [("yyyy","==",sY),("mm","==",sM),("dd",">=",sD)],
+            [("yyyy","==",eY),("mm","==",eM),("dd","<=",eD)],
+        ]
+        if sY == eY:
+            for m in range(sM+1, eM):
+                clauses.append([("yyyy","==",sY),("mm","==",m)])
+            return clauses
+        for m in range(sM+1, 13):
+            clauses.append([("yyyy","==",sY),("mm","==",m)])
+        for y in range(sY+1, eY):
+            clauses.append([("yyyy","==",y)])
+        for m in range(1, eM):
+            clauses.append([("yyyy","==",eY),("mm","==",m)])
+        return clauses