flowtask 5.12.3__tar.gz → 5.12.5__tar.gz

{flowtask-5.12.3 → flowtask-5.12.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flowtask
-Version: 5.12.3
+Version: 5.12.5
 Summary: Framework for Task orchestration
 Author-email: Jesus Lara <jesuslarag@gmail.com>
 License: Apache-2.0
@@ -37,7 +37,7 @@ Requires-Dist: selenium>=4.35.0
 Requires-Dist: snapshot-selenium==0.0.2
 Requires-Dist: httpx[http2,socks]>=0.26.0
 Requires-Dist: urllib3[socks]>=2.3.0
-Requires-Dist: requests>=2.33.0
+Requires-Dist: requests<2.34,>=2.33.0
 Requires-Dist: h2>=4.2.0
 Requires-Dist: blinker>=1.9.0
 Requires-Dist: webdriver-manager==4.0.2
@@ -64,10 +64,9 @@ Requires-Dist: datatable==1.1.0
 Requires-Dist: imageio==2.37.0
 Requires-Dist: decorator>=5.1.1
 Requires-Dist: emoji==2.14.1
-Requires-Dist: caio==0.9.24
 Requires-Dist: Wand==0.6.13
 Requires-Dist: pylibdmtx==0.1.10
-Requires-Dist: aiofile>=3.9.0
+Requires-Dist: aiofiles>=24.1.0
 Requires-Dist: filetype==1.2.0
 Requires-Dist: proxylists>=0.14.0
 Requires-Dist: aioftp==0.23.1
@@ -127,10 +126,12 @@ Requires-Dist: kr8s>=0.16.0; extra == "k8s"
 Provides-Extra: git
 Requires-Dist: gitpython>=3.1.50; extra == "git"
 Provides-Extra: ai
-Requires-Dist: ai-parrot[agents,images,integrations,llms,scheduler]>=0.20.4; extra == "ai"
+Requires-Dist: ai-parrot[agents,graphindex,images,llms,retrieval]>=0.20.4; extra == "ai"
 Requires-Dist: ai-parrot-loaders[audio,document,ebook,ml,pdf,scraping,video,web,youtube]>=0.1.0; extra == "ai"
 Requires-Dist: ai-parrot-tools[analysis,arxiv,aws,codeinterpreter,db,docker,excel,finance,flowtask,git,google,jira,messaging,pdf,pulumi,sandbox,scraping,sitesearch,slack,weather,wikipedia]>=0.1.0; extra == "ai"
 Requires-Dist: ai-parrot-pipelines>=0.1.0; extra == "ai"
+Requires-Dist: ai-parrot-server[all]>=0.1.0; extra == "ai"
+Requires-Dist: ai-parrot-integrations[all]>=0.0.17; extra == "ai"
 Provides-Extra: dev
 Requires-Dist: pytest>=5.4.0; extra == "dev"
 Requires-Dist: coverage; extra == "dev"

{flowtask-5.12.3 → flowtask-5.12.5}/flowtask/components/CallAnalysis.py

@@ -2,7 +2,7 @@ from collections.abc import Callable
 import asyncio
 import pandas as pd
 import orjson
-from ..exceptions import DataNotFound
+from ..exceptions import DataNotFound, ComponentError
 from ..interfaces.ParrotBot import ParrotBot
 from ..interfaces.flow import FlowComponent
 
@@ -236,6 +236,49 @@ class CallAnalysis(ParrotBot, FlowComponent):
         self.skip_existing: bool = kwargs.get('skip_existing', True)
         self.lookup_schema_column: str = kwargs.get('schema_column', 'program_slug')
 
+        # Credentials/auth abort state (set per bot_evaluation run).
+        # When a non-retryable credentials error is detected, _auth_error holds
+        # the offending exception and _abort_event signals every in-flight/queued
+        # task to bail out so no further API calls are attempted.
+        self._auth_error: Exception | None = None
+        self._abort_event: asyncio.Event | None = None
+
+    # Substrings that flag a non-retryable credentials/permission failure.
+    # Retrying these only wastes API quota — the key/permissions must be fixed.
+    _CREDENTIALS_ERROR_MARKERS = (
+        "PERMISSION_DENIED",
+        "API_KEY_SERVICE_BLOCKED",
+        "API_KEY_INVALID",
+        "API key not valid",
+        "API_KEY_HTTP_REFERRER_BLOCKED",
+        "UNAUTHENTICATED",
+        "CONSUMER_INVALID",
+    )
+
+    def _is_credentials_error(self, exc: Exception) -> bool:
+        """Return True if ``exc`` is a non-retryable credentials/auth failure.
+
+        Detects the Google GenAI ``403 PERMISSION_DENIED`` /
+        ``API_KEY_SERVICE_BLOCKED`` family (and equivalent auth errors) so the
+        batch can fail fast instead of retrying a blocked API key.
+
+        Args:
+            exc: The exception raised by the LLM provider call.
+
+        Returns:
+            bool: True when the error indicates blocked/invalid credentials.
+        """
+        # HTTP status carried on the exception (google-genai / aiohttp styles).
+        status = (
+            getattr(exc, 'status_code', None)
+            or getattr(exc, 'code', None)
+            or getattr(exc, 'status', None)
+        )
+        if status in (401, 403):
+            return True
+        message = str(exc)
+        return any(marker in message for marker in self._CREDENTIALS_ERROR_MARKERS)
+
     def load_from_file(
         self,
         df: pd.DataFrame,
@@ -651,6 +694,13 @@ class CallAnalysis(ParrotBot, FlowComponent):
         """
         call_key = call_key or call_identifier
         async with semaphore:
+            # A fatal credentials error already aborted the batch — bail out
+            # without making another doomed API call.
+            if self._abort_event is not None and self._abort_event.is_set():
+                self._logger.debug(
+                    f"Skipping {call_identifier} - batch aborted due to credentials error"
+                )
+                return (call_key, None)
             for attempt in range(self.retry_attempts):
                 try:
                     # Wait minimum interval before request to respect RPM limit
@@ -673,6 +723,23 @@ class CallAnalysis(ParrotBot, FlowComponent):
 
 
                 except Exception as e:
+                    # Credentials / permission errors are NOT retryable: retrying
+                    # only burns API quota against a blocked key. Flag the batch
+                    # for abort so no further calls are attempted and surface it
+                    # clearly (distinct from DataNotFound) in bot_evaluation.
+                    if self._is_credentials_error(e):
+                        if self._auth_error is None:
+                            self._auth_error = e
+                            self._logger.error(
+                                f"🔑 Credentials/permission error from LLM provider while "
+                                f"analyzing {call_identifier}: {e}. Aborting batch — the API "
+                                f"key is blocked/invalid or lacks permission. No further "
+                                f"calls will be attempted."
+                            )
+                        if self._abort_event is not None:
+                            self._abort_event.set()
+                        return (call_key, None)
+
                     if attempt < self.retry_attempts - 1:
                         # Calculate exponential backoff wait time
                         wait_time = self.retry_backoff_factor ** attempt
@@ -704,6 +771,10 @@ class CallAnalysis(ParrotBot, FlowComponent):
             A Pandas Dataframe with the Call Analysis results.
 
         """
+        # Reset credentials-abort state for this run.
+        self._auth_error = None
+        self._abort_event = asyncio.Event()
+
         # Determine if we should group the data or process row by row
         should_group = self.group_analysis
         schema_col_present = (
@@ -944,6 +1015,17 @@ class CallAnalysis(ParrotBot, FlowComponent):
             # Add the analysis results as a new column
             result_df[self.output_column] = analysis_results
 
+        # If a fatal credentials/permission error aborted the batch, surface it
+        # explicitly. Do NOT let it fall through to the DataNotFound below — that
+        # status (204, "nothing to analyze") would mask a blocked/invalid API key.
+        if self._auth_error is not None:
+            raise ComponentError(
+                f"{self._bot_name}: aborted analysis due to a credentials/permission "
+                f"error from the LLM provider ({self._auth_error}). The API key is "
+                f"blocked, invalid, or lacks permission to call the model. Fix the "
+                f"credentials and retry — no further API calls were attempted."
+            )
+
         # Clean up JSON formatting in the output column if present
         # Only clean strings, preserve dicts/lists that are already valid JSON objects
         if self.output_column in result_df.columns and not result_df[self.output_column].isna().all():

{flowtask-5.12.3 → flowtask-5.12.5}/flowtask/components/OpenWithBase.py

@@ -56,15 +56,13 @@ def detect_encoding(filename, encoding: str = "utf-8"):
         try:
             result = chardet.detect(raw)
             if result["encoding"] in ("ASCII", "ascii"):
-                # let me try to repair the file:
-                content = None
-                with open(filename, "rb") as fp:
-                    content = fp.read()
-                decoded = content.decode(encoding, "ignore")
-                output = StringIO()
-                output.write(decoded)
-                output.seek(0)
-                return [output, encoding]
+                # ASCII is a strict subset of UTF-8, so the declared encoding
+                # is safe to use as-is.  Do NOT slurp the whole file here: the
+                # only thing every caller consumes from this function is the
+                # encoding string (the buffer is always discarded), and reading
+                # a multi-GB file just to throw it away caused massive memory
+                # pressure / apparent hangs on large CSVs.
+                return [None, encoding]
             else:
                 enc = result["encoding"]
         except UnicodeDecodeError:
@@ -573,7 +571,7 @@ class OpenWithBase(FlowComponent):
                 encoding = self.encoding
             else:
                 count = 0
-                # migrate to aiofile
+                # migrate to aiofiles
                 bt = min(32, os.path.getsize(filename))
                 with open(filename, "rb") as f:
                     line = f.read(bt)

{flowtask-5.12.3 → flowtask-5.12.5}/flowtask/components/OpenWithPandas.py

@@ -539,11 +539,13 @@ class OpenWithPandas(OpenWithBase):
                     raise ComponentError(
                         f"Error Combining Resultset Dataframes: {err}"
                     ) from err
-        # When --debug is on, print the RAW DataFrame exactly as read, BEFORE
-        # any post-processing (drop_empty/trim/dropna). This is the place to
-        # validate which columns came in and which are all-empty (and would be
-        # dropped below by drop_empty).
-        if self._debug is True:
+        # When is_debug is on, log the RAW read shape/columns BEFORE any
+        # post-processing (drop_empty/trim/dropna) — useful to validate which
+        # columns came in and which are all-empty (and would be dropped below
+        # by drop_empty). Gated behind the explicit is_debug YAML flag (NOT the
+        # global --debug) so normal debug runs don't flood the logs with the
+        # per-column non-null dump on every OpenWithPandas step.
+        if getattr(self, 'is_debug', False) is True:
             try:
                 nonnull = df.notna().sum()
                 empty_cols = [c for c in df.columns if nonnull[c] == 0]
@@ -656,11 +658,15 @@ class OpenWithPandas(OpenWithBase):
         self.add_metric("NUMROWS", numrows)
         self.add_metric("OPENED_FILES", self._filenames)
         if getattr(self, 'is_debug', False) is True:
-            self._logger.debug("DataFrame result:\n%s", df.to_string())
+            # Do NOT render the full DataFrame (df.to_string() with no row limit
+            # materialises every row into one giant string — CPU-bound for
+            # minutes/hours on multi-GB frames, the apparent "hang"). Instead,
+            # show the shape and a single representative row: every column with
+            # its dtype and its first value.
+            self._logger.debug("DataFrame result: shape=%s", df.shape)
             self._logger.debug("::: Column Information ===")
             for column, t in df.dtypes.items():
-                self._logger.debug(
-                    "%s -> %s -> %s", column, t, df[column].iloc[0]
-                )
+                sample = df[column].iloc[0] if len(df.index) else "<empty>"
+                self._logger.debug("%s -> %s -> %s", column, t, sample)
             self._logger.debug("Opened File(s) with Pandas %s", self._filenames)
         return self._result

flowtask-5.12.5/flowtask/components/ParrotKnowledgeBuilder/__init__.py

@@ -0,0 +1,5 @@
+"""ParrotKnowledgeBuilder — builds GraphIndex/PageIndex knowledge structures from documents."""
+from .loader import ParrotKnowledgeBuilder
+from .payload import KnowledgeArtifact
+
+__all__ = ["KnowledgeArtifact", "ParrotKnowledgeBuilder"]