classifyre-cli 0.4.8__tar.gz → 0.4.9__tar.gz

{classifyre_cli-0.4.8 → classifyre_cli-0.4.9}/.turbo/turbo-build.log

@@ -1,3 +1,3 @@
 $ uv sync
-Resolved 256 packages in 185ms
+Resolved 256 packages in 201ms
 Checked 50 packages in 1ms

{classifyre_cli-0.4.8 → classifyre_cli-0.4.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: classifyre-cli
-Version: 0.4.8
+Version: 0.4.9
 Summary: Classifyre CLI — scan and classify unstructured data sources
 License: MIT
 Keywords: data,ingestion,metadata,pii,secrets,unstructured

{classifyre_cli-0.4.8 → classifyre_cli-0.4.9}/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@classifyre/cli",
-  "version": "0.4.8",
+  "version": "0.4.9",
   "private": true,
   "scripts": {
     "build": "uv sync",

{classifyre_cli-0.4.8 → classifyre_cli-0.4.9}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "classifyre-cli"
-version = "0.4.8"
+version = "0.4.9"
 description = "Classifyre CLI — scan and classify unstructured data sources"
 readme = "README.md"
 requires-python = ">=3.12"

{classifyre_cli-0.4.8 → classifyre_cli-0.4.9}/src/outputs/rest.py

@@ -1,17 +1,85 @@
 from __future__ import annotations
 
 import logging
+import random
 from typing import Any, Literal, cast
 from urllib.parse import urljoin
 
 import requests  # type: ignore[import-untyped]
 from pydantic import BaseModel, ConfigDict, Field
+from requests.adapters import HTTPAdapter
+from urllib3.util.retry import Retry  # type: ignore[import-untyped]
 
 from .base import OutputRuntimeContext, OutputType
 
 logger = logging.getLogger(__name__)
 
 
+class _JitteredRetry(Retry):
+    """urllib3 Retry subclass that adds ±25 % multiplicative jitter to the
+    computed backoff so that multiple concurrent CLI jobs do not all retry
+    at exactly the same moment (thundering-herd mitigation).
+
+    The jitter is applied *after* the standard exponential backoff formula
+    and the backoff_max cap, so it never pushes the delay above
+    backoff_max * 1.25.
+    """
+
+    _JITTER_FACTOR: float = 0.25
+
+    def get_backoff_time(self) -> float:  # type: ignore[override]
+        base = super().get_backoff_time()
+        if base == 0:
+            return 0.0
+        lo = base * (1 - self._JITTER_FACTOR)
+        hi = base * (1 + self._JITTER_FACTOR)
+        return random.uniform(lo, hi)
+
+
+# Retry policy for CLI → API REST calls.
+#
+# What we retry and why:
+#   connect=8  — pod restarted / not yet ready (RemoteDisconnected, ConnectionReset,
+#                ConnectTimeout). Request never reached the application.
+#   read=8     — API is under load and slow to respond (ReadTimeout). Safe to retry
+#                because all endpoints are idempotent (bulk ingest is upsert-based,
+#                status/findings updates are set-operations).
+#   status=8   — transient HTTP errors from an overloaded or restarting API:
+#                  408 Request Timeout   - API-level timeout
+#                  429 Too Many Requests - rate-limited / backpressure
+#                  502 Bad Gateway       - proxy has no upstream yet
+#                  503 Service Unavail.  - under-pressure / pod not ready
+#                  504 Gateway Timeout   - upstream took too long
+#
+# backoff_factor=2, backoff_max=60: exponential cap at 60 s, with ±25 % jitter
+# (see _JitteredRetry). Approximate wait schedule between attempts:
+#   attempt 1 → immediate (0 s)
+#   attempt 2 → ~2 s
+#   attempt 3 → ~4 s
+#   attempt 4 → ~8 s
+#   attempt 5 → ~16 s
+#   attempt 6 → ~32 s
+#   attempt 7 → ~60 s  (capped)
+#   attempt 8 → ~60 s  (capped)
+# Total extra wait: ~182 s (~3 min) — covers extended load spikes on a
+# single-node VPS before event-loop pressure drops. Worst-case a single
+# call costs 8 * 120 s + 182 s = ~18 min, acceptable for long-running scans.
+#
+# POST and PATCH are explicitly allowed: without this urllib3 only retries
+# idempotent methods (GET/HEAD) by default.
+_RETRY_POLICY = _JitteredRetry(
+    total=8,
+    connect=8,
+    read=8,
+    status=8,
+    backoff_factor=2,
+    backoff_max=60,
+    status_forcelist={408, 429, 502, 503, 504},
+    allowed_methods={"GET", "POST", "PUT", "PATCH", "DELETE", "HEAD", "OPTIONS"},
+    raise_on_status=False,
+)
+
+
 def _drop_none_recursive(value: Any) -> Any:
     if isinstance(value, dict):
         return {key: _drop_none_recursive(item) for key, item in value.items() if item is not None}
@@ -63,6 +131,9 @@ class RestOutputSink:
         self.base_url = base_url.rstrip("/")
         self.timeout_sec = timeout_sec
         self.session = requests.Session()
+        adapter = HTTPAdapter(max_retries=_RETRY_POLICY)
+        self.session.mount("http://", adapter)
+        self.session.mount("https://", adapter)
         self._runner_id = context.runner_id
         self._seen_hashes: set[str] = set()
 

{classifyre_cli-0.4.8 → classifyre_cli-0.4.9}/tests/test_outputs.py

@@ -158,6 +158,9 @@ class _FakeSession:
         self.responses = responses
         self.calls: list[dict[str, Any]] = []
 
+    def mount(self, prefix: str, adapter: Any) -> None:  # noqa: ARG002
+        """Absorb HTTPAdapter.mount() calls from RestOutputSink.__init__."""
+
     def request(
         self,
         method: str,