fix: handle SoundCloud API 5xx errors with client_id refresh, backoff, and cursor fallback

SoundCloud began rejecting the fabricated pagination cursor with 500
errors. Fixed cursor user_id padding (zfill 22→20) to match the
documented format, added 5xx retry with exponential backoff in _api_get,
and added a fallback in fetch_likes that drops the fabricated cursor
when it causes persistent 500s.

Made-with: Cursor

src/ntr_fetcher/soundcloud.py

@@ -1,3 +1,4 @@
+import asyncio
 import json
 import logging
 import re
@@ -17,7 +18,7 @@ HYDRATION_PATTERN = re.compile(r"__sc_hydration\s*=\s*(\[.*?\])\s*;", re.DOTALL)
 
 def _build_cursor(until: datetime, user_id: int) -> str:
     ts = until.strftime("%Y-%m-%dT%H:%M:%S.000Z")
-    padded_user = str(user_id).zfill(22)
+    padded_user = str(user_id).zfill(20)
     return f"{ts},user-track-likes,000-{padded_user}-99999999999999999999"
 
 
@@ -55,19 +56,27 @@ class SoundCloudClient:
         params = dict(params or {})
         params["client_id"] = client_id
 
-        for attempt in range(3):
+        max_attempts = 3
+        for attempt in range(max_attempts):
             resp = await self._http.get(url, params=params)
-            if resp.status_code == 401:
-                logger.warning("Got 401 from SoundCloud API, refreshing client_id (attempt %d)", attempt + 1)
+
+            if resp.status_code == 401 or resp.status_code >= 500:
+                logger.warning(
+                    "Got %d from SoundCloud API, refreshing client_id (attempt %d/%d)",
+                    resp.status_code, attempt + 1, max_attempts,
+                )
                 self.invalidate_client_id()
+                if resp.status_code >= 500:
+                    await asyncio.sleep(2 ** attempt)
                 client_id = await self._extract_client_id()
                 params["client_id"] = client_id
                 continue
+
             resp.raise_for_status()
             return resp
 
         raise httpx.HTTPStatusError(
-            "Failed after 3 attempts (401)",
+            f"Failed after {max_attempts} attempts (last status: {resp.status_code})",
             request=resp.request,
             response=resp,
         )
@@ -86,15 +95,24 @@ class SoundCloudClient:
         until: datetime,
         limit: int = 50,
     ) -> list[Track]:
-        cursor = _build_cursor(until, user_id)
+        cursor: str | None = _build_cursor(until, user_id)
         collected: list[Track] = []
+        used_fabricated_cursor = True
 
         while True:
             params: dict = {"limit": limit}
             if cursor:
                 params["offset"] = cursor
 
-            resp = await self._api_get(f"{API_BASE}/users/{user_id}/likes", params=params)
+            try:
+                resp = await self._api_get(f"{API_BASE}/users/{user_id}/likes", params=params)
+            except httpx.HTTPStatusError as exc:
+                if used_fabricated_cursor and cursor and exc.response.status_code >= 500:
+                    logger.warning("Fabricated cursor rejected (HTTP %d), retrying without cursor", exc.response.status_code)
+                    cursor = None
+                    used_fabricated_cursor = False
+                    continue
+                raise
             data = resp.json()
             collection = data.get("collection", [])