fix: release cooldown on local media failures

rf4_research/media_cli.py

@@ -7,7 +7,7 @@ from pathlib import Path
 import urllib.error
 import urllib.request
 
-from .community_cli import MIN_FETCH_INTERVAL_SECONDS, USER_AGENT, _StrictRedirectHandler, _read_state, _validate_url_before_io, check_and_reserve, fetch_html, fetch_site_key
+from .community_cli import MIN_FETCH_INTERVAL_SECONDS, USER_AGENT, _StrictRedirectHandler, _read_state, _validate_url_before_io, check_and_reserve, fetch_html, fetch_site_key, release_reservation
 from .media_assets import approve_stored_assets, audit_media_catalog, compare_quality_upgrades, extract_media_candidates, generate_media_derivatives, generate_quality_contact_sheets, media_coverage, media_quality_report, merge_manifest, publish_quality_upgrades, queue_quality_upgrades, reconcile_queued_duplicates, reclassify_manifest, review_asset, rollback_quality_upgrade, store_asset
 
 
@@ -104,6 +104,11 @@ def _attempt_asset(root: Path, manifest_path: Path, manifest: dict, queued: dict
     return f"stored {url} as {relative}", None
 
 
+def _response_was_received(error: Exception) -> bool:
+    """Only an HTTP response or validated body is allowed to consume cooldown."""
+    return isinstance(error, (urllib.error.HTTPError, ValueError))
+
+
 def _download_one(root: Path, state_file: Path, asset_url: str | None = None) -> str:
     manifest_path = root / "manifest.json"
     manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
@@ -112,9 +117,11 @@ def _download_one(root: Path, state_file: Path, asset_url: str | None = None) ->
         return "queue is empty" if asset_url is None else "asset is not queued"
     url = queued["asset_url"]
     _validate_url_before_io(url)
-    check_and_reserve(fetch_site_key(url), state_file=state_file)
+    reserved_at = check_and_reserve(fetch_site_key(url), state_file=state_file)
     message, error = _attempt_asset(root, manifest_path, manifest, queued)
     if error is not None:
+        if not _response_was_received(error):
+            release_reservation(fetch_site_key(url), state_file=state_file, reserved_at=reserved_at)
         raise error
     return message
 
@@ -142,6 +149,7 @@ def _download_batch(root: Path, state_file: Path, *, limit: int = MAX_BATCH_ASSE
             domains[site] = {"attempted": 0, "stored": 0, "failed": 0, "skipped": str(exc)}
             continue
         stored = failed = consecutive_failures = 0
+        response_received = False
         messages: list[str] = []
         stopped_reason: str | None = None
         for item in items[:limit]:
@@ -149,17 +157,24 @@ def _download_batch(root: Path, state_file: Path, *, limit: int = MAX_BATCH_ASSE
             messages.append(message)
             if error is None:
                 stored += 1
+                response_received = True
                 consecutive_failures = 0
                 continue
             failed += 1
             consecutive_failures += 1
             status = item.get("status")
+            if _response_was_received(error):
+                response_received = True
             if status in {"blocked", "queued"}:
                 stopped_reason = f"domain stopped after {status} response"
                 break
             if consecutive_failures >= MAX_CONSECUTIVE_FAILURES:
                 stopped_reason = f"domain stopped after {MAX_CONSECUTIVE_FAILURES} consecutive failures"
                 break
+        if not response_received and failed and stored == 0:
+            # No HTTP response was observed in this reservation window. It is
+            # safe to release this process's reservation for a later retry.
+            release_reservation(site, state_file=state_file, reserved_at=reserved_at)
         domains[site] = {
             "attempted": stored + failed, "stored": stored, "failed": failed,
             "stopped_reason": stopped_reason, "messages": messages,