test(providers): Mistral batch retry + retry-log detail + comment (#893 r4)

Round-4 review on PR #893 (no blockers, minor items):
- Document why Mistral's _is_transient is SDK-level only (429/5xx): a bare
  connection drop the SDK surfaces as httpx/ConnectionError isn't an SDKError
  and isn't retried here by design — the pod-rollover target is the gateway
  (OpenAI-compatible) path, which does cover connection errors.
- Include the last error (%r) in the retry helper's "not resolved after N
  attempts" error log.
- Add test_mistral_embed_batch_retries_on_5xx (batch path parity with embed()).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

nextcloud_mcp_server/providers/_retry.py

@@ -77,10 +77,11 @@ def retry_on_transient(
                         retry_delay = min(retry_delay * 2, MAX_RETRY_DELAY)
 
             logger.error(
-                "%s %s not resolved after %d attempts",
+                "%s %s not resolved after %d attempts: %r",
                 provider_name,
                 label,
                 MAX_RETRIES,
+                last_error,
             )
             if last_error is None:  # pragma: no cover — loop above always sets this
                 raise RuntimeError("retry loop exited without capturing an error")

nextcloud_mcp_server/providers/mistral.py

@@ -31,7 +31,15 @@ _NO_EMBEDDING_MODEL_MSG = "Embedding not supported - no embedding_model configur
 
 
 def _is_transient(exc: BaseException) -> bool:
-    """Retry HTTP 429 (rate limit) and 5xx (server/transient) SDKErrors."""
+    """Retry HTTP 429 (rate limit) and 5xx (server/transient) SDKErrors.
+
+    Scope is deliberately SDK-level: only ``SDKError`` (an HTTP-status error) is
+    caught by the decorator, so a pure connection drop that the Mistral SDK
+    surfaces as a bare ``httpx``/``ConnectionError`` is NOT retried here. The
+    primary pod-rollover resilience target (card 309) is the gateway path via
+    the OpenAI-compatible client, which does cover connection errors; direct
+    Mistral is a self-hoster fallback where 429/5xx is the common transient.
+    """
     status = getattr(exc, "status_code", None)
     return status == 429 or (isinstance(status, int) and status >= 500)
 

tests/unit/providers/test_mistral.py

@@ -353,3 +353,23 @@ async def test_mistral_embed_retries_on_5xx(mock_mistral_client, monkeypatch):
     embedding = await provider.embed("hello")
     assert embedding == [0.1, 0.2, 0.3]
     assert mock_mistral_client.embeddings.create_async.await_count == 2
+
+
+@pytest.mark.unit
+async def test_mistral_embed_batch_retries_on_5xx(mock_mistral_client, monkeypatch):
+    """The batch path (_embed_batch_request) shares the transient retry too."""
+    from nextcloud_mcp_server.providers import _retry
+
+    monkeypatch.setattr(_retry.anyio, "sleep", AsyncMock(return_value=None))
+
+    err = SDKError.__new__(SDKError)
+    err.status_code = 503
+
+    mock_mistral_client.embeddings.create_async = AsyncMock(
+        side_effect=[err, _make_response([[0.1, 0.2], [0.3, 0.4]])]
+    )
+    provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
+
+    embeddings = await provider.embed_batch(["a", "b"])
+    assert embeddings == [[0.1, 0.2], [0.3, 0.4]]
+    assert mock_mistral_client.embeddings.create_async.await_count == 2