fix: bound chat latency and clarify blackjack flow

packages/jwebot-kernel/pyproject.toml

@@ -4,13 +4,13 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "jwebot-kernel"
-version = "0.6.1a0+1015"
+version = "0.6.2a0+1015"
 description = "Stable routing and extension authority for JWeBot"
 requires-python = ">=3.9"
 dependencies = [
-  "jwebot-config==0.6.1a0+1015",
-  "jwebot-sdk==0.6.1a0+1015",
-  "jwebot-official-extensions==0.6.1a0+1015",
+  "jwebot-config==0.6.2a0+1015",
+  "jwebot-sdk==0.6.2a0+1015",
+  "jwebot-official-extensions==0.6.2a0+1015",
   "tomli>=2; python_version < '3.11'",
 ]
 

packages/jwebot-kernel/src/jwebot_kernel/__init__.py

@@ -107,4 +107,4 @@ __all__ = [
     "verify_installed_release_signature",
 ]
 
-__version__ = "0.6.1a0+1015"
+__version__ = "0.6.2a0+1015"

packages/jwebot-kernel/src/jwebot_kernel/edge_runtime.py

@@ -92,6 +92,10 @@ MAX_REPLY_CHARS = 900
 MAX_CONTEXT_TURNS = 6
 MAX_OUTBOX_ATTEMPTS = 3
 DIRECT_CHAT_GENERATION_WORKERS = 2
+DIRECT_API_SOCKET_TIMEOUT_SECONDS = 18.0
+DIRECT_API_WALL_TIMEOUT_SECONDS = 30.0
+DIRECT_API_MAX_RESPONSE_BYTES = 2_000_000
+DIRECT_API_READ_CHUNK_BYTES = 65_536
 FAST_ACKNOWLEDGEMENT = "喵......"
 PROFILE_MIN_AUTOMATIC_OBSERVATIONS = 3
 PROFILE_MAX_FACTS = 12
@@ -626,6 +630,59 @@ def deepseek_response_content(raw: bytes) -> str:
     return str(content)
 
 
+def read_direct_api_response(
+    response: Any,
+    *,
+    wall_timeout_seconds: float = DIRECT_API_WALL_TIMEOUT_SECONDS,
+    max_bytes: int = DIRECT_API_MAX_RESPONSE_BYTES,
+) -> bytes:
+    """Read an SSE response with an absolute deadline, not an idle timeout."""
+
+    deadline = time.monotonic() + wall_timeout_seconds
+    chunks: list[bytes] = []
+    total = 0
+    stream_tail = b""
+    done_marker = re.compile(
+        rb"(?:^|\r?\n)data:\s*\[DONE\]\s*(?:\r?\n|$)"
+    )
+    socket_connection = getattr(
+        getattr(getattr(response, "fp", None), "raw", None),
+        "_sock",
+        None,
+    )
+    read_chunk = getattr(response, "read1", response.read)
+    while True:
+        remaining = deadline - time.monotonic()
+        if remaining <= 0:
+            raise DeepSeekUnavailable("direct API exceeded total timeout")
+        if socket_connection is not None:
+            socket_connection.settimeout(
+                max(
+                    0.001,
+                    min(DIRECT_API_SOCKET_TIMEOUT_SECONDS, remaining),
+                )
+            )
+        try:
+            chunk = read_chunk(
+                min(DIRECT_API_READ_CHUNK_BYTES, max_bytes - total + 1)
+            )
+        except (TimeoutError, OSError) as error:
+            raise DeepSeekUnavailable("direct API transport failure") from error
+        if not chunk:
+            break
+        if not isinstance(chunk, bytes):
+            raise DeepSeekUnavailable("direct API returned invalid bytes")
+        total += len(chunk)
+        if total > max_bytes:
+            raise DeepSeekUnavailable("direct API response exceeded size limit")
+        chunks.append(chunk)
+        stream_scan = stream_tail + chunk
+        if done_marker.search(stream_scan):
+            break
+        stream_tail = stream_scan[-64:]
+    return b"".join(chunks)
+
+
 def load_environment_file(path: Path = ENVIRONMENT_FILE) -> None:
     """Load the active generated profile, then the legacy compatibility file."""
 
@@ -3422,8 +3479,11 @@ class DeepSeekClient:
             method="POST",
         )
         try:
-            with urllib.request.urlopen(request, timeout=18) as response:
-                raw = response.read(2_000_000)
+            with urllib.request.urlopen(
+                request,
+                timeout=DIRECT_API_SOCKET_TIMEOUT_SECONDS,
+            ) as response:
+                raw = read_direct_api_response(response)
         except urllib.error.HTTPError as error:
             raise DeepSeekUnavailable(f"direct API HTTP {error.code}") from error
         except (urllib.error.URLError, TimeoutError, OSError) as error: