feat: enhance batch processing and warmup logic in proxy server and domain fronter

src/core/constants.py

@@ -73,9 +73,9 @@ WARM_POOL_COUNT         = 30
 
 
 # ── Batch windows ─────────────────────────────────────────────────────────
-BATCH_WINDOW_MICRO      = 0.005   # 5 ms
+BATCH_WINDOW_MICRO      = 0.015   # 15 ms: captures short asset bursts
-BATCH_WINDOW_MACRO      = 0.050   # 50 ms
+BATCH_WINDOW_MACRO      = 0.120   # 120 ms: balances batching and latency
-BATCH_MAX               = 50
+BATCH_MAX               = 64      # enough headroom for browser fan-out waves
 
 
 # ── Fan-out relay (parallel Apps Script instances) ────────────────────────
@@ -201,7 +201,7 @@ TRACE_HOST_SUFFIXES: tuple[str, ...] = (
 STATIC_EXTS: tuple[str, ...] = (
     ".css", ".js", ".mjs", ".woff", ".woff2", ".ttf", ".eot",
     ".png", ".jpg", ".jpeg", ".gif", ".webp", ".svg", ".ico",
-    ".mp3", ".mp4", ".webm", ".wasm", ".avif",
+    ".mp3", ".mp4", ".webm", ".wasm", ".avif", ".json", ".map",
 )
 LARGE_FILE_EXTS = frozenset({
     ".bin",

src/proxy/proxy_server.py

@@ -105,6 +105,8 @@ class ProxyServer:
         self._download_max_chunks = self._cfg_int(
             config, "chunked_download_max_chunks", 256, minimum=1,
         )
+        self._warmup_before_listen = True
+        self._warmup_timeout = 20.0
         self._download_extensions, self._download_any_extension = (
             self._normalize_download_extensions(
                 config.get(
@@ -228,6 +230,20 @@ class ProxyServer:
         return self.fronter._is_static_asset_url(url)
 
     async def start(self):
+        if self._warmup_before_listen:
+            log.info(
+                "Relay warmup in progress... waiting up to %.0fs before opening listeners",
+                self._warmup_timeout,
+            )
+            ready = await self.fronter.wait_until_warm(timeout=self._warmup_timeout)
+            if ready:
+                log.info("Relay warmup complete — enabling HTTP/SOCKS listeners")
+            else:
+                log.warning(
+                    "Relay warmup timed out after %.0fs — starting listeners anyway",
+                    self._warmup_timeout,
+                )
+
         http_srv = await asyncio.start_server(self._on_client, self.host, self.port)
         socks_srv = None
 

src/relay/domain_fronter.py

@@ -201,6 +201,12 @@ class DomainFronter:
             "Execution monitor enabled: reporting total every %.0fs",
             self._execution_report_interval,
         )
+        log.info(
+            "Batch config: micro=%.0fms macro=%.0fms max=%d",
+            self._batch_window_micro * 1000.0,
+            self._batch_window_macro * 1000.0,
+            self._batch_max,
+        )
 
         # Exit node — optional second-hop relay with a non-Google exit IP.
         # Useful for sites that block GCP/Apps Script IPs (e.g. ChatGPT).
@@ -881,6 +887,23 @@ class DomainFronter:
         # active its _keepalive_loop skips the ping; they do not double-fire.
         self._spawn(self._h1_container_keepalive())
 
+    async def wait_until_warm(self, timeout: float | None = None) -> bool:
+        """Start warmup and wait until the initial pool-open phase finishes.
+
+        Returns True if warmup finished before timeout, else False.
+        """
+        await self._warm_pool()
+        if self._pool_ready.is_set():
+            return True
+        try:
+            if timeout is None or timeout <= 0:
+                await self._pool_ready.wait()
+            else:
+                await asyncio.wait_for(self._pool_ready.wait(), timeout=timeout)
+            return True
+        except asyncio.TimeoutError:
+            return False
+
     def _spawn(self, coro) -> asyncio.Task:
         """Create a task and keep a strong reference for clean cancellation."""
         task = asyncio.create_task(coro)
@@ -1801,20 +1824,32 @@ class DomainFronter:
         if method not in {"GET", "HEAD"} or body:
             return True
 
+        # Static assets are safe to batch in parallel as independent requests.
+        is_static = cls._is_static_asset_url(url)
+
         if headers:
-            for name in STATEFUL_HEADER_NAMES:
+            # Hard stateful markers: preserve strict ordering / isolation.
+            for name in ("cookie", "authorization", "proxy-authorization"):
                 if cls._header_value(headers, name):
                     return True
 
             accept = cls._header_value(headers, "accept").lower()
-            if "text/html" in accept or "application/json" in accept:
+            if "text/html" in accept:
                 return True
 
             fetch_mode = cls._header_value(headers, "sec-fetch-mode").lower()
-            if fetch_mode in {"navigate", "cors"}:
+            if fetch_mode == "navigate":
                 return True
 
-        return not cls._is_static_asset_url(url)
+            fetch_dest = cls._header_value(headers, "sec-fetch-dest").lower()
+            if fetch_dest in {"document", "iframe", "frame"}:
+                return True
+
+            # Non-static JSON/API calls are treated as stateful by default.
+            if (not is_static) and "application/json" in accept:
+                return True
+
+        return not is_static
 
     # ── Batch collector ───────────────────────────────────────────
 
@@ -2241,14 +2276,9 @@ class DomainFronter:
                           payloads: list[dict]) -> list[bytes]:
         """Parse a batch response body into individual results."""
         text = resp_body.decode(errors="replace").strip()
-        try:
+        # Apps Script can wrap JSON inside an HTML shell; reuse the same
-            data = json.loads(text)
+        # robust loader used by single-response parsing.
-        except json.JSONDecodeError:
+        data = load_relay_json(text)
-            m = re.search(r'\{.*\}', text, re.DOTALL)
-            try:
-                data = json.loads(m.group()) if m else None
-            except json.JSONDecodeError:
-                data = None
         if not data:
             raise RuntimeError(f"Bad batch response: {text[:200]}")