sf-veritas 0.10.3__cp314-cp314-manylinux_2_28_x86_64.whl

sf_veritas/request_utils.py

@@ -0,0 +1,550 @@
+# request_utils.py
+import logging
+import os
+import threading
+import time
+from typing import Any, Callable, List, Optional, Tuple
+
+import orjson
+
+from .env_vars import SF_DEBUG
+from .server_status import server_running
+from .shutdown_flag import is_shutting_down
+from .thread_local import _thread_locals, suppress_logs, suppress_network_recording
+
+# ==========================================================
+# Tunables
+# ==========================================================
+BATCH_MAX = int(os.getenv("SF_NBPOST_BATCH_MAX", "512"))
+BATCH_FLUSH_MS = float(os.getenv("SF_NBPOST_FLUSH_MS", "2"))
+CONNECT_TIMEOUT_S = float(os.getenv("SF_NBPOST_CONNECT_TIMEOUT", "0.1"))
+READ_TIMEOUT_S = float(os.getenv("SF_NBPOST_READ_TIMEOUT", "0.3"))
+TOTAL_TIMEOUT_S = float(os.getenv("SF_NBPOST_TOTAL_TIMEOUT", "0.7"))
+
+# Check if h2 is available for HTTP/2 support
+_HAS_H2 = False
+try:
+    import h2  # noqa: F401
+    _HAS_H2 = True
+except ImportError:
+    pass
+
+# Only enable HTTP/2 if h2 is installed AND env var is set
+HTTP2_ENABLED = os.getenv("SF_NBPOST_HTTP2", "0") == "1" and _HAS_H2
+DISABLE_BATCHING = os.getenv("SF_NBPOST_DISABLE_BATCHING", "0") == "1"
+# Keep gzip OFF by default per perf tests
+GZIP_ENABLED = os.getenv("SF_NBPOST_GZIP", "0") == "1"
+
+
+# ==========================================================
+# Minimal public helpers
+# ==========================================================
+def get_header(request, header_name):
+    return request.headers.get(header_name)
+
+
+def set_header(request, header_name, header_value):
+    request.headers[header_name] = header_value
+
+
+def is_server_running(url="http://localhost:8000/healthz"):
+    """
+    Lightweight liveness probe using stdlib to avoid async spin in caller thread.
+    """
+    global server_running
+    if server_running:
+        return True
+    try:
+        import urllib.request
+
+        with suppress_network_recording(), suppress_logs():
+            with urllib.request.urlopen(url, timeout=0.5) as r:  # nosec
+                if getattr(r, "status", 0) == 200:
+                    server_running = True
+                    return True
+    except Exception:
+        pass
+    return False
+
+
+# ==========================================================
+# Queue & worker
+# ==========================================================
+try:
+    from queue import SimpleQueue  # C fast path
+except Exception:  # pragma: no cover
+    from queue import Queue as SimpleQueue  # fallback
+
+# item variants:
+#   ("POST", url, op, query, variables, wants_response, future)
+#   ("DEFER", builder_callable)  # builder returns (url, op, query, variables)
+_Item = Tuple[Any, ...]
+
+_q: "SimpleQueue[_Item]" = SimpleQueue()
+_started = False
+_start_lock = threading.Lock()
+
+# Worker backends
+_HAS_PYCURL = False
+try:
+    import pycurl  # type: ignore
+
+    _HAS_PYCURL = True
+except Exception:
+    _HAS_PYCURL = False
+
+# httpx client (fallback path)
+_client = None  # type: ignore[assignment]
+
+
+def _ensure_started():
+    global _started
+    if _started:
+        return
+    with _start_lock:
+        if _started:
+            return
+        t = threading.Thread(target=_bg_thread, name="nbpost-batcher", daemon=True)
+        t.start()
+        _started = True
+        if SF_DEBUG:
+            http2_status = HTTP2_ENABLED
+            if not _HAS_H2 and os.getenv("SF_NBPOST_HTTP2", "0") == "1":
+                http2_status = "disabled (h2 not installed)"
+            print(
+                f"[nbpost] started batcher (HTTP/2={http2_status}, batching={'off' if DISABLE_BATCHING else 'on'}, backend={'pycurl' if _HAS_PYCURL else 'httpx'})",
+                log=False,
+            )
+
+
+def _bg_thread():
+    """
+    Background thread: pycurl multi (preferred, C fast path). Falls back to httpx.
+    """
+    if _HAS_PYCURL:
+        _run_worker_pycurl()
+    else:
+        # optional uvloop for httpx path
+        try:
+            import uvloop  # type: ignore
+
+            uvloop.install()
+        except Exception:
+            pass
+        import asyncio
+
+        asyncio.run(_run_worker_httpx())
+
+
+# ==========================================================
+# pycurl backend (C, HTTP/2 via libcurl/nghttp2)
+# ==========================================================
+def _run_worker_pycurl():
+    import pycurl  # type: ignore
+
+    # CRITICAL: Set suppress flag for the ENTIRE thread since this thread is dedicated to sending telemetry
+    # This prevents the C preload library from capturing our telemetry requests
+    from .thread_local import suppress_network_recording_ctx
+    suppress_network_recording_ctx.set(True)
+    if SF_DEBUG:
+        print(f"[pycurl worker] Set suppress_network_recording_ctx to True, current value: {suppress_network_recording_ctx.get()}", log=False)
+
+    m = pycurl.CurlMulti()
+    m.setopt(pycurl.M_MAX_HOST_CONNECTIONS, 1024)
+    m.setopt(pycurl.M_MAXCONNECTS, 1024)
+
+    # CRITICAL: Add marker header so C preload library can identify and skip telemetry requests
+    base_headers = [
+        "Content-Type: application/json",
+        "X-Sf3-TelemetryOutbound: True"
+    ]
+    in_flight = {}
+
+    last_flush = time.monotonic()
+    batch: List[dict] = []
+    url_for_batch: Optional[str] = None
+
+    def _add_easy(url: str, body: bytes, future):
+        c = pycurl.Curl()
+        c.setopt(pycurl.CONNECTTIMEOUT_MS, int(CONNECT_TIMEOUT_S * 1000))
+        c.setopt(pycurl.TIMEOUT_MS, int(TOTAL_TIMEOUT_S * 1000))
+        if HTTP2_ENABLED:
+            c.setopt(pycurl.HTTP_VERSION, pycurl.CURL_HTTP_VERSION_2TLS)
+        c.setopt(pycurl.URL, url.encode("utf-8"))
+        c.setopt(pycurl.NOPROGRESS, True)
+        c.setopt(pycurl.NOSIGNAL, 1)
+        c.setopt(pycurl.POST, 1)
+        c.setopt(pycurl.POSTFIELDS, body)
+        c.setopt(pycurl.POSTFIELDSIZE, len(body))
+        c.setopt(pycurl.WRITEFUNCTION, lambda _b: len(_b))
+        c.setopt(pycurl.HEADERFUNCTION, lambda _b: len(_b))
+
+        headers = list(base_headers)
+        if GZIP_ENABLED:
+            headers.append("Content-Encoding: gzip")
+        c.setopt(pycurl.HTTPHEADER, headers)
+
+        m.add_handle(c)
+        if future is not None:
+            in_flight[c] = future
+
+    def _flush_batch(url: Optional[str], batch_payload: List[dict]):
+        if not batch_payload or not url:
+            return
+        body = orjson.dumps(batch_payload)
+        if GZIP_ENABLED:
+            import gzip
+
+            body = gzip.compress(body)
+        # Note: suppress_network_recording_ctx is already set for the entire thread
+        with suppress_logs():
+            _add_easy(url, body, future=None)
+
+    def _send_one(url: str, op: str, query: str, variables: dict, future):
+        payload = {"query": query, "variables": variables, "operationName": op}
+        body = orjson.dumps(payload)
+        if GZIP_ENABLED:
+            import gzip
+
+            body = gzip.compress(body)
+        # Note: suppress_network_recording_ctx is already set for the entire thread
+        with suppress_logs():
+            _add_easy(url, body, future=future)
+
+    while True:
+        now = time.monotonic()
+        elapsed_ms = (now - last_flush) * 1000.0
+        should_flush = len(batch) >= BATCH_MAX or (
+            batch and elapsed_ms >= BATCH_FLUSH_MS
+        )
+
+        drained = 0
+        try:
+            item = _q.get(timeout=max(0.0005, BATCH_FLUSH_MS / 1000.0))
+            drained += 1
+
+            tag = item[0]
+            if tag == "DEFER":
+                # Build tuple on the worker; keeps request thread extremely light
+                builder: Callable[[], Tuple[str, str, str, dict]] = item[1]
+                url, op, query, variables = builder()
+                if DISABLE_BATCHING:
+                    _send_one(url, op, query, variables, None)
+                else:
+                    if url_for_batch is None:
+                        url_for_batch = url
+                    elif url != url_for_batch:
+                        _flush_batch(url_for_batch, batch)
+                        last_flush = time.monotonic()
+                        batch.clear()
+                        url_for_batch = url
+                    batch.append(
+                        {"query": query, "variables": variables, "operationName": op}
+                    )
+            else:
+                # ("POST", url, op, query, variables, wants_response, future)
+                _, url, op, query, variables, wants_response, fut = item
+                if getattr(_thread_locals, "reentrancy_guard_logging_preactive", False):
+                    variables["reentrancyGuardPreactive"] = True
+                if wants_response and fut is not None:
+                    _send_one(url, op, query, variables, fut)
+                else:
+                    if DISABLE_BATCHING:
+                        _send_one(url, op, query, variables, None)
+                    else:
+                        if url_for_batch is None:
+                            url_for_batch = url
+                        elif url != url_for_batch:
+                            _flush_batch(url_for_batch, batch)
+                            last_flush = time.monotonic()
+                            batch.clear()
+                            url_for_batch = url
+                        batch.append(
+                            {
+                                "query": query,
+                                "variables": variables,
+                                "operationName": op,
+                            }
+                        )
+        except Exception:
+            pass
+
+        if should_flush and not DISABLE_BATCHING:
+            _flush_batch(url_for_batch, batch)
+            last_flush = time.monotonic()
+            batch.clear()
+            url_for_batch = None
+
+        # Pump the multi interface
+        import pycurl as _pc  # local alias for speed
+
+        while True:
+            stat, _ = m.perform()
+            if stat != _pc.E_CALL_MULTI_PERFORM:
+                break
+
+        # Completed transfers
+        while True:
+            num_q, ok_list, err_list = m.info_read()
+            for c in ok_list:
+                fut = in_flight.pop(c, None)
+                if fut is not None:
+                    try:
+                        code = c.getinfo(_pc.RESPONSE_CODE)
+                        fut.set_result(code == 200)
+                    except Exception as e:
+                        fut.set_exception(e)
+                m.remove_handle(c)
+                c.close()
+
+            for c, errno, errmsg in err_list:
+                fut = in_flight.pop(c, None)
+                if fut is not None:
+                    try:
+                        fut.set_result(False)
+                    except Exception:
+                        pass
+                if SF_DEBUG:
+                    print(f"[nbpost] pycurl error {errno}: {errmsg}", log=False)
+                m.remove_handle(c)
+                c.close()
+
+            if num_q == 0:
+                break
+
+        if drained == 0 and not batch:
+            try:
+                m.select(0.01)  # wait for activity (max 10ms)
+            except Exception:
+                pass
+
+
+# ==========================================================
+# httpx backend (fallback)
+# ==========================================================
+async def _run_worker_httpx():
+    global _client
+    import asyncio
+
+    import httpx
+
+    # CRITICAL: Set suppress flag for the ENTIRE thread since this thread is dedicated to sending telemetry
+    # This prevents the C preload library from capturing our telemetry requests
+    from .thread_local import suppress_network_recording_ctx
+    suppress_network_recording_ctx.set(True)
+    if SF_DEBUG:
+        print(f"[httpx worker] Set suppress_network_recording_ctx to True, current value: {suppress_network_recording_ctx.get()}", log=False)
+
+    limits = httpx.Limits(
+        max_connections=1024,
+        max_keepalive_connections=1024,
+        keepalive_expiry=30.0,
+    )
+    timeout = httpx.Timeout(
+        connect=CONNECT_TIMEOUT_S,
+        read=READ_TIMEOUT_S,
+        write=READ_TIMEOUT_S,
+        pool=TOTAL_TIMEOUT_S,
+    )
+    # CRITICAL: Add marker header so C preload library can identify and skip telemetry requests
+    _client = httpx.AsyncClient(
+        http2=HTTP2_ENABLED,
+        limits=limits,
+        timeout=timeout,
+        headers={
+            "Content-Type": "application/json",
+            "X-Sf3-TelemetryOutbound": "True"
+        },
+    )
+
+    try:
+        last_flush = time.monotonic()
+        batch: List[dict] = []
+        url_for_batch: Optional[str] = None
+
+        while True:
+            now = time.monotonic()
+            elapsed_ms = (now - last_flush) * 1000.0
+            should_flush = len(batch) >= BATCH_MAX or (
+                batch and elapsed_ms >= BATCH_FLUSH_MS
+            )
+
+            drained = 0
+            try:
+                item = _q.get(timeout=max(0.0005, BATCH_FLUSH_MS / 1000.0))
+                drained += 1
+
+                tag = item[0]
+                if tag == "DEFER":
+                    builder: Callable[[], Tuple[str, str, str, dict]] = item[1]
+                    url, op, query, variables = builder()
+                    if DISABLE_BATCHING:
+                        await _send_one_httpx(_client, url, op, query, variables, None)
+                    else:
+                        if url_for_batch is None:
+                            url_for_batch = url
+                        elif url != url_for_batch:
+                            await _flush_batch_httpx(_client, url_for_batch, batch)
+                            last_flush = time.monotonic()
+                            batch.clear()
+                            url_for_batch = url
+                        batch.append(
+                            {
+                                "query": query,
+                                "variables": variables,
+                                "operationName": op,
+                            }
+                        )
+                else:
+                    _, url, op, query, variables, wants_response, fut = item
+                    if getattr(
+                        _thread_locals, "reentrancy_guard_logging_preactive", False
+                    ):
+                        variables["reentrancyGuardPreactive"] = True
+                    if wants_response and fut is not None:
+                        await _send_one_httpx(_client, url, op, query, variables, fut)
+                    else:
+                        if DISABLE_BATCHING:
+                            await _send_one_httpx(
+                                _client, url, op, query, variables, None
+                            )
+                        else:
+                            if url_for_batch is None:
+                                url_for_batch = url
+                            elif url != url_for_batch:
+                                await _flush_batch_httpx(_client, url_for_batch, batch)
+                                last_flush = time.monotonic()
+                                batch.clear()
+                                url_for_batch = url
+                            batch.append(
+                                {
+                                    "query": query,
+                                    "variables": variables,
+                                    "operationName": op,
+                                }
+                            )
+            except Exception:
+                pass
+
+            if should_flush and not DISABLE_BATCHING:
+                if batch:
+                    await _flush_batch_httpx(_client, url_for_batch, batch)
+                    last_flush = time.monotonic()
+                    batch.clear()
+                    url_for_batch = None
+
+            if drained == 0 and not batch:
+                await asyncio.sleep(0)
+    finally:
+        try:
+            await _client.aclose()
+        except Exception:
+            pass
+        _client = None
+
+
+async def _flush_batch_httpx(client, url: Optional[str], batch: List[dict]):
+    if not batch or not url:
+        return
+    import httpx
+
+    body = orjson.dumps(batch)
+    headers = {}
+    if GZIP_ENABLED:
+        import gzip
+
+        body = gzip.compress(body)
+        headers["Content-Encoding"] = "gzip"
+    try:
+        # Note: suppress_network_recording_ctx is already set for the entire thread
+        with suppress_logs():
+            r: httpx.Response = await client.post(url, content=body, headers=headers)
+        try:
+            await r.aclose()
+        except Exception:
+            pass
+        if SF_DEBUG:
+            print(f"[nbpost] batch -> {r.status_code}, items={len(batch)}", log=False)
+    except Exception as e:
+        if SF_DEBUG:
+            print(f"[nbpost] batch POST failed: {e}", log=False)
+
+
+async def _send_one_httpx(
+    client, url: str, op: str, query: str, variables: dict, future=None
+):
+    import httpx
+
+    payload = {"query": query, "variables": variables, "operationName": op}
+    body = orjson.dumps(payload)
+    headers = {}
+    if GZIP_ENABLED:
+        import gzip
+
+        body = gzip.compress(body)
+        headers["Content-Encoding"] = "gzip"
+    try:
+        # Note: suppress_network_recording_ctx is already set for the entire thread
+        with suppress_logs():
+            r: httpx.Response = await client.post(url, content=body, headers=headers)
+        status = r.status_code
+        try:
+            await r.aclose()
+        except Exception:
+            pass
+        if future is not None:
+            future.set_result(status == 200)
+    except Exception as e:
+        if future is not None:
+            future.set_exception(e)
+        if SF_DEBUG:
+            print(f"[nbpost] POST failed: {op} {e}", log=False)
+
+
+# ==========================================================
+# Public API (hot path)
+# ==========================================================
+def non_blocking_post(url, operation_name, query, variables):
+    """Enqueue and return immediately."""
+    if is_shutting_down:
+        return None
+    _ensure_started()
+    try:
+        _q.put_nowait(("POST", url, operation_name, query, variables, False, None))
+    except Exception:
+        if SF_DEBUG:
+            print("[nbpost] queue put failed", log=False)
+    return None
+
+
+def non_blocking_post_with_response(url, operation_name, query, variables):
+    """Enqueue and return a Future-like that resolves to bool (success)."""
+    if is_shutting_down:
+        return None
+    _ensure_started()
+    from concurrent.futures import Future
+
+    fut = Future()
+    try:
+        _q.put_nowait(("POST", url, operation_name, query, variables, True, fut))
+    except Exception as e:
+        fut.set_exception(e)
+    return fut
+
+
+def non_blocking_post_deferred(builder: Callable[[], Tuple[str, str, str, dict]]):
+    """
+    Enqueue a zero-alloc builder closure; the worker will call it to build
+    (url, operation_name, query, variables) just-in-time before sending.
+
+    This moves dict construction off the request thread entirely.
+    """
+    if is_shutting_down:
+        return None
+    _ensure_started()
+    try:
+        _q.put_nowait(("DEFER", builder))
+    except Exception:
+        if SF_DEBUG:
+            print("[nbpost] queue put failed (deferred)", log=False)
+    return None

sf_veritas/server_status.py

@@ -0,0 +1 @@
+server_running = False

sf_veritas/shutdown_flag.py

@@ -0,0 +1,11 @@
+import atexit
+
+is_shutting_down = False
+
+
+def set_shutdown_flag():
+    global is_shutting_down
+    is_shutting_down = True
+
+
+atexit.register(set_shutdown_flag)

sf_veritas/subprocess_startup.py

@@ -0,0 +1,3 @@
+from sf_veritas import setup_interceptors
+
+setup_interceptors()  # Set up the interceptors immediately