fix(ci): silence sidecar flush-timeout log noise in client_side_stats test

.gitlab/generate-tracer.php

@@ -367,6 +367,16 @@ function before_script_steps($with_docker_auth = false) {
     ARCH: "amd64"
   timeout: 120m
   script:
+    # Start an always-200 sink on the default agent endpoint (localhost:8126). The bulk of
+    # the suite runs with the ddtrace default agent (localhost:8126), which has no listener
+    # here -> those sidecar trace sends ECONNREFUSED and retry, clogging the shared sidecar
+    # trace-flusher and stalling unrelated tests' synchronous flush past the 10s IPC timeout
+    # (flaky client_side_stats "Kind(TimedOut)" / "wait for replay timeout"). The sink makes
+    # those sends succeed instantly. It changes no DD_* env var, so config-observing tests
+    # (telemetry/config, read_c_configuration) and the RequestReplayer helper are unaffected.
+    # (make test_extension_ci copies ~hundreds of test files before any test runs, so the
+    # sink's <100ms bind is comfortably ready before the first trace is sent.)
+    - python3 .gitlab/scripts/localhost-trace-sink.py &
     - make test_extension_ci
 <?php after_script("tmp/build_extension", has_test_agent: true); ?>
 

.gitlab/scripts/localhost-trace-sink.py

@@ -0,0 +1,55 @@
+#!/usr/bin/env python3
+"""Always-200 trace sink bound to 127.0.0.1:8126 for test_extension_ci.
+
+Most of the extension test suite runs with the ddtrace default agent endpoint
+(DD_AGENT_HOST=localhost, DD_TRACE_AGENT_PORT=8126), which has no listener in the
+CI container -- the real services (request-replayer, test-agent) run as separate
+containers on their own hostnames. Those sidecar trace sends therefore get
+ECONNREFUSED and retry (max_retries=5, exponential backoff). Because the sidecar's
+trace flusher is shared across all sessions, those doomed sends stall the
+synchronous flush of unrelated tests past the 10s IPC read timeout, producing the
+flaky "Failed synchronously flushing traces: Kind(TimedOut)" /
+"wait for replay timeout" failures (e.g. client_side_stats).
+
+This sink answers every request instantly with 200 and an empty JSON body, so those
+default sends succeed immediately instead of retry-storming and never clog the shared
+flusher. An instant in-process responder (vs redirecting traffic at a real agent)
+cannot itself become the bottleneck.
+
+It deliberately changes no DD_* environment variable, so config-observing tests
+(telemetry/config, read_c_configuration) and the RequestReplayer test helper -- which
+read DD_AGENT_HOST / DD_TRACE_AGENT_PORT -- are unaffected.
+"""
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+BODY = b"{}"  # empty rate_by_service / info; enough for the sidecar to treat the flush as succeeded
+
+
+class Handler(BaseHTTPRequestHandler):
+    protocol_version = "HTTP/1.1"  # keep-alive so the sidecar can reuse connections
+
+    def _respond(self):
+        # Drain the request body so keep-alive connections stay in sync.
+        length = int(self.headers.get("Content-Length") or 0)
+        if length:
+            try:
+                self.rfile.read(length)
+            except Exception:
+                pass
+        try:
+            self.send_response(200)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(BODY)))
+            self.end_headers()
+            self.wfile.write(BODY)
+        except Exception:
+            pass
+
+    do_GET = do_POST = do_PUT = do_DELETE = do_PATCH = _respond
+
+    def log_message(self, format, *args):  # silence per-request logging
+        pass
+
+
+if __name__ == "__main__":
+    ThreadingHTTPServer(("127.0.0.1", 8126), Handler).serve_forever()