From 33d9e1c453795035d517fb887da1060571a78722 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Pawe=C5=82=20Chojnacki?= Date: Mon, 20 Apr 2026 20:38:00 +0200 Subject: [PATCH] Rewrite readiness-probe comment to be bug-agnostic The previous comment explained the TCP-over-HTTP choice by citing a specific bug in the DatadogTracing Off scope-merge path. That bug is now fixed, so the comment reads as if a live bug is still in the tree. Restate the rationale in terms that will still be true after further mod_datadog changes: any HTTP probe runs through the module's handlers and can emit a trace into the per-test agent session; a bare TCP connect can't. --- test/integration-test/conftest.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/test/integration-test/conftest.py b/test/integration-test/conftest.py index 4390031..f7fea71 100644 --- a/test/integration-test/conftest.py +++ b/test/integration-test/conftest.py @@ -94,11 +94,11 @@ def load_configuration(self, conf_path: str) -> bool: # Continue anyway - sometimes apachectl returns non-zero but Apache starts # We'll verify below with HTTP requests - # Probe readiness with a bare TCP connect rather than an HTTP GET — - # mod_datadog's `DatadogTracing Off` inside a is silently - # dropped by the config merge in common_conf.cpp (child Off is ORed - # with parent On), so any HTTP probe would emit a trace and pollute - # the per-test agent session. + # Probe readiness with a bare TCP connect rather than an HTTP GET: + # HTTP probes run through mod_datadog's handlers and can emit a + # trace into the per-test agent session, contaminating assertions + # tests make about per-request trace counts. A raw TCP connect + # doesn't reach any request handler, so it stays out of traces. import socket max_wait = 5 start_time = time.time()