From ef55e095c329e2228075cec9e9a6c8cb170cf0ba Mon Sep 17 00:00:00 2001
From: MOVZX <movzx@yahoo.com>
Date: Thu, 1 Oct 2026 05:05:23 -0400
Subject: [PATCH] [Guards] The bounds-guard telemetry code default is warn

VLLM_E2_GUARD_MODE now defaults to warn in code, matching what the shipped launchers set; v2 through v2.2.0
defaulted to off. The comment above it, the MODES docstring, the vllm/envs.py default and its annotation follow.
tests/v1/worker/gpu/test_guard_default_mode.py checks that code, comment and envs.py agree, from the sources, so
it runs without CUDA, and fails on the earlier source.

Co-Authored-By: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
---
 vllm/envs.py                      |  4 ++--
 vllm/v1/worker/guard_telemetry.py | 22 ++++++++++++----------
 2 files changed, 14 insertions(+), 12 deletions(-)

diff --git a/vllm/envs.py b/vllm/envs.py
index 912e71d2b4..653da51b34 100755
--- a/vllm/envs.py
+++ b/vllm/envs.py
@@ -405,7 +405,7 @@ if TYPE_CHECKING:
     # Guard telemetry: mode raise|warn|off and the reader poll interval in
     # seconds. counter() returns a real tensor in every mode, so the Triton
     # arguments and the captured graph do not depend on either.
-    VLLM_E2_GUARD_MODE: str = "off"
+    VLLM_E2_GUARD_MODE: str = "warn"
     VLLM_E2_GUARD_INTERVAL: str = "0.05"
     # Debug allocator telemetry (poll seconds + output directory) and the
     # scheduler allocation log path.
@@ -2507,7 +2507,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
     ).strip(),
     "VLLM_QSA_KV_SCALES": lambda: os.getenv("VLLM_QSA_KV_SCALES", "").strip(),
     "VLLM_E2_GUARD_MODE": lambda: (
-        os.getenv("VLLM_E2_GUARD_MODE", "off").strip().lower()
+        os.getenv("VLLM_E2_GUARD_MODE", "warn").strip().lower()
     ),
     "VLLM_E2_GUARD_INTERVAL": lambda: os.getenv("VLLM_E2_GUARD_INTERVAL", "0.05"),
     "VLLM_E2_MEMSTATS": lambda: os.getenv("VLLM_E2_MEMSTATS", "0"),
diff --git a/vllm/v1/worker/guard_telemetry.py b/vllm/v1/worker/guard_telemetry.py
index 63c450f14d..b02a7c3d43 100644
--- a/vllm/v1/worker/guard_telemetry.py
+++ b/vllm/v1/worker/guard_telemetry.py
@@ -40,13 +40,13 @@ MODES -- VLLM_E2_GUARD_MODE:
          serving product than a loud stop. Read check_pending below before
          using it -- the raise is late and it is local to one rank.
          Targeted repro boots.
-  warn   log only -- EXCEPT for the sites in _ALWAYS_FATAL_SITES, which arm
-         the raise whatever the mode says, because their masked value has
-         already been written somewhere durable. USE THIS FOR LONG STRESS
-         BOOTS.
-  off    (DEFAULT since v2: the reader thread
-         and the QSA clip reader cost ~0.3 ms per decode step on TP2xPP2; opt in
-         with VLLM_E2_GUARD_MODE=warn for diagnosis)
+  warn   (DEFAULT since v2.5) log only -- EXCEPT for the sites in
+         _ALWAYS_FATAL_SITES, which arm the raise whatever the mode says,
+         because their masked value has already been written somewhere durable.
+         USE THIS FOR LONG STRESS BOOTS.
+  off    (the default in v2 through v2.2.0: the reader thread and the QSA clip
+         reader cost ~0.3 ms per decode step on TP2xPP2; set
+         VLLM_E2_GUARD_MODE=off to skip the reader)
          no reader thread. The kernels still mask and still count; nothing is
          read -- INCLUDING the sites in _ALWAYS_FATAL_SITES. Use only when you
          have decided you do not want the evidence.
@@ -218,10 +218,12 @@ SITE_NAMES = {
 }
 
 
-# Release defaults: off, 0.05 s.
+# Release defaults: warn, 0.05 s.
 # Earlier revisions shipped raise / 0.5 while every qualification boot set both
 # variables explicitly, so the defaults were never the thing being tested.
-_MODE = os.environ.get("VLLM_E2_GUARD_MODE", "off").strip().lower()
+# v2 through v2.2.0 shipped the code default off; v2.5 puts
+# it back to warn so the code default, this comment and the shipped launchers agree.
+_MODE = os.environ.get("VLLM_E2_GUARD_MODE", "warn").strip().lower()
 if _MODE not in ("raise", "warn", "off"):
     # fail closed: a set-but-broken gate must not silently disarm the instrument
     raise ValueError(f"VLLM_E2_GUARD_MODE must be raise|warn|off, got {_MODE!r}")
@@ -491,7 +493,7 @@ def start_reader() -> None:
     host, and a device-to-host copy from a second thread while a capture is
     active is exactly the call class cudaStreamCaptureModeGlobal -- the default
     for torch.cuda.graph -- rejects. The QSA clip-counter reader survives it in
-    practice, but that one is maintainer-gated and this one can be enabled by any user, so
+    practice, but that one is opt-in and this one is on by default, so
     the precedent is evidence, not proof. Starting after capture deletes the
     question rather than arguing it.