From 56726503c93db6149c19282c62fa8afc11e41d0a Mon Sep 17 00:00:00 2001
From: Zvi Grinberg <zgrinber@redhat.com>
Date: Thu, 9 Jul 2026 17:19:09 +0300
Subject: [PATCH] fix: switch liveness probe from httpGet to exec to prevent
 pod crashes under heavy CPU load

The Python event loop (GIL-bound) cannot serve HTTP /health responses when
the CPU is saturated by concurrent workloads, causing the httpGet liveness
probe to fail repeatedly and triggering pod restarts. The exec probe checks
that PID 1 is still alive via /proc/1/status, which runs outside the Python
process and is unaffected by event loop stalls.
---
 kustomize/base/exploit_iq_service.yaml | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)

diff --git a/kustomize/base/exploit_iq_service.yaml b/kustomize/base/exploit_iq_service.yaml
index 3b5759c94..df47b101a 100644
--- a/kustomize/base/exploit_iq_service.yaml
+++ b/kustomize/base/exploit_iq_service.yaml
@@ -66,18 +66,14 @@ spec:
               protocol: TCP
               containerPort: 8080
           livenessProbe:
-            failureThreshold: 24
-            timeoutSeconds: 10
-            periodSeconds: 20
-            httpGet:
-              path: /health
-              port: http
-# If still doing problems under heavy cpu loads and heavy concurrent workloads , kindly uncomment the tcpSocket block below,
-# and comment out the httpGet group as a temporary workaround.
-# Upgrading the agent to python version 3.14 will solve this issue as it will
-# be a real parallel multi-threading environment without GIL ( Global Interpreter Lock) restrictions.
-#           tcpSocket:
-#             port: 8080
+            failureThreshold: 3
+            timeoutSeconds: 5
+            periodSeconds: 30
+            exec:
+              command:
+                - /bin/sh
+                - -c
+                - cat /proc/1/status | grep -q '^State:.*[RSDs]'
 
           resources:
             limits: