From 287c3f463987e5eee5c52079bf136a976523a48c Mon Sep 17 00:00:00 2001
From: Yadnesh Kulkarni <ykulkarn@redhat.com>
Date: Tue, 26 Nov 2024 06:28:07 -0500
Subject: [PATCH] Explicitly enable metrics needed for Kepler dashboard

- Disable estimated idle power metrics
- Enable container, vm and process metrics
- Timeout healthcheck curl command after 5 seconds
---
 .../files/healthchecks/exporter/healthcheck                | 3 ++-
 roles/edpm_telemetry_power_monitoring/tasks/install.yml    | 3 +--
 .../templates/kepler.json.j2                               | 7 ++++---
 3 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/roles/edpm_telemetry_power_monitoring/files/healthchecks/exporter/healthcheck b/roles/edpm_telemetry_power_monitoring/files/healthchecks/exporter/healthcheck
index 5f6baf851..53a669f60 100644
--- a/roles/edpm_telemetry_power_monitoring/files/healthchecks/exporter/healthcheck
+++ b/roles/edpm_telemetry_power_monitoring/files/healthchecks/exporter/healthcheck
@@ -16,9 +16,10 @@
 #    under the License.
 
 URL="http://0.0.0.0:8888/healthz"
+TIMEOUT=5 # Timeout in seconds
 
 # Get the HTTP status code and response body using curl
-RESPONSE=$(curl -s -w "%{http_code}" $URL)
+RESPONSE=$(curl -s -w "%{http_code}" $URL --max-time $TIMEOUT)
 BODY=${RESPONSE:0:-3}  # Extract the body (all but the last 3 characters)
 HTTP_CODE=${RESPONSE: -3}  # Extract the last 3 characters as the HTTP status code
 
diff --git a/roles/edpm_telemetry_power_monitoring/tasks/install.yml b/roles/edpm_telemetry_power_monitoring/tasks/install.yml
index 9d6aa547e..16dfc31ea 100644
--- a/roles/edpm_telemetry_power_monitoring/tasks/install.yml
+++ b/roles/edpm_telemetry_power_monitoring/tasks/install.yml
@@ -63,8 +63,7 @@
   become: true
   ansible.builtin.systemd:
     name: edpm_kepler.service
-    enabled: true
-    state: restarted
+    state: started
 
 - name: List deployed health check scripts
   ansible.builtin.find:
diff --git a/roles/edpm_telemetry_power_monitoring/templates/kepler.json.j2 b/roles/edpm_telemetry_power_monitoring/templates/kepler.json.j2
index 8c7e04e1e..3b839ab63 100644
--- a/roles/edpm_telemetry_power_monitoring/templates/kepler.json.j2
+++ b/roles/edpm_telemetry_power_monitoring/templates/kepler.json.j2
@@ -3,13 +3,14 @@
     "privileged": "true",
     "restart": "always",
     "ports": ["8888:8888"],
-    "command": "-v 2",
+    "command": "-v=2",
     "recreate": true,
     "environment": {
         "ENABLE_GPU": "true",
+        "EXPOSE_CONTAINER_METRICS": "true",
         "ENABLE_PROCESS_METRICS": "true",
-        "EXPOSE_ESTIMATED_IDLE_POWER_METRICS": "true",
         "EXPOSE_VM_METRICS": "true",
+        "EXPOSE_ESTIMATED_IDLE_POWER_METRICS": "false",
         "LIBVIRT_METADATA_URI": "http://openstack.org/xmlns/libvirt/nova/1.1"
     },
 {% if edpm_telemetry_power_monitoring_healthcheck %}
@@ -21,7 +22,7 @@
     "volumes": [
         "/lib/modules:/lib/modules:ro",
         "/run/libvirt:/run/libvirt:shared,ro",
-        "/sys/:/sys/",
+        "/sys:/sys",
         "/proc:/proc"
     ]
 }