From 845381800c3522cf1b702eb2459d5a7d4253b432 Mon Sep 17 00:00:00 2001
From: Kerri Wait <kerri.wait@monash.edu>
Date: Wed, 7 Oct 2020 14:55:01 +1100
Subject: [PATCH] Fix deployment of telegraf for monitoring mlx hw counters
---
roles/telegraf/tasks/main.yml | 17 +++++++++++++-
.../templates/inputs.multifile_mlx.conf.j2 | 22 +++++++++++++++++++
2 files changed, 38 insertions(+), 1 deletion(-)
create mode 100644 roles/telegraf/templates/inputs.multifile_mlx.conf.j2
diff --git a/roles/telegraf/tasks/main.yml b/roles/telegraf/tasks/main.yml
index 9226fa68..5ca8af1f 100644
--- a/roles/telegraf/tasks/main.yml
+++ b/roles/telegraf/tasks/main.yml
@@ -71,6 +71,20 @@
tags:
- configuration
+- name: Install multifile plugin for mlx hw_counters
+ template:
+ src: inputs.multifile_mlx.conf.j2
+ dest: /etc/telegraf/telegraf.d/inputs.multifile_mlx.conf
+ owner: telegraf
+ group: telegraf
+ mode: '640'
+ notify:
+ - "restart telegraf"
+ become: true
+ become_user: root
+ tags:
+ - configuration
+
- name: Install nvidia-smi plugin
template:
src: inputs.nvidia_smi.conf.j2
@@ -84,4 +98,5 @@
become_user: root
tags:
- configuration
- - gpu
\ No newline at end of file
+ - gpu
+ when: "'VisNodes' in group_names"
\ No newline at end of file
diff --git a/roles/telegraf/templates/inputs.multifile_mlx.conf.j2 b/roles/telegraf/templates/inputs.multifile_mlx.conf.j2
new file mode 100644
index 00000000..c6adbfd9
--- /dev/null
+++ b/roles/telegraf/templates/inputs.multifile_mlx.conf.j2
@@ -0,0 +1,22 @@
+# Read mlx hardware counters
+{% if hwcounterlist %}
+{% for interface in hwcounterlist %}
+[[inputs.multifile]]
+ name_override = 'infiniband'
+ base_dir = '/sys/class/infiniband'
+ interval = '60s'
+
+ [[inputs.multifile.tags]]
+ device = '{{ interface }}'
+ port = '1'
+ type = 'hw_counters'
+
+ {% for counter in hwcounterlist[interface] | sort %}
+[[inputs.multifile.file]]
+ file = '{{ interface }}/ports/1/hw_counters/{{ counter }}'
+ conversion = 'int'
+
+ {% endfor %}
+{% endfor %}
+{% endif %}
+
--
GitLab