add 2 new health statuses, fix gpu panels to avg by host and gpu, remove jb, jsonnet files and download them ondemand

Author: sam-andaluri

playbooks/roles/grafana/files/jb

@@ -86,6 +86,8 @@ local health_status = [
 { expr1: 'rdma_device_status{hostname=~"$hostname", oci_name=~"$oci_name"}==0', expr2: 'rdma_device_status{hostname=~"$hostname", oci_name=~"$oci_name"}==1', legend_format: '{{hostname}}:{{rdma_device}}', title: 'RDMA Device Status', unit: 'none' },
 { expr1: 'rdma_link_noflap{hostname=~"$hostname", oci_name=~"$oci_name"}==0', expr2: 'rdma_link_noflap{hostname=~"$hostname", oci_name=~"$oci_name"}==1', legend_format: '{{hostname}}:{{rdma_device}}', title: 'RDMA Link flapping', unit: 'none' },
 { expr1: 'rttcc_status{hostname=~"$hostname", oci_name=~"$oci_name"}==0', expr2: 'rttcc_status{hostname=~"$hostname", oci_name=~"$oci_name"}==1', legend_format: '{{hostname}}:{{rdma_device}}', title: 'RTTCC Status', unit: 'none' },
+{ expr1: 'gpu_count{hostname=~"$hostname", oci_name=~"$oci_name"}==0', expr2: 'gpu_count{hostname=~"$hostname", oci_name=~"$oci_name"}==1', legend_format: '{{hostname}}:{{instance_shape}}', title: 'GPU Status', unit: 'none' },
+{ expr1: 'oca_version{hostname=~"$hostname", oci_name=~"$oci_name"}==0', expr2: 'oca_version{hostname=~"$hostname", oci_name=~"$oci_name"}==1', legend_format: '{{hostname}}:{{version}}', title: 'OCA Version', unit: 'none' },
 ];
 
 local nfs_metrics = [
@@ -211,9 +213,9 @@ g.dashboard.new('Cluster Dashboard')
         + g.panel.timeSeries.queryOptions.withTargets([
             g.query.prometheus.new(
                 '$PROMETHEUS_DS',
-                'avg by(Hostname) (' + metric.name + '{Hostname=~"$hostname", oci_name=~"$oci_name"})',
+                'avg by(Hostname, gpu) (' + metric.name + '{Hostname=~"$hostname", oci_name=~"$oci_name"})',
             )
-            + g.query.prometheus.withLegendFormat('{{ Hostname }}')
+            + g.query.prometheus.withLegendFormat('{{ Hostname }}:{{ gpu }}')
         ])
         + g.panel.timeSeries.standardOptions.withUnit(metric.unit)
         + g.panel.timeSeries.gridPos.withW(24)
@@ -227,9 +229,9 @@ g.dashboard.new('Cluster Dashboard')
         + g.panel.timeSeries.queryOptions.withTargets([
             g.query.prometheus.new(
                 '$PROMETHEUS_DS',
-                'avg by(Hostname) (' + metric.name + '{Hostname=~"$hostname", oci_name=~"$oci_name"})',
+                'avg by(Hostname, gpu) (' + metric.name + '{Hostname=~"$hostname", oci_name=~"$oci_name"})',
             )
-            + g.query.prometheus.withLegendFormat('{{ Hostname }}')
+            + g.query.prometheus.withLegendFormat('{{ Hostname }}:{{ gpu }}')
         ])
         + g.panel.timeSeries.standardOptions.withUnit(metric.unit)
         + g.panel.timeSeries.gridPos.withW(24)

playbooks/roles/grafana/files/jsonnet

@@ -1,15 +1,24 @@
 ---
-- name: Copy binaries
-  copy:
-    src: "{{ item }}"
-    dest: /usr/local/bin
-    owner: "{{ user }}"
-    group: "{{ user }}"
+- name: Download jb
+  become: true
+  get_url:
+    url: https://github.com/jsonnet-bundler/jsonnet-bundler/releases/download/v0.6.0/jb-linux-amd64
+    dest: /usr/local/bin/jb
     mode: '0755'
-  loop:
-    - jb
-    - jsonnet
+
+- name: Download jsonnet
+  become: true
+  get_url:
+    url: https://github.com/google/go-jsonnet/releases/download/v0.20.0/go-jsonnet_0.20.0_Linux_x86_64.tar.gz
+    dest: /tmp
+
+- name: Extract jsonnet
   become: true
+  unarchive:
+    src: /tmp/go-jsonnet_0.20.0_Linux_x86_64.tar.gz
+    dest: /usr/local/bin
+    mode: '0755'
+    
 
 - name: Delete existing build directory
   file: