Sync upstream latest PRs

grafana · Nov 1, 2024 · 8327300 · 8327300
1 parent 4c355b0
commit 8327300
Show file tree

Hide file tree

Showing 17 changed files with 276 additions and 18 deletions.
diff --git a/docs/node-mixin/lib/aix/README.md b/docs/node-mixin/lib/aix/README.md
@@ -0,0 +1,42 @@
+# AIX exporter observability lib
+
+This jsonnet observability lib can be used to generate observability package for node exporter(AIX).
+
+## Import
+
+```sh
+jb init
+jb install https://github.com/prometheus/node_exporter/docs/node-mixin/lib/aix
+```
+
+## Examples
+
+### Example 1: Basic example
+
+You can use observ-lib to fill in monitoring-mixin structure:
+
+```jsonnet
+// mixin.libsonnet file
+local aixlib = import 'aix/main.libsonnet';
+
+local aix =
+  aixlib.new()
+  + aixlib.withConfigMixin({
+    filteringSelector: 'job=~".*aix.*"',
+    groupLabels: ['job'],
+    instanceLabels: ['instance'],
+    dashboardNamePrefix: 'AIX / ',
+    dashboardTags: ['aix-mixin'],
+    uid: 'aix',
+    // enable loki logs
+    enableLokiLogs: true,
+  });
+
+{
+  grafanaDashboards+:: aix.grafana.dashboards,
+  prometheusAlerts+:: aix.prometheus.alerts,
+  prometheusRules+:: aix.prometheus.recordingRules,
+}
+
+```
+For more examples see [node-mixin/lib/linux](../linux).
diff --git a/docs/node-mixin/lib/aix/alerts.libsonnet b/docs/node-mixin/lib/aix/alerts.libsonnet
@@ -0,0 +1,23 @@
+{
+  new(this, parentPrometheus):
+    {
+      groups:
+        //keep only alerts listed in alertsKeep
+        std.filter(
+          function(group) std.length(group.rules) > 0,
+          [
+            {
+              name: group.name,
+              rules: [
+                rule
+                for rule in group.rules
+                if std.length(std.find(rule.alert, this.config.alertsKeep)) > 0
+              ],
+            }
+            for group in parentPrometheus.alerts.groups
+          ],
+
+        ),
+
+    },
+}
diff --git a/docs/node-mixin/lib/aix/config.libsonnet b/docs/node-mixin/lib/aix/config.libsonnet
@@ -0,0 +1,27 @@
+{
+  // Rest of the config is imported from linux
+  filteringSelector: 'job="aix"',
+  dashboardNamePrefix: 'MacOS / ',
+  //uid prefix
+  uid: 'aix',
+
+  dashboardTags: ['aix-mixin'],
+
+
+  // Alerts to keep from node-observ-lib:
+  alertsKeep: [
+    'NodeFilesystemAlmostOutOfSpace',
+    'NodeNetworkReceiveErrs',
+    'NodeNetworkTransmitErrs',
+    'NodeTextFileCollectorScrapeError',
+    'NodeFilesystemFilesFillingUp',
+    'NodeFilesystemAlmostOutOfFiles',
+    'NodeCPUHighUsage',
+    'NodeSystemSaturation',
+    'NodeMemoryHighUtilization',
+    'NodeDiskIOSaturation',
+  ],
+  // logs lib related
+  enableLokiLogs: false,
+
+}
diff --git a/docs/node-mixin/lib/aix/main.libsonnet b/docs/node-mixin/lib/aix/main.libsonnet
@@ -0,0 +1,65 @@
+local g = import '../g.libsonnet';
+local nodelib = import '../linux/main.libsonnet';
+local alerts = import './alerts.libsonnet';
+local config = import './config.libsonnet';
+local panels = import './panels.libsonnet';
+local targets = import './targets.libsonnet';
+
+
+// inherit nodelib
+nodelib
+{
+
+  new():
+    super.new()
+    + nodelib.withConfigMixin(config)
+    +
+    {
+      local this = self,
+      local parentGrafana = super.grafana,
+      local parentPrometheus = super.prometheus,
+
+      grafana+: {
+        // drop backToFleet link
+        links+: {
+          local link = g.dashboard.link,
+          backToFleet:: {},
+          backToOverview:
+            link.link.new('Back to ' + this.config.dashboardNamePrefix + 'overview', '/d/' + this.grafana.dashboards['nodes-darwin.json'].uid)
+            + link.link.options.withKeepTime(true),
+        },
+        annotations: {
+          // keep only reboot annotation
+          reboot: parentGrafana.annotations.reboot,
+        },
+        // override targets (memory)
+        targets+: targets.new(this),
+        // override panels (update description and targets in panels)
+        panels+: panels.new(this),
+
+        // keep only overview and logs(optionally) dashes
+        dashboards:
+          {
+            'nodes-aix.json':
+              parentGrafana.dashboards['nodes.json']
+              + g.dashboard.withUid(
+                (if this.config.uid == 'aix' then std.md5('nodes-aix.json') else this.config.uid + '-overview')
+              ),
+          }
+          +
+          (
+            if this.config.enableLokiLogs
+            then
+              {
+                'logs-aix.json': parentGrafana.dashboards['logs.json'],
+              }
+            else {}
+          ),
+      },
+      prometheus+: {
+        recordingRules: {},
+        alerts: alerts.new(this, parentPrometheus),
+      },
+    },
+
+}
diff --git a/docs/node-mixin/lib/aix/panels.libsonnet b/docs/node-mixin/lib/aix/panels.libsonnet
@@ -0,0 +1,24 @@
+local g = import '../g.libsonnet';
+local commonlib = import 'common-lib/common/main.libsonnet';
+
+{
+  new(this):
+    {
+      local t = this.grafana.targets,
+      local table = g.panel.table,
+      local fieldOverride = g.panel.table.fieldOverride,
+      local instanceLabel = this.config.instanceLabels[0],
+
+      // override description and targets
+      memory+: {
+        memoryUsageTsBytes+:
+          g.panel.timeSeries.queryOptions.withTargets([
+            t.memory.memoryUsedBytes,
+            t.memory.memoryTotalBytes,
+            t.memory.memorySwapUsedBytes,
+          ])
+          + commonlib.panels.generic.timeSeries.threshold.stylizeByRegexp('Physical memory'),
+      },
+
+    },
+}
diff --git a/docs/node-mixin/lib/aix/targets.libsonnet b/docs/node-mixin/lib/aix/targets.libsonnet
@@ -0,0 +1,61 @@
+local g = import '../g.libsonnet';
+local prometheusQuery = g.query.prometheus;
+local lokiQuery = g.query.loki;
+
+{
+  new(this): {
+    local variables = this.grafana.variables.main,
+    local config = this.config,
+    local prometheusDatasource = '${' + variables.datasources.prometheus.name + '}',
+    local lokiDatasource = '${' + variables.datasources.loki.name + '}',
+
+    // override memory targets (other metrics in macos)
+    memory+: {
+      memoryTotalBytes:
+        prometheusQuery.new(
+          prometheusDatasource,
+          'node_memory_total_bytes{%(queriesSelector)s}' % variables
+        )
+        + prometheusQuery.withLegendFormat('Physical memory'),
+
+      memoryUsedBytes:
+        prometheusQuery.new(
+          prometheusDatasource,
+          |||
+            (
+              node_memory_total_bytes{%(queriesSelector)s} -
+              node_memory_available_bytes{%(queriesSelector)s}
+            )
+          ||| % variables
+        )
+        + prometheusQuery.withLegendFormat('Memory used'),
+
+      memoryUsagePercent:
+        prometheusQuery.new(
+          prometheusDatasource,
+          |||
+            (
+              (
+                node_memory_total_bytes{%(queriesSelector)s} -
+                node_memory_available_bytes{%(queriesSelector)s}
+              ) 
+              /avg(node_memory_total_bytes{%(queriesSelector)s})
+            ) * 100
+          |||
+          % variables,
+        ),
+      memorySwapTotal:
+        prometheusQuery.new(
+          prometheusDatasource,
+          'node_memory_swap_total_bytes{%(queriesSelector)s}' % variables
+        ),
+
+      memorySwapUsedBytes:
+        prometheusQuery.new(
+          prometheusDatasource,
+          'node_memory_swap_used_bytes{%(queriesSelector)s}' % variables
+        )
+        + prometheusQuery.withLegendFormat('Swap used'),
+    },
+  },
+}
diff --git a/docs/node-mixin/lib/linux/alerts/alerts.libsonnet b/docs/node-mixin/lib/linux/alerts/alerts.libsonnet
@@ -395,7 +395,7 @@
               annotations: {
                 summary: 'Disk IO queue is high.',
                 description: |||
-                  Disk IO queue (aqu-sq) is high on {{ $labels.device }} at {{ $labels.instance }}, has been above %(diskIOSaturationThreshold)d for the last 15 minutes, is currently at {{ printf "%%.2f" $value }}.
+                  Disk IO queue (aqu-sq) is high on {{ $labels.device }} at {{ $labels.instance }}, has been above %(diskIOSaturationThreshold)d for the last 30 minutes, is currently at {{ printf "%%.2f" $value }}.
                   This symptom might indicate disk saturation.
                 ||| % this.config,
               },

diff --git a/docs/node-mixin/lib/linux/config.libsonnet b/docs/node-mixin/lib/linux/config.libsonnet
@@ -10,6 +10,7 @@
   groupLabels: ['job'],
   instanceLabels: ['instance'],
   dashboardNamePrefix: 'Node exporter / ',
+  //uid prefix
   uid: 'node',
   dashboardTags: ['node-exporter-mixin'],
 
@@ -76,7 +77,7 @@
   fsSpaceAvailableWarningThreshold: 5,
   fsSpaceAvailableCriticalThreshold: 3,
 
-  // Memory utilzation (%) level on which to trigger the
+  // Memory utilization (%) level on which to trigger the
   // 'NodeMemoryHighUtilization' alert.
   memoryHighUtilizationThreshold: 90,
 

diff --git a/docs/node-mixin/lib/linux/dashboards.libsonnet b/docs/node-mixin/lib/linux/dashboards.libsonnet
@@ -24,7 +24,7 @@ local logslib = import 'github.com/grafana/jsonnet-libs/logs-lib/logs/main.libso
         )
         // hide link to self
         + root.applyCommon(vars.multiInstance, uid + '-fleet', tags, links { backToFleet+:: {}, backToOverview+:: {} }, annotations, timezone, refresh, period),
-      'overview.json':
+      'nodes.json':
         g.dashboard.new(prefix + 'overview')
         + g.dashboard.withPanels(
           g.util.panel.resolveCollapsedFlagOnRows(
@@ -46,7 +46,7 @@ local logslib = import 'github.com/grafana/jsonnet-libs/logs-lib/logs/main.libso
           )
         )
         // defaults to uid=nodes for backward compatibility with old node-mixins
-        + root.applyCommon(vars.singleInstance, (if uid == 'node' then 'nodes' else uid + '-overview'), tags, links { backToOverview+:: {} }, annotations, timezone, refresh, period),
+        + root.applyCommon(vars.singleInstance, (if uid == 'node' then std.md5('nodes.json') else uid + '-overview'), tags, links { backToOverview+:: {} }, annotations, timezone, refresh, period),
       'network.json':
         g.dashboard.new(prefix + 'network')
         + g.dashboard.withPanels(
@@ -120,7 +120,7 @@ local logslib = import 'github.com/grafana/jsonnet-libs/logs-lib/logs/main.libso
             )
           )
         )
-        + root.applyCommon(this.grafana.variables.use.singleInstance, uid + '-rsrc-use', tags, links, annotations, timezone, refresh, period),
+        + root.applyCommon(this.grafana.variables.use.singleInstance, std.md5(uid + '-cluster-rsrc-use.json'), tags, links, annotations, timezone, refresh, period),
 
       'node-cluster-rsrc-use.json':
         g.dashboard.new(prefix + 'USE method / cluster')
@@ -137,15 +137,15 @@ local logslib = import 'github.com/grafana/jsonnet-libs/logs-lib/logs/main.libso
             )
           )
         )
-        + root.applyCommon(this.grafana.variables.useCluster.singleInstance, uid + '-cluster-rsrc-use', tags, links, annotations, timezone, refresh, period),
+        + root.applyCommon(this.grafana.variables.useCluster.singleInstance, std.md5(uid + '-cluster-rsrc-use.json'), tags, links, annotations, timezone, refresh, period),
     }
     +
     (
       if this.config.showMultiCluster
       then
         {
           'node-multicluster-rsrc-use.json':
-            g.dashboard.new(prefix + 'USE method / cluster')
+            g.dashboard.new(prefix + 'USE method / Multi-cluster')
             + g.dashboard.withPanels(
               g.util.panel.resolveCollapsedFlagOnRows(
                 g.util.grid.wrapPanels(
@@ -159,7 +159,7 @@ local logslib = import 'github.com/grafana/jsonnet-libs/logs-lib/logs/main.libso
                 )
               )
             )
-            + root.applyCommon(this.grafana.variables.useCluster.multiInstance, uid + '-multicluster-rsrc-use', tags, links, annotations, timezone, refresh, period),
+            + root.applyCommon(this.grafana.variables.useCluster.multiInstance, std.md5(uid + '-multicluster-rsrc-use.json'), tags, links, annotations, timezone, refresh, period),
         }
       else {}
     )

diff --git a/docs/node-mixin/lib/linux/links.libsonnet b/docs/node-mixin/lib/linux/links.libsonnet
@@ -8,7 +8,7 @@ local commonlib = import 'common-lib/common/main.libsonnet';
         link.link.new('Back to ' + this.config.dashboardNamePrefix + 'fleet', '/d/' + this.grafana.dashboards['fleet.json'].uid)
         + link.link.options.withKeepTime(true),
       backToOverview:
-        link.link.new('Back to ' + this.config.dashboardNamePrefix + 'overview', '/d/' + this.grafana.dashboards['overview.json'].uid)
+        link.link.new('Back to ' + this.config.dashboardNamePrefix + 'overview', '/d/' + this.grafana.dashboards['nodes.json'].uid)
         + link.link.options.withKeepTime(true),
       otherDashboards:
         link.dashboards.new('All ' + this.config.dashboardNamePrefix + ' dashboards', this.config.dashboardTags)

diff --git a/docs/node-mixin/lib/linux/panels/cpu.libsonnet b/docs/node-mixin/lib/linux/panels/cpu.libsonnet
@@ -19,7 +19,7 @@ local xtd = import 'github.com/jsonnet-libs/xtd/main.libsonnet';
         target=t.cpu.cpuUsage,
         topk=25,
         instanceLabels=this.config.instanceLabels,
-        drillDownDashboardUid=this.grafana.dashboards['overview.json'].uid,
+        drillDownDashboardUid=this.grafana.dashboards['nodes.json'].uid,
       ),
       cpuUsageStat: commonlib.panels.cpu.stat.usage.new(targets=[t.cpu.cpuUsage]),
       cpuUsageByMode: commonlib.panels.cpu.timeSeries.utilizationByMode.new(

diff --git a/docs/node-mixin/lib/linux/panels/disk.libsonnet b/docs/node-mixin/lib/linux/panels/disk.libsonnet
@@ -83,7 +83,7 @@ local xtd = import 'github.com/jsonnet-libs/xtd/main.libsonnet';
         target=t.disk.diskUsagePercent,
         topk=25,
         instanceLabels=this.config.instanceLabels + ['volume'],
-        drillDownDashboardUid=this.grafana.dashboards['overview.json'].uid,
+        drillDownDashboardUid=this.grafana.dashboards['nodes.json'].uid,
       ),
       diskIOBytesPerSec: commonlib.panels.disk.timeSeries.ioBytesPerSec.new(
         targets=[t.disk.diskIOreadBytesPerSec, t.disk.diskIOwriteBytesPerSec, t.disk.diskIOutilization]
@@ -94,7 +94,7 @@ local xtd = import 'github.com/jsonnet-libs/xtd/main.libsonnet';
           target=t.disk.diskIOutilization,
           topk=25,
           instanceLabels=this.config.instanceLabels + ['volume'],
-          drillDownDashboardUid=this.grafana.dashboards['overview.json'].uid,
+          drillDownDashboardUid=this.grafana.dashboards['nodes.json'].uid,
         ),
       diskIOps:
         commonlib.panels.disk.timeSeries.iops.new(

diff --git a/docs/node-mixin/lib/linux/panels/fleet.libsonnet b/docs/node-mixin/lib/linux/panels/fleet.libsonnet
@@ -83,7 +83,7 @@ local xtd = import 'github.com/jsonnet-libs/xtd/main.libsonnet';
             {
               targetBlank: false,
               title: 'Drill down to ${__field.name} ${__value.text}',
-              url: 'd/%s?var-%s=${__data.fields.%s}&${__url_time_range}&${datasource:queryparam}' % [this.grafana.dashboards['overview.json'].uid, instanceLabel, instanceLabel],
+              url: 'd/%s?var-%s=${__data.fields.%s}&${__url_time_range}&${datasource:queryparam}' % [this.grafana.dashboards['nodes.json'].uid, instanceLabel, instanceLabel],
             },
           ]),
           fieldOverride.byRegexp.new(std.join('|', std.map(utils.toSentenceCase, this.config.groupLabels)))

diff --git a/docs/node-mixin/lib/linux/panels/memory.libsonnet b/docs/node-mixin/lib/linux/panels/memory.libsonnet
@@ -30,7 +30,7 @@ local xtd = import 'github.com/jsonnet-libs/xtd/main.libsonnet';
         target=t.memory.memoryUsagePercent,
         topk=25,
         instanceLabels=this.config.instanceLabels,
-        drillDownDashboardUid=this.grafana.dashboards['overview.json'].uid,
+        drillDownDashboardUid=this.grafana.dashboards['nodes.json'].uid,
       ),
       memoryUsageTsBytes:
         commonlib.panels.memory.timeSeries.usageBytes.new(

diff --git a/docs/node-mixin/lib/macos/config.libsonnet b/docs/node-mixin/lib/macos/config.libsonnet
@@ -1,10 +1,11 @@
 {
   // Rest of the config is imported from linux
-  filteringSelector: 'job="integrations/macos-node"',
+  filteringSelector: 'job=macos"',
   dashboardNamePrefix: 'MacOS / ',
+  //uid prefix
   uid: 'darwin',
 
-  dashboardTags: [self.uid],
+  dashboardTags: ['macos-mixin'],
 
 
   // Alerts to keep from node-observ-lib: