Apache Mesos Mixin Modernization

apache-mesos-mixin/alerts/alerts.libsonnet → apache-mesos-mixin/alerts.libsonnet

@@ -1,5 +1,5 @@
 {
-  prometheusAlerts+:: {
+  new(this): {
     groups+: [
       {
         name: 'apache-mesos',
@@ -8,7 +8,7 @@
             alert: 'ApacheMesosHighMemoryUsage',
             expr: |||
               min without(instance, job, type) (mesos_master_mem{type="percent"}) > %(alertsWarningMemoryUsage)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'warning',
@@ -19,14 +19,14 @@
                 (
                   '{{ printf "%%.0f" $value }} percent memory usage on {{$labels.mesos_cluster}}, ' +
                   'which is above the threshold of %(alertsWarningMemoryUsage)s.'
-                ) % $._config,
+                ) % this.config,
             },
           },
           {
             alert: 'ApacheMesosHighDiskUsage',
             expr: |||
               min without(instance, job, type) (mesos_master_disk{type="percent"}) > %(alertsCriticalDiskUsage)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'critical',
@@ -37,14 +37,14 @@
                 (
                   '{{ printf "%%.0f" $value }} percent disk usage on {{$labels.mesos_cluster}}, ' +
                   'which is above the threshold of %(alertsCriticalDiskUsage)s.'
-                ) % $._config,
+                ) % this.config,
             },
           },
           {
             alert: 'ApacheMesosUnreachableTasks',
             expr: |||
               max without(instance, job, state) (mesos_master_task_states_current{state="unreachable"}) > %(alertsWarningUnreachableTask)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'warning',
@@ -55,14 +55,14 @@
                 (
                   '{{ printf "%%.0f" $value }} unreachable tasks on {{$labels.mesos_cluster}}, ' +
                   'which is above the threshold of %(alertsWarningUnreachableTask)s.'
-                ) % $._config,
+                ) % this.config,
             },
           },
           {
             alert: 'ApacheMesosNoLeaderElected',
             expr: |||
               max without(instance, job) (mesos_master_elected) == 0
-            ||| % $._config,
+            ||| % this.config,
             'for': '1m',
             labels: {
               severity: 'critical',
@@ -72,14 +72,14 @@
               description:
                 (
                   'There is no cluster coordinator on {{$labels.mesos_cluster}}.'
-                ) % $._config,
+                ) % this.config,
             },
           },
           {
             alert: 'ApacheMesosInactiveAgents',
             expr: |||
               max without(instance, job, state) (mesos_master_slaves_state{state=~"connected_inactive|disconnected_inactive"}) > 1
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'warning',
@@ -89,7 +89,7 @@
               description:
                 (
                   '{{ printf "%%.0f" $value }} inactive agent clients over the last 5m which is above the threshold of 1.'
-                ) % $._config,
+                ) % this.config,
             },
           },
         ],

apache-mesos-mixin/config.libsonnet

@@ -1,17 +1,36 @@
 {
-  _config+:: {
-    dashboardTags: ['apache-mesos-mixin'],
-    dashboardPeriod: 'now-1h',
-    dashboardTimezone: 'default',
-    dashboardRefresh: '1m',
 
-    // alerts thresholds
-    alertsWarningMemoryUsage: 90,
-    alertsCriticalDiskUsage: 90,
-    alertsWarningUnreachableTask: 3,
-    enableLokiLogs: true,
-    enableMultiCluster: false,
-    mesosSelector: if self.enableMultiCluster then 'job=~"$job", cluster=~"$cluster"' else 'job=~"$job"',
-    multiclusterSelector: 'job=~"$job"',
+  local this = self,
+  filteringSelector: 'job="integrations/apache-mesos"',
+  groupLabels: ['job', 'mesos_cluster', 'cluster'],
+  instanceLabels: ['instance'],
+
+  dashboardTags: [self.uid + '-mixin'],
+  uid: 'apache-mesos',
+  dashboardNamePrefix: 'Apache Mesos',
+  dashboardPeriod: 'now-1h',
+  dashboardTimezone: 'default',
+  dashboardRefresh: '1m',
+
+  // Logging configuration
+  enableLokiLogs: true,
+  logLabels: ['job', 'cluster', 'instance'],
+  extraLogLabels: ['level'],  // Required by logs-lib
+  logsVolumeGroupBy: 'level',
+  showLogsVolume: true,
+
+  // alerts thresholds
+  alertsWarningMemoryUsage: 90,
+  alertsCriticalDiskUsage: 90,
+  alertsWarningUnreachableTask: 3,
+
+  // metrics source for signals library
+  metricsSource: 'prometheus',
+
+  // signals configuration
+  signals+: {
+    overview: (import './signals/overview.libsonnet')(this),
+    master: (import './signals/master.libsonnet')(this),
+    agent: (import './signals/agent.libsonnet')(this),
   },
 }

apache-mesos-mixin/dashboards.libsonnet

@@ -0,0 +1,77 @@
+local g = import './g.libsonnet';
+local logslib = import 'logs-lib/logs/main.libsonnet';
+
+{
+  local root = self,
+  new(this)::
+    local links = this.grafana.links;
+    local tags = this.config.dashboardTags;
+    local uid = g.util.string.slugify(this.config.uid);
+    local vars = this.grafana.variables;
+    local annotations = this.grafana.annotations;
+    local prefix = this.config.dashboardNamePrefix;
+    local refresh = this.config.dashboardRefresh;
+    local period = this.config.dashboardPeriod;
+    local timezone = this.config.dashboardTimezone;
+
+
+    {
+
+      'apache-mesos-overview.json':
+        g.dashboard.new(this.config.dashboardNamePrefix + ' overview')
+        + g.dashboard.withPanels(
+          g.util.panel.resolveCollapsedFlagOnRows(
+            g.util.grid.wrapPanels([
+              this.grafana.rows.masterOverview,
+              this.grafana.rows.agentOverview,
+            ])
+          )
+        ) + root.applyCommon(
+          vars.multiInstance,
+          uid + '_overview',
+          tags,
+          links { apacheMesosOverview+:: {} },
+          annotations,
+          timezone,
+          refresh,
+          period,
+        ),
+    }
+    + if this.config.enableLokiLogs then {
+      'apache-mesos-logs.json':
+        logslib.new(
+          prefix + ' logs',
+          datasourceName=this.grafana.variables.datasources.loki.name,
+          datasourceRegex=this.grafana.variables.datasources.loki.regex,
+          filterSelector=this.config.filteringSelector,
+          labels=this.config.groupLabels + this.config.extraLogLabels,
+          formatParser=null,
+          showLogsVolume=this.config.showLogsVolume,
+        ) {
+          dashboards+: {
+            logs+:
+              root.applyCommon(super.logs.templating.list, uid=uid + '-logs', tags=tags, links=links { logs+:: {} }, annotations=annotations, timezone=timezone, refresh=refresh, period=period),
+          },
+          panels+: {
+            logs+:
+              g.panel.logs.options.withEnableLogDetails(true)
+              + g.panel.logs.options.withShowTime(false)
+              + g.panel.logs.options.withWrapLogMessage(false),
+          },
+          variables+: {
+            toArray+: [
+              this.grafana.variables.datasources.prometheus { hide: 2 },
+            ],
+          },
+        }.dashboards.logs,
+    } else {},
+
+  applyCommon(vars, uid, tags, links, annotations, timezone, refresh, period):
+    g.dashboard.withTags(tags)
+    + g.dashboard.withUid(uid)
+    + g.dashboard.withLinks(std.objectValues(links))
+    + g.dashboard.withTimezone(timezone)
+    + g.dashboard.withRefresh(refresh)
+    + g.dashboard.time.withFrom(period)
+    + g.dashboard.withVariables(vars),
+}