MSSQL Modernization

mssql-mixin/alerts/alerts.libsonnet → mssql-mixin/alerts.libsonnet

@@ -1,30 +1,30 @@
 {
-  prometheusAlerts+:: {
-    groups+: [
+  new(this): {
+    groups: [
       {
         name: 'MSSQLAlerts',
         rules: [
           {
             alert: 'MSSQLHighNumberOfDeadlocks',
             expr: |||
               increase(mssql_deadlocks_total{}[5m]) > %(alertsWarningDeadlocks5m)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'warning',
             },
             annotations: {
-              summary: 'There are deadlocks ocurring in the database.',
+              summary: 'There are deadlocks occurring in the database.',
               description:
                 ('{{ printf "%%.2f" $value }} deadlocks have occurred over the last 5 minutes on {{$labels.instance}}, ' +
-                 'which is above threshold of %(alertsWarningDeadlocks5m)s deadlocks.') % $._config,
+                 'which is above threshold of %(alertsWarningDeadlocks5m)s deadlocks.') % this.config,
             },
           },
           {
             alert: 'MSSQLModerateReadStallTime',
             expr: |||
               1000 * increase(mssql_io_stall_seconds_total{operation="read"}[5m]) > %(alertsWarningModerateReadStallTimeMS)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'warning',
@@ -33,14 +33,14 @@
               summary: 'There is a moderate amount of IO stall for database reads.',
               description:
                 ('{{ printf "%%.2f" $value }}ms of IO read stall has occurred on {{$labels.instance}}, ' +
-                 'which is above threshold of %(alertsWarningModerateReadStallTimeMS)sms.') % $._config,
+                 'which is above threshold of %(alertsWarningModerateReadStallTimeMS)sms.') % this.config,
             },
           },
           {
             alert: 'MSSQLHighReadStallTime',
             expr: |||
               1000 * increase(mssql_io_stall_seconds_total{operation="read"}[5m]) > %(alertsCriticalHighReadStallTimeMS)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'critical',
@@ -49,14 +49,14 @@
               summary: 'There is a high amount of IO stall for database reads.',
               description:
                 ('{{ printf "%%.2f" $value }}ms of IO read stall has occurred on {{$labels.instance}}, ' +
-                 'which is above threshold of %(alertsCriticalHighReadStallTimeMS)sms.') % $._config,
+                 'which is above threshold of %(alertsCriticalHighReadStallTimeMS)sms.') % this.config,
             },
           },
           {
             alert: 'MSSQLModerateWriteStallTime',
             expr: |||
               1000 * increase(mssql_io_stall_seconds_total{operation="write"}[5m]) > %(alertsWarningModerateWriteStallTimeMS)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'warning',
@@ -65,14 +65,14 @@
               summary: 'There is a moderate amount of IO stall for database writes.',
               description:
                 ('{{ printf "%%.2f" $value }}ms of IO write stall has occurred on {{$labels.instance}}, ' +
-                 'which is above threshold of %(alertsWarningModerateWriteStallTimeMS)sms.') % $._config,
+                 'which is above threshold of %(alertsWarningModerateWriteStallTimeMS)sms.') % this.config,
             },
           },
           {
             alert: 'MSSQLHighWriteStallTime',
             expr: |||
               1000 * increase(mssql_io_stall_seconds_total{operation="write"}[5m]) > %(alertsCriticalHighWriteStallTimeMS)s
-            ||| % $._config,
+            ||| % this.config,
             'for': '5m',
             labels: {
               severity: 'critical',
@@ -81,7 +81,7 @@
               summary: 'There is a high amount of IO stall for database writes.',
               description:
                 ('{{ printf "%%.2f" $value }}ms of IO write stall has occurred on {{$labels.instance}}, ' +
-                 'which is above threshold of %(alertsCriticalHighWriteStallTimeMS)sms.') % $._config,
+                 'which is above threshold of %(alertsCriticalHighWriteStallTimeMS)sms.') % this.config,
             },
           },
         ],

mssql-mixin/config.libsonnet

@@ -1,21 +1,29 @@
 {
-  _config+:: {
-    enableMultiCluster: false,
-    multiclusterSelector: 'job=~"$job"',
-    mssqlSelector: if self.enableMultiCluster then 'job=~"$job", cluster=~"$cluster"' else 'job=~"$job"',
-    dashboardTags: ['mssql-mixin'],
-    dashboardPeriod: 'now-1h',
-    dashboardTimezone: 'default',
-    dashboardRefresh: '1m',
+  enableMultiCluster: false,
+  filteringSelector: 'job=~"integrations/mssql"',
+  groupLabels: if self.enableMultiCluster then ['job', 'cluster'] else ['job'],
+  instanceLabels: ['instance'],
+  dashboardTags: ['mssql-mixin'],
+  legendLabels: ['instance'],
+  uid: 'mssql',
+  dashboardNamePrefix: 'MSSQL',
 
-    // alert thresholds
-    alertsWarningDeadlocks5m: 10,
-    alertsWarningModerateReadStallTimeMS: 200,
-    alertsCriticalHighReadStallTimeMS: 400,
-    alertsWarningModerateWriteStallTimeMS: 200,
-    alertsCriticalHighWriteStallTimeMS: 400,
+  // additional params
+  dashboardPeriod: 'now-1h',
+  dashboardTimezone: 'default',
+  dashboardRefresh: '1m',
 
-    // enable Loki logs
-    enableLokiLogs: true,
-  },
+  // logs lib related
+  enableLokiLogs: true,
+  logLabels: if self.enableMultiCluster then ['job', 'instance', 'cluster', 'level'] else ['job', 'instance', 'level'],
+  extraLogLabels: [],  // Required by logs-lib
+  logsVolumeGroupBy: 'level',
+  showLogsVolume: true,
+
+  // alert thresholds
+  alertsWarningDeadlocks5m: 10,
+  alertsWarningModerateReadStallTimeMS: 200,
+  alertsCriticalHighReadStallTimeMS: 400,
+  alertsWarningModerateWriteStallTimeMS: 200,
+  alertsCriticalHighWriteStallTimeMS: 400,
 }

mssql-mixin/dashboards.libsonnet

@@ -0,0 +1,115 @@
+local g = import './g.libsonnet';
+local logslib = import 'logs-lib/logs/main.libsonnet';
+{
+  local root = self,
+  new(this)::
+    local prefix = this.config.dashboardNamePrefix;
+    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 refresh = this.config.dashboardRefresh;
+    local period = this.config.dashboardPeriod;
+    local timezone = this.config.dashboardTimezone;
+    local panels = this.grafana.panels;
+
+    {
+      mssql_overview:
+        g.dashboard.new(prefix + ' overview')
+        + g.dashboard.withPanels(
+          g.util.panel.resolveCollapsedFlagOnRows(
+            g.util.grid.wrapPanels(
+              [
+                panels.connectionsPanel { gridPos+: { w: 12 } },
+                panels.batchRequestsPanel { gridPos+: { w: 12 } },
+                panels.severeErrorsPanel { gridPos+: { w: 12 } },
+                panels.deadlocksPanel { gridPos+: { w: 12 } },
+                panels.osMemoryUsagePanel { gridPos+: { w: 24 } },
+                panels.memoryManagerPanel { gridPos+: { w: 16 } },
+                panels.committedMemoryUtilizationPanel { gridPos+: { w: 8 } },
+              ] +
+              [this.grafana.rows.database + g.panel.row.withCollapsed(false)],
+            ),
+          )
+        )
+        + root.applyCommon(
+          vars.singleInstance,
+          uid + '_mssql_overview',
+          tags,
+          links { mssqlOverview+:: {} },
+          annotations,
+          timezone,
+          refresh,
+          period
+        ),
+
+      mssql_pages:
+        g.dashboard.new(prefix + ' pages')
+        + g.dashboard.withPanels(
+          g.util.grid.wrapPanels(
+            [
+              panels.pageFileMemoryPanel { gridPos+: { w: 12 } },
+              panels.bufferCacheHitPercentagePanel { gridPos+: { w: 12 } },
+              panels.pageCheckpointsPanel { gridPos+: { w: 12 } },
+              panels.pageFaultsPanel { gridPos+: { w: 12 } },
+            ]
+          )
+        )
+        + root.applyCommon(
+          vars.singleInstance,
+          uid + '_mssql_pages',
+          tags,
+          links { mssqlPages+:: {} },
+          annotations,
+          timezone,
+          refresh,
+          period
+        ),
+
+    }
+    +
+    if this.config.enableLokiLogs then
+      {
+        logs:
+          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)
+    + g.dashboard.withAnnotations(std.objectValues(annotations)),
+}