Merge branch 'master' into pgbench

cloud-bulldozer · Feb 13, 2024 · 439d3cc · 439d3cc
2 parents c84f140 + 45abdfd
commit 439d3cc
Show file tree

Hide file tree

Showing 36 changed files with 3,268 additions and 11 deletions.
diff --git a/assets/api-performance-overview/panels.libsonnet b/assets/api-performance-overview/panels.libsonnet
@@ -0,0 +1,63 @@
+local g = import 'github.com/grafana/grafonnet/gen/grafonnet-latest/main.libsonnet';
+
+{
+    timeSeries: {
+       local timeSeries = g.panel.timeSeries,
+        local custom = timeSeries.fieldConfig.defaults.custom,
+        local options = timeSeries.options,
+
+        base(title, unit, targets, gridPos):
+            timeSeries.new(title)
+            + timeSeries.queryOptions.withTargets(targets)
+            + timeSeries.datasource.withType('prometheus')
+            + timeSeries.datasource.withUid('$Datasource')
+            + timeSeries.standardOptions.withUnit(unit)
+            + timeSeries.gridPos.withX(gridPos.x)
+            + timeSeries.gridPos.withY(gridPos.y)
+            + timeSeries.gridPos.withH(gridPos.h)
+            + timeSeries.gridPos.withW(gridPos.w)
+            + custom.withDrawStyle("line")
+            + custom.withLineInterpolation("linear")
+            + custom.withBarAlignment(0)
+            + custom.withLineWidth(1)
+            + custom.withFillOpacity(10)
+            + custom.withGradientMode("none")
+            + custom.withSpanNulls(false)
+            + custom.withPointSize(5)
+            + custom.withSpanNulls(false)
+            + custom.stacking.withGroup("A")
+            + custom.stacking.withMode("none")
+            + custom.withShowPoints('never')
+            + options.tooltip.withSort('desc')
+            + timeSeries.queryOptions.withTimeFrom(null)
+            + timeSeries.queryOptions.withTimeShift(null)
+            + options.legend.withSortDesc(true),
+
+        legendRightPlacement(title, unit, targets, gridPos):
+            self.base(title, unit, targets, gridPos)
+            + options.legend.withCalcs([
+                'max'
+            ])
+            + options.legend.withShowLegend(true)
+            + options.legend.withDisplayMode('table')
+            + options.legend.withPlacement('right')
+            + options.legend.withAsTable(true)
+            + options.tooltip.withMode('multi'),
+
+        legendBottomPlacement(title, unit, targets, gridPos):
+            self.base(title, unit, targets, gridPos)
+            + options.tooltip.withMode('multi')
+            + options.legend.withShowLegend(true)
+            + options.legend.withDisplayMode('list')
+            + options.legend.withPlacement('bottom')
+            + options.tooltip.withMode('multi'),
+
+        withRequestWaitDurationAggregations(title, unit, targets, gridPos):
+            self.legendRightPlacement(title, unit, targets, gridPos)
+            + options.legend.withCalcs([
+                'mean',
+                'max',
+                'lastNotNull'
+            ])  
+    }
+}
diff --git a/assets/api-performance-overview/queries.libsonnet b/assets/api-performance-overview/queries.libsonnet
@@ -0,0 +1,189 @@
+local g = import 'github.com/grafana/grafonnet/gen/grafonnet-latest/main.libsonnet';
+local variables = import './variables.libsonnet';
+local prometheus = g.query.prometheus;
+
+{
+    request_duration_99th_quantile: {
+        query():
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",subresource!="log",verb!~"WATCH|WATCHLIST|PROXY"}[$interval])) by(verb,le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{verb}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestRateByInstance: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_request_total{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",code=~"$code",verb=~"$verb"}[$interval])) by(instance)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{instance}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestDuarationByResource: {
+        query():
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",subresource!="log",verb!~"WATCH|WATCHLIST|PROXY"}[$interval])) by(resource,le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{resource}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestDurationBy99Quatile: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_request_total{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",code=~"$code",verb=~"$verb"}[$interval])) by(resource)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{resource}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestDurationReadWrite: {
+        query():
+            [ prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",verb=~"LIST|GET"}[$interval])) by(le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('read')
+            + prometheus.withDatasource('$Datasource'),
+
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",verb=~"POST|PUT|PATCH|UPDATE|DELETE"}[$interval])) by(le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('write')
+            + prometheus.withDatasource('$Datasource')]
+    },
+
+    requestRateReadWrite: {
+        query():
+            [prometheus.withExpr('sum(rate(apiserver_request_total{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",verb=~"LIST|GET"}[$interval]))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('read')
+            + prometheus.withDatasource('$Datasource'),
+
+            prometheus.withExpr('sum(rate(apiserver_request_total{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",verb=~"POST|PUT|PATCH|UPDATE|DELETE"}[$interval]))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('write')
+            + prometheus.withDatasource('$Datasource')
+            ]
+    },
+
+    requestRateDropped: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_dropped_requests_total{instance=~"$instance"}[$interval])) by (requestKind)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestRateTerminated: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_request_terminations_total{instance=~"$instance",resource=~"$resource",code=~"$code"}[$interval])) by(component)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestRateStatus: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_request_total{apiserver=~"$apiserver",instance=~"$instance",resource=~"$resource",verb=~"$verb",code=~"$code"}[$interval])) by(code)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{code}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestsLongRunning: {
+        query():
+            prometheus.withExpr('sum(apiserver_longrunning_gauge{instance=~"$instance",resource=~"$resource",verb=~"$verb"}) by(instance)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{instance}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestInFlight: {
+        query():
+            prometheus.withExpr('sum(apiserver_current_inflight_requests{instance=~"$instance"}) by (instance,requestKind)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{requestKind}}-{{instance}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestRejectPandF: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_flowcontrol_rejected_requests_total{instance=~"$instance",flowSchema=~"$flowSchema",priorityLevel=~"$priorityLevel"}[$interval])) by (reason)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    responseSize99Quatile: {
+        query():
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_response_sizes_bucket{instance=~"$instance",resource=~"$resource",verb=~"$verb"}[$interval])) by(instance,le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{instance}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestQueueLengthPandF: {
+        query():
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_flowcontrol_request_queue_length_after_enqueue_bucket{instance=~"$instance",flowSchema=~"$flowSchema",priorityLevel=~"$priorityLevel"}[$interval])) by(flowSchema, priorityLevel, le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{flowSchema}}:{{priorityLevel}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestWaitDuration99QuatilePandF: {
+        query():
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_flowcontrol_request_wait_duration_seconds_bucket{instance=~"$instance"}[5m])) by(flow_schema, priority_level, le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestDispatchRatePandF: {
+        query():
+            prometheus.withExpr('sum(rate(apiserver_flowcontrol_dispatched_requests_total{instance=~"$instance",flowSchema=~"$flowSchema",priorityLevel=~"$priorityLevel"}[$interval])) by(flowSchema,priorityLevel)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{flowSchema}}:{{priorityLevel}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    requestExecutionDurationPandF: {
+        query():
+            prometheus.withExpr('histogram_quantile(0.99, sum(rate(apiserver_flowcontrol_request_execution_seconds_bucket{instance=~"$instance",flowSchema=~"$flowSchema",priorityLevel=~"$priorityLevel"}[$interval])) by(flowSchema, priorityLevel, le))')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{flowSchema}}:{{priorityLevel}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    pendingInQueuePandF: {
+        query():
+            prometheus.withExpr('sum(apiserver_flowcontrol_current_inqueue_requests{instance=~"$instance",flowSchema=~"$flowSchema",priorityLevel=~"$priorityLevel"}) by (flowSchema,priorityLevel)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('{{flowSchema}}:{{priorityLevel}}')
+            + prometheus.withDatasource('$Datasource')
+    },
+
+    concurrencyLimitByKubeapiserverPandF: {
+        query():
+            prometheus.withExpr('sum(apiserver_flowcontrol_request_concurrency_limit{instance=~".*:6443",priorityLevel=~"$priorityLevel"}) by (instance,priorityLevel)')
+            + prometheus.withFormat('time_series')
+            + prometheus.withIntervalFactor(2)
+            + prometheus.withLegendFormat('')
+            + prometheus.withDatasource('$Datasource')
+    }
+}
diff --git a/assets/api-performance-overview/variables.libsonnet b/assets/api-performance-overview/variables.libsonnet
@@ -0,0 +1,74 @@
+local g = import 'github.com/grafana/grafonnet/gen/grafonnet-latest/main.libsonnet';
+local var = g.dashboard.variable;
+
+{
+    Datasource:
+        var.datasource.new('Datasource','prometheus')
+        + var.datasource.withRegex("")
+        + var.query.generalOptions.withLabel('Datasource')
+        + var.query.withRefresh(1),
+    apiserver:
+        var.query.new('apiserver','label_values(apiserver_request_duration_seconds_bucket, apiserver)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('apisever')
+        + var.query.withRefresh(2),
+
+    instance: 
+        var.query.new('instance','label_values(apiserver_request_total, instance)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('instance')
+        + var.query.withRefresh(2),
+
+    resource: 
+        var.query.new('resource','label_values(apiserver_request_duration_seconds_bucket, resource)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('resource')
+        + var.query.withRefresh(2),
+
+    code:
+        var.query.new('code','label_values(code)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('code')
+        + var.query.withRefresh(2),
+
+    verb: 
+        var.query.new('verb','label_values(verb)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('verb')
+        + var.query.withRefresh(2),
+
+    flowSchema:
+        var.query.new('flowSchema','label_values(flowSchema)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('flow-schema')
+        + var.query.withRefresh(2),
+
+    priorityLevel: 
+        var.query.new('priorityLevel','label_values(priorityLevel)')
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+        + var.query.generalOptions.withLabel('priority-level')
+        + var.query.withRefresh(2),
+
+    interval:
+        var.interval.new('interval',['1m','5m'])
+        + var.query.withDatasourceFromVariable(self.Datasource)
+        + var.interval.generalOptions.withLabel('interval')
+        + var.interval.withAutoOption(count=30, minInterval='10s')
+        + var.query.withRefresh(2)
+        + var.query.selectionOptions.withMulti(false)
+        + var.query.selectionOptions.withIncludeAll(true)
+}