Add units to timeseries panel

The yaxes field doesn't seem to do anything in the timeseries panel and was replaced with the units field instead. So I defaulted the units to be short and allowed it to be set for the panel. Signed-off-by: Charlie Le <[email protected]>
cortexproject · Oct 22, 2024 · f6a9025 · f6a9025
1 parent 3ab33de
commit f6a9025
Show file tree

Hide file tree

Showing 10 changed files with 173 additions and 141 deletions.
diff --git a/cortex-mixin/dashboards/alertmanager.libsonnet b/cortex-mixin/dashboards/alertmanager.libsonnet
@@ -73,7 +73,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
         )
       )
       .addPanel(
-        $.timeseriesPanel('Latency') +
+        $.timeseriesPanel('Latency', unit='ms') +
         $.latencyPanel('cortex_alertmanager_notification_latency_seconds', '{%s}' % $.jobMatcher($._config.job_names.alertmanager))
       )
     )
@@ -84,7 +84,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
         $.qpsPanel('cortex_request_duration_seconds_count{%s, route=~"api_v1_alerts|alertmanager"}' % $.jobMatcher($._config.job_names.gateway))
       )
       .addPanel(
-        $.timeseriesPanel('Latency') +
+        $.timeseriesPanel('Latency', unit='ms') +
         utils.latencyRecordingRulePanel('cortex_request_duration_seconds', $.jobSelector($._config.job_names.gateway) + [utils.selector.re('route', 'api_v1_alerts|alertmanager')])
       )
     )
@@ -166,7 +166,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
         }
       )
       .addPanel(
-        $.timeseriesPanel('Initial sync duration') +
+        $.timeseriesPanel('Initial sync duration', unit='s') +
         $.latencyPanel('cortex_alertmanager_state_initial_sync_duration_seconds', '{%s}' % $.jobMatcher($._config.job_names.alertmanager)) + {
           targets: [
             target {

diff --git a/cortex-mixin/dashboards/compactor.libsonnet b/cortex-mixin/dashboards/compactor.libsonnet
@@ -14,7 +14,6 @@ local utils = import 'mixin-utils/utils.libsonnet';
           'sum(rate(cortex_compactor_runs_failed_total{%s}[$__rate_interval]))' % $.jobMatcher($._config.job_names.compactor)
         ) +
         $.bars +
-        { yaxes: $.yaxes('ops') } +
         $.panelDescription(
           'Per-instance runs',
           |||
@@ -44,9 +43,8 @@ local utils = import 'mixin-utils/utils.libsonnet';
     .addRow(
       $.row('')
       .addPanel(
-        $.timeseriesPanel('Compacted blocks / sec') +
+        $.timeseriesPanel('Compacted blocks / sec', unit='ops') +
         $.queryPanel('sum(rate(prometheus_tsdb_compactions_total{%s}[$__rate_interval]))' % $.jobMatcher($._config.job_names.compactor), 'blocks') +
-        { yaxes: $.yaxes('ops') } +
         $.panelDescription(
           'Compacted blocks / sec',
           |||
@@ -55,7 +53,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
         ),
       )
       .addPanel(
-        $.timeseriesPanel('Per-block compaction duration') +
+        $.timeseriesPanel('Per-block compaction duration', unit='s') +
         $.latencyPanel('prometheus_tsdb_compaction_duration_seconds', '{%s}' % $.jobMatcher($._config.job_names.compactor)) +
         $.panelDescription(
           'Per-block compaction duration',
@@ -85,9 +83,8 @@ local utils = import 'mixin-utils/utils.libsonnet';
     .addRow(
       $.row('Garbage Collector')
       .addPanel(
-        $.timeseriesPanel('Blocks marked for deletion / sec') +
-        $.queryPanel('sum(rate(cortex_compactor_blocks_marked_for_deletion_total{%s}[$__rate_interval]))' % $.jobMatcher($._config.job_names.compactor), 'blocks') +
-        { yaxes: $.yaxes('ops') },
+        $.timeseriesPanel('Blocks marked for deletion / sec', unit='ops') +
+        $.queryPanel('sum(rate(cortex_compactor_blocks_marked_for_deletion_total{%s}[$__rate_interval]))' % $.jobMatcher($._config.job_names.compactor), 'blocks'),
       )
       .addPanel(
         $.successFailurePanel(
@@ -111,7 +108,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
         ) + { yaxes: $.yaxes('ops') }
       )
       .addPanel(
-        $.timeseriesPanel('Metadata Sync Duration') +
+        $.timeseriesPanel('Metadata Sync Duration', unit='ms') +
         // This metric tracks the duration of a per-tenant metadata sync.
         $.latencyPanel('cortex_compactor_meta_sync_duration_seconds', '{%s}' % $.jobMatcher($._config.job_names.compactor)),
       )

diff --git a/cortex-mixin/dashboards/config.libsonnet b/cortex-mixin/dashboards/config.libsonnet
@@ -8,19 +8,17 @@ local utils = import 'mixin-utils/utils.libsonnet';
     .addRow(
       $.row('Startup config file')
       .addPanel(
-        $.timeseriesPanel('Startup config file hashes') +
+        $.timeseriesPanel('Startup config file hashes', unit='instances') +
         $.queryPanel('count(cortex_config_hash{%s}) by (sha256)' % $.namespaceMatcher(), 'sha256:{{sha256}}') +
-        $.stack +
-        { yaxes: $.yaxes('instances') },
+        $.stack,
       )
     )
     .addRow(
       $.row('Runtime config file')
       .addPanel(
-        $.timeseriesPanel('Runtime config file hashes') +
+        $.timeseriesPanel('Runtime config file hashes', unit='instances') +
         $.queryPanel('count(cortex_runtime_config_hash{%s}) by (sha256)' % $.namespaceMatcher(), 'sha256:{{sha256}}') +
-        $.stack +
-        { yaxes: $.yaxes('instances') },
+        $.stack,
       )
     ),
 }
diff --git a/cortex-mixin/dashboards/dashboard-utils.libsonnet b/cortex-mixin/dashboards/dashboard-utils.libsonnet
@@ -62,6 +62,44 @@ local utils = import 'mixin-utils/utils.libsonnet';
                .addTemplate('cluster', 'cortex_build_info', 'cluster')
                .addTemplate('namespace', 'cortex_build_info{cluster=~"$cluster"}', 'namespace'),
     },
+  timeseriesPanel(title, unit='short'):: {
+    datasource: '$datasource',
+    fieldConfig: {
+      defaults: {
+        custom: {
+          drawStyle: 'line',
+          fillOpacity: 1,
+          lineWidth: 1,
+          pointSize: 5,
+          showPoints: 'never',
+          spanNulls: false,
+          stacking: {
+            group: 'A',
+            mode: 'none',
+          },
+        },
+        thresholds: {
+          mode: 'absolute',
+          steps: [],
+        },
+        unit: unit,
+      },
+      overrides: [],
+    },
+    options: {
+      legend: {
+        showLegend: true,
+      },
+      tooltip: {
+        mode: 'single',
+        sort: 'none',
+      },
+    },
+    links: [],
+    targets: [],
+    title: title,
+    type: 'timeseries',
+  },
 
   // The mixin allow specialism of the job selector depending on if its a single binary
   // deployment or a namespaced one.
@@ -108,6 +146,35 @@ local utils = import 'mixin-utils/utils.libsonnet';
         }
         for target in super.targets
       ],
+      fieldConfig+: {
+        defaults+: {
+          custom+: {
+            lineWidth: 0,
+            fillOpacity: 100,  // Get solid fill.
+            stacking: {
+              mode: 'normal',
+              group: 'A',
+            },
+          },
+          unit: 'reqps',
+          min: 0,
+        },
+        overrides+: [{
+          matcher: {
+            id: 'byName',
+            options: status,
+          },
+          properties: [
+            {
+              id: 'color',
+              value: {
+                mode: 'fixed',
+                fixedColor: $.httpStatusColors[status],
+              },
+            },
+          ],
+        } for status in std.objectFieldsAll($.httpStatusColors)],
+      },
     },
 
   latencyPanel(metricName, selector, multiplier='1e3')::
@@ -121,7 +188,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     },
 
   successFailurePanel(title, successMetric, failureMetric)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='short') +
     $.queryPanel([successMetric, failureMetric], ['successful', 'failed']) +
     $.stack + {
       aliasColors: {
@@ -132,7 +199,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
 
   // Displays started, completed and failed rate.
   startedCompletedFailedPanel(title, startedMetric, completedMetric, failedMetric)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='ops') +
     $.queryPanel([startedMetric, completedMetric, failedMetric], ['started', 'completed', 'failed']) +
     $.stack + {
       aliasColors: {
@@ -160,7 +227,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     },
 
   containerMemoryWorkingSetPanel(title, containerName)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='bytes') +
     $.queryPanel([
       // We use "max" instead of "sum" otherwise during a rolling update of a statefulset we will end up
       // summing the memory of the old instance/pod (whose metric will be stale for 5m) to the new instance/pod.
@@ -180,7 +247,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     },
 
   containerNetworkPanel(title, metric, instanceName)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='Bps') +
     $.queryPanel(
       'sum by(%(instance)s) (rate(%(metric)s{%(namespace)s,%(instance)s=~"%(instanceName)s"}[$__rate_interval]))' % {
         namespace: $.namespaceMatcher(),
@@ -199,7 +266,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     $.containerNetworkPanel('Transmit Bandwidth', 'container_network_transmit_bytes_total', instanceName),
 
   containerDiskWritesPanel(title, containerName)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='Bps') +
     $.queryPanel(
       |||
         sum by(%s, %s, device) (
@@ -220,7 +287,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     { yaxes: $.yaxes('Bps') },
 
   containerDiskReadsPanel(title, containerName)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='Bps') +
     $.queryPanel(
       |||
         sum by(%s, %s, device) (
@@ -239,7 +306,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     { yaxes: $.yaxes('Bps') },
 
   containerDiskSpaceUtilization(title, containerName)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='percentunit') +
     $.queryPanel(
       |||
         max by(persistentvolumeclaim) (
@@ -266,7 +333,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     else 'label_name="%s"' % containerName,
 
   goHeapInUsePanel(title, jobName)::
-    $.timeseriesPanel(title) +
+    $.timeseriesPanel(title, unit='bytes') +
     $.queryPanel(
       'sum by(%s) (go_memstats_heap_inuse_bytes{%s})' % [$._config.per_instance_label, $.jobMatcher(jobName)],
       '{{%s}}' % $._config.per_instance_label
@@ -361,13 +428,12 @@ local utils = import 'mixin-utils/utils.libsonnet';
   getObjectStoreRows(title, component):: [
     super.row(title)
     .addPanel(
-      $.timeseriesPanel('Operations / sec') +
+      $.timeseriesPanel('Operations / sec', unit='rps') +
       $.queryPanel('sum by(operation) (rate(thanos_objstore_bucket_operations_total{%s,component="%s"}[$__rate_interval]))' % [$.namespaceMatcher(), component], '{{operation}}') +
-      $.stack +
-      { yaxes: $.yaxes('rps') },
+      $.stack
     )
     .addPanel(
-      $.timeseriesPanel('Error rate') +
+      $.timeseriesPanel('Error rate', unit='percentunit') +
       $.queryPanel('sum by(operation) (rate(thanos_objstore_bucket_operation_failures_total{%s,component="%s"}[$__rate_interval])) / sum by(operation) (rate(thanos_objstore_bucket_operations_total{%s,component="%s"}[$__rate_interval]))' % [$.namespaceMatcher(), component, $.namespaceMatcher(), component], '{{operation}}') +
       { yaxes: $.yaxes('percentunit') },
     )
@@ -406,7 +472,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
     };
     super.row(title)
     .addPanel(
-      $.timeseriesPanel('Requests / sec') +
+      $.timeseriesPanel('Requests / sec', unit='ops') +
       $.queryPanel(
         |||
           sum by(operation) (
@@ -439,7 +505,7 @@ local utils = import 'mixin-utils/utils.libsonnet';
       )
     )
     .addPanel(
-      $.timeseriesPanel('Hit ratio') +
+      $.timeseriesPanel('Hit ratio', unit='percentunit') +
       $.queryPanel(
         |||
           sum(

diff --git a/cortex-mixin/dashboards/object-store.libsonnet b/cortex-mixin/dashboards/object-store.libsonnet
@@ -7,29 +7,25 @@ local utils = import 'mixin-utils/utils.libsonnet';
     .addRow(
       $.row('Components')
       .addPanel(
-        $.timeseriesPanel('RPS / component') +
+        $.timeseriesPanel('RPS / component', unit='rps') +
         $.queryPanel('sum by(component) (rate(thanos_objstore_bucket_operations_total{%s}[$__rate_interval]))' % $.namespaceMatcher(), '{{component}}') +
-        $.stack +
-        { yaxes: $.yaxes('rps') },
+        $.stack,
       )
       .addPanel(
-        $.timeseriesPanel('Error rate / component') +
-        $.queryPanel('sum by(component) (rate(thanos_objstore_bucket_operation_failures_total{%s}[$__rate_interval])) / sum by(component) (rate(thanos_objstore_bucket_operations_total{%s}[$__rate_interval]))' % [$.namespaceMatcher(), $.namespaceMatcher()], '{{component}}') +
-        { yaxes: $.yaxes('percentunit') },
+        $.timeseriesPanel('Error rate / component', unit='percentunit') +
+        $.queryPanel('sum by(component) (rate(thanos_objstore_bucket_operation_failures_total{%s}[$__rate_interval])) / sum by(component) (rate(thanos_objstore_bucket_operations_total{%s}[$__rate_interval]))' % [$.namespaceMatcher(), $.namespaceMatcher()], '{{component}}')
       )
     )
     .addRow(
       $.row('Operations')
       .addPanel(
-        $.timeseriesPanel('RPS / operation') +
+        $.timeseriesPanel('RPS / operation', unit='rps') +
         $.queryPanel('sum by(operation) (rate(thanos_objstore_bucket_operations_total{%s}[$__rate_interval]))' % $.namespaceMatcher(), '{{operation}}') +
-        $.stack +
-        { yaxes: $.yaxes('rps') },
+        $.stack,
       )
       .addPanel(
-        $.timeseriesPanel('Error rate / operation') +
-        $.queryPanel('sum by(operation) (rate(thanos_objstore_bucket_operation_failures_total{%s}[$__rate_interval])) / sum by(operation) (rate(thanos_objstore_bucket_operations_total{%s}[$__rate_interval]))' % [$.namespaceMatcher(), $.namespaceMatcher()], '{{operation}}') +
-        { yaxes: $.yaxes('percentunit') },
+        $.timeseriesPanel('Error rate / operation', unit='percentunit') +
+        $.queryPanel('sum by(operation) (rate(thanos_objstore_bucket_operation_failures_total{%s}[$__rate_interval])) / sum by(operation) (rate(thanos_objstore_bucket_operations_total{%s}[$__rate_interval]))' % [$.namespaceMatcher(), $.namespaceMatcher()], '{{operation}}')
       )
     )
     .addRow(