CLDSRV-520: display utilization service uptime as timeserie

scality · Apr 17, 2024 · 54d36b3 · 54d36b3
1 parent 1ddad9d
commit 54d36b3
Show file tree

Hide file tree

Showing 2 changed files with 75 additions and 60 deletions.
diff --git a/monitoring/dashboard.json b/monitoring/dashboard.json
@@ -2701,60 +2701,68 @@
       "error": false,
       "fieldConfig": {
         "defaults": {
-          "calcs": [
-            "mean"
-          ],
-          "decimals": null,
-          "limit": null,
-          "links": [],
+          "color": {
+            "mode": "palette-classic"
+          },
+          "custom": {
+            "axisLabel": "",
+            "axisPlacement": "auto",
+            "barAlignment": 0,
+            "drawStyle": "line",
+            "fillOpacity": 30,
+            "gradientMode": "none",
+            "hideFrom": {
+              "legend": false,
+              "tooltip": false,
+              "viz": false
+            },
+            "lineInterpolation": "stepAfter",
+            "lineWidth": 1,
+            "pointSize": 5,
+            "scaleDistribution": {
+              "log": 2,
+              "type": "linear"
+            },
+            "showPoints": "auto",
+            "spanNulls": false,
+            "stacking": {},
+            "thresholdsStyle": {
+              "mode": "off"
+            }
+          },
           "mappings": [],
-          "max": 100,
-          "min": 0,
-          "noValue": "-",
-          "override": {},
           "thresholds": {
             "mode": "absolute",
             "steps": [
               {
-                "color": "#808080",
+                "color": "green",
                 "index": 0,
                 "line": true,
                 "op": "gt",
                 "value": "null",
                 "yaxis": "left"
               },
-              {
-                "color": "red",
-                "index": 1,
-                "line": true,
-                "op": "gt",
-                "value": 0.0,
-                "yaxis": "left"
-              },
               {
                 "color": "orange",
-                "index": 2,
+                "index": 1,
                 "line": true,
                 "op": "gt",
                 "value": 90.0,
                 "yaxis": "left"
               },
               {
-                "color": "green",
-                "index": 3,
+                "color": "red",
+                "index": 2,
                 "line": true,
                 "op": "gt",
-                "value": 95.0,
+                "value": 0.0,
                 "yaxis": "left"
               }
             ]
           },
-          "title": null,
-          "unit": "percent",
-          "values": false
+          "unit": "bool_on_off"
         },
-        "showThresholdLabels": false,
-        "showThresholdMarkers": true
+        "overrides": []
       },
       "gridPos": {
         "h": 8,
@@ -2767,10 +2775,13 @@
       "links": [],
       "maxDataPoints": 100,
       "options": {
-        "reduceOptions": {
-          "calcs": [
-            "mean"
-          ]
+        "legend": {
+          "calcs": [],
+          "displayMode": "list",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "single"
         }
       },
       "targets": [
@@ -2782,7 +2793,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Online rate",
+          "legendFormat": "State",
           "metric": "",
           "refId": "",
           "step": 10,
@@ -2792,7 +2803,7 @@
       "title": "Quota service uptime",
       "transformations": [],
       "transparent": false,
-      "type": "gauge"
+      "type": "timeseries"
     },
     {
       "datasource": "${DS_PROMETHEUS}",
@@ -3274,9 +3285,13 @@
       "maxDataPoints": 100,
       "options": {
         "legend": {
-          "calcs": [],
-          "displayMode": "list",
-          "placement": "bottom"
+          "calcs": [
+            "min",
+            "mean",
+            "max"
+          ],
+          "displayMode": "table",
+          "placement": "right"
         },
         "tooltip": {
           "mode": "single"
@@ -3305,7 +3320,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Data write (bucket, success)",
+          "legendFormat": "Data write (bucket)",
           "metric": "",
           "refId": "",
           "step": 10,
@@ -3319,7 +3334,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Data write (bucket, quota exceeded)",
+          "legendFormat": "Data write (bucket, exceeded)",
           "metric": "",
           "refId": "",
           "step": 10,
@@ -3333,7 +3348,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Data write (account, success)",
+          "legendFormat": "Data write (account)",
           "metric": "",
           "refId": "",
           "step": 10,
@@ -3347,7 +3362,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Data write (account, quota exceeded)",
+          "legendFormat": "Data write (account, exceeded)",
           "metric": "",
           "refId": "",
           "step": 10,
@@ -3361,7 +3376,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Data write (account and bucket, success)",
+          "legendFormat": "Data write (account & bucket)",
           "metric": "",
           "refId": "",
           "step": 10,
@@ -3375,7 +3390,7 @@
           "instant": false,
           "interval": "",
           "intervalFactor": 1,
-          "legendFormat": "Data write (account and bucket, quota exceeded)",
+          "legendFormat": "Data write (account & bucket, exceeded)",
           "metric": "",
           "refId": "",
           "step": 10,

diff --git a/monitoring/dashboard.py b/monitoring/dashboard.py
@@ -558,27 +558,24 @@ def top10_errors_by_bucket(title, code):
 )
 
 
-quotaHealth = GaugePanel(
+quotaHealth = TimeSeries(
     title="Quota service uptime",
     dataSource="${DS_PROMETHEUS}",
-    calc="mean",
-    format=UNITS.PERCENT_FORMAT,
-    min=0,
-    max=100,
-    noValue="-",
+    lineInterpolation="stepAfter",
+    fillOpacity=30,
+    unit="bool_on_off",
     targets=[Target(
         expr="\n".join([
             'sum(rate(s3_cloudserver_utilization_service_state{namespace="${namespace}", state="healthy"}[$__rate_interval])) * 100',  # noqa: E501
             "   /",
             'sum(rate(s3_cloudserver_utilization_service_state{namespace="${namespace}", state=~"unhealthy|healthy"}[$__rate_interval]))',  # noqa: E501
         ]),
-        legendFormat="Online rate",
+        legendFormat="State",
     )],
     thresholds=[
-        Threshold("#808080", 0, 0.0),
-        Threshold("red",     1, 0.0),
-        Threshold("orange",  2, 90.0),
-        Threshold("green",   3, 95.0),
+        Threshold("green", 0, 95.0),
+        Threshold("orange", 1, 90.0),
+        Threshold("red", 2, 0.0),
     ],
 )
 
@@ -626,22 +623,25 @@ def top10_errors_by_bucket(title, code):
     dataSource="${DS_PROMETHEUS}",
     lineInterpolation="smooth",
     spanNulls=3*60*1000,
+    legendDisplayMode="table",
+    legendPlacement="right",
+    legendValues=["min", "mean", "max"],
     unit=UNITS.SECONDS,
     targets=[
         average_quota_latency_target(title="Overall"),
         average_quota_latency_target(
-            title="Data write (bucket, success)", type='"bucket"', code='~"2.."'),
+            title="Data write (bucket)", type='"bucket"', code='~"2.."'),
         average_quota_latency_target(
-            title="Data write (bucket, quota exceeded)", type='"bucket"', code='"429"'),
+            title="Data write (bucket, exceeded)", type='"bucket"', code='"429"'),
         average_quota_latency_target(
-            title="Data write (account, success)", type='"account"', code='~"2.."'),
+            title="Data write (account)", type='"account"', code='~"2.."'),
         average_quota_latency_target(
-            title="Data write (account, quota exceeded)", type='"account"', code='"429"'),
+            title="Data write (account, exceeded)", type='"account"', code='"429"'),
         average_quota_latency_target(
-            title="Data write (account and bucket, success)", type='"bucketAccount"',
+            title="Data write (account & bucket)", type='"bucketAccount"',
             code='~"2.."'),
         average_quota_latency_target(
-            title="Data write (account and bucket, quota exceeded)", type='"bucketAccount"',
+            title="Data write (account & bucket, exceeded)", type='"bucketAccount"',
             code='"429"'),
         average_quota_latency_target(
             title="Data deletion", type='"delete"'),