Merge branch 'main' into js-version

ydb-platform · Mar 6, 2024 · 339f9f0 · 339f9f0
2 parents a7fd881 + cfda13d
commit 339f9f0
Show file tree

Hide file tree

Showing 8 changed files with 816 additions and 72 deletions.
diff --git a/k8s/helms/grafana.yaml b/k8s/helms/grafana.yaml
@@ -96,17 +96,6 @@ dashboards:
           "links": [],
           "liveNow": false,
           "panels": [
-            {
-              "gridPos": {
-                "h": 1,
-                "w": 24,
-                "x": 0,
-                "y": 0
-              },
-              "id": 9,
-              "title": "Row title",
-              "type": "row"
-            },
             {
               "datasource": {
                 "type": "prometheus",
@@ -168,14 +157,14 @@ dashboards:
                 "h": 9,
                 "w": 12,
                 "x": 0,
-                "y": 1
+                "y": 0
               },
-              "id": 2,
+              "id": 12,
               "options": {
                 "legend": {
                   "calcs": [],
-                  "displayMode": "table",
-                  "placement": "right",
+                  "displayMode": "list",
+                  "placement": "bottom",
                   "showLegend": true
                 },
                 "tooltip": {
@@ -190,19 +179,7 @@ dashboards:
                     "uid": "prometheus"
                   },
                   "editorMode": "builder",
-                  "expr": "inflight > 0",
-                  "format": "time_series",
-                  "legendFormat": "({{sdk}}-{{sdkVersion}}) {{jobName}} In flight",
-                  "range": true,
-                  "refId": "A"
-                },
-                {
-                  "datasource": {
-                    "type": "prometheus",
-                    "uid": "prometheus"
-                  },
-                  "editorMode": "builder",
-                  "expr": "oks > 0",
+                  "expr": "rate(oks[$__rate_interval]) > 0",
                   "hide": false,
                   "legendFormat": "({{sdk}}-{{sdkVersion}}) {{jobName}} OK",
                   "range": true,
@@ -214,14 +191,14 @@ dashboards:
                     "uid": "prometheus"
                   },
                   "editorMode": "builder",
-                  "expr": "not_oks > 0",
+                  "expr": "rate(not_oks[$__rate_interval]) > 0",
                   "hide": false,
                   "legendFormat": "({{sdk}}-{{sdkVersion}}) {{jobName}} not OK",
                   "range": true,
                   "refId": "C"
                 }
               ],
-              "title": "SLO Requests",
+              "title": "SLO Requests RPS",
               "transformations": [],
               "type": "timeseries"
             },
@@ -240,6 +217,7 @@ dashboards:
                     "axisColorMode": "text",
                     "axisLabel": "",
                     "axisPlacement": "auto",
+                    "axisSoftMin": 0,
                     "barAlignment": 0,
                     "drawStyle": "line",
                     "fillOpacity": 0,
@@ -265,6 +243,7 @@ dashboards:
                       "mode": "off"
                     }
                   },
+                  "decimals": 0,
                   "mappings": [],
                   "thresholds": {
                     "mode": "absolute",
@@ -286,14 +265,14 @@ dashboards:
                 "h": 9,
                 "w": 12,
                 "x": 12,
-                "y": 1
+                "y": 0
               },
-              "id": 12,
+              "id": 14,
               "options": {
                 "legend": {
                   "calcs": [],
-                  "displayMode": "table",
-                  "placement": "right",
+                  "displayMode": "list",
+                  "placement": "bottom",
                   "showLegend": true
                 },
                 "tooltip": {
@@ -308,38 +287,14 @@ dashboards:
                     "uid": "prometheus"
                   },
                   "editorMode": "builder",
-                  "expr": "delta(inflight[2s]) / 2 > 0",
-                  "format": "time_series",
-                  "legendFormat": "({{sdk}}-{{sdkVersion}}) {{jobName}} In flight",
-                  "range": true,
-                  "refId": "A"
-                },
-                {
-                  "datasource": {
-                    "type": "prometheus",
-                    "uid": "prometheus"
-                  },
-                  "editorMode": "builder",
-                  "expr": "delta(oks[2s]) / 2 > 0",
-                  "hide": false,
-                  "legendFormat": "({{sdk}}-{{sdkVersion}}) {{jobName}} OK",
-                  "range": true,
-                  "refId": "B"
-                },
-                {
-                  "datasource": {
-                    "type": "prometheus",
-                    "uid": "prometheus"
-                  },
-                  "editorMode": "builder",
-                  "expr": "changes(not_oks[2s]) > 0",
+                  "expr": "histogram_quantile(1, rate(attempts_bucket[$__rate_interval]))",
                   "hide": false,
-                  "legendFormat": "({{sdk}}-{{sdkVersion}}) {{jobName}} not OK",
+                  "legendFormat": "{{sdk}}-{{sdkVersion}} {{jobName}}-{{status}}",
                   "range": true,
-                  "refId": "C"
+                  "refId": "A"
                 }
               ],
-              "title": "SLO Requests RPS",
+              "title": "Attempts",
               "transformations": [],
               "type": "timeseries"
             },
@@ -349,7 +304,7 @@ dashboards:
                 "h": 1,
                 "w": 24,
                 "x": 0,
-                "y": 10
+                "y": 9
               },
               "id": 7,
               "panels": [],
@@ -417,7 +372,7 @@ dashboards:
                 "h": 8,
                 "w": 12,
                 "x": 0,
-                "y": 11
+                "y": 10
               },
               "id": 4,
               "options": {
@@ -510,7 +465,7 @@ dashboards:
                 "h": 8,
                 "w": 12,
                 "x": 12,
-                "y": 11
+                "y": 10
               },
               "id": 5,
               "options": {
@@ -602,7 +557,7 @@ dashboards:
                 "h": 8,
                 "w": 12,
                 "x": 0,
-                "y": 19
+                "y": 18
               },
               "id": 10,
               "options": {
@@ -694,7 +649,7 @@ dashboards:
                 "h": 8,
                 "w": 12,
                 "x": 12,
-                "y": 19
+                "y": 18
               },
               "id": 11,
               "options": {
@@ -726,16 +681,30 @@ dashboards:
               "type": "timeseries"
             }
           ],
-          "refresh": false,
-          "schemaVersion": 37,
+          "refresh": "",
+          "revision": 1,
+          "schemaVersion": 38,
           "style": "dark",
           "tags": [],
           "templating": {
-            "list": []
+            "list": [
+              {
+                "datasource": {
+                  "type": "prometheus",
+                  "uid": "prometheus"
+                },
+                "filters": [],
+                "hide": 0,
+                "label": "",
+                "name": "filter",
+                "skipUrlSync": false,
+                "type": "adhoc"
+              }
+            ]
           },
           "time": {
-            "from": "2023-03-30T12:36:47.587Z",
-            "to": "2023-03-30T12:43:11.246Z"
+            "from": "now-15m",
+            "to": "now"
           },
           "timepicker": {},
           "timezone": "",

diff --git a/playground/.gitignore b/playground/.gitignore
@@ -0,0 +1 @@
+data/
diff --git a/playground/README.md b/playground/README.md
@@ -0,0 +1,40 @@
+# SLO playground
+
+Playground may be used for testing SLO workloads locally
+
+It has several services:
+
+- `prometheus` - storage for metrics
+- `prometheus-pushgateway` - push acceptor for prometheus
+- `grafana` - provides chats for metrics
+- `ydb` - local instance of ydb-database to run workload with
+
+## Network addresses
+
+- Grafana dashboard: http://localhost:3000
+- Prometheus pushgateway: http://localhost:9091
+- YDB monitoring: http://localhost:8765
+- YDB GRPC: grpc://localhost:2136
+- YDB GRPC TLS: grpcs://localhost:2135
+
+## Start
+
+```shell
+docker-compose up -d
+```
+
+## Stop
+
+```shell
+docker-compose down
+```
+
+## Configs
+
+Grafana's dashboards stored in `configs/grafana/provisioning/dashboards`
+
+## Data
+
+YDB databases are not persistent
+
+All other data like metrics and certs stored in `data/`
diff --git a/playground/configs/grafana/provisioning/dashboards/dashboard.yml b/playground/configs/grafana/provisioning/dashboards/dashboard.yml
@@ -0,0 +1,6 @@
+apiVersion: 1
+
+providers:
+  - name: 'SLO'
+    options:
+      path: /etc/grafana/provisioning/dashboards