From 7a1324de72f4cf80d14252c6134e1f7c775befbb Mon Sep 17 00:00:00 2001 From: riskyadventure Date: Thu, 21 Nov 2024 15:05:19 -0800 Subject: [PATCH] version 1.1 --- .../eks/infrastructure/kcm.json | 83 +++++++++---------- .../eks/infrastructure/ksh.json | 65 +++++++-------- 2 files changed, 66 insertions(+), 82 deletions(-) diff --git a/artifacts/grafana-dashboards/eks/infrastructure/kcm.json b/artifacts/grafana-dashboards/eks/infrastructure/kcm.json index 42ff4ff..d4a38e5 100644 --- a/artifacts/grafana-dashboards/eks/infrastructure/kcm.json +++ b/artifacts/grafana-dashboards/eks/infrastructure/kcm.json @@ -1,14 +1,5 @@ { - "__inputs": [ - { - "name": "DS_PROMETHEUS", - "label": "prometheus", - "description": "", - "type": "datasource", - "pluginId": "prometheus", - "pluginName": "Prometheus" - } - ], + "__inputs": [], "__elements": {}, "__requires": [ { @@ -86,7 +77,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "${datasource}" }, "fieldConfig": { "defaults": { @@ -169,12 +160,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "topk(5, rate(workqueue_queue_duration_seconds_sum{job=\"$job\"}[5m]) / rate(workqueue_queue_duration_seconds_count{job=\"$job\"}[5m]))", + "expr": "topk(5, rate(workqueue_queue_duration_seconds_sum{job=\"$job\"}[$__rate_interval]) / rate(workqueue_queue_duration_seconds_count{job=\"$job\"}[$__rate_interval]))", "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{name}}", "range": true, "refId": "A" @@ -187,7 +178,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -271,13 +262,13 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "topk(5, rate(workqueue_work_duration_seconds_sum{job=\"$job\"}[5m]) / rate(workqueue_work_duration_seconds_count{job=\"$job\"}[5m])) > 0", + "expr": "topk(5, rate(workqueue_work_duration_seconds_sum{job=\"$job\"}[$__rate_interval]) / rate(workqueue_work_duration_seconds_count{job=\"$job\"}[$__rate_interval])) > 0", "hide": false, "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{name}}", "range": true, "refId": "B" @@ -290,7 +281,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -319,7 +310,6 @@ } }, "mappings": [], - "noValue": "No Value", "thresholds": { "mode": "absolute", "steps": [ @@ -374,12 +364,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "topk(5, sum_over_time(workqueue_depth{job=\"$job\"}[5m])) > 0", + "expr": "topk(5, sum_over_time(workqueue_depth{job=\"$job\"}[$__rate_interval])) > 0", "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{name}}", "range": true, "refId": "A" @@ -405,7 +395,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "description": "Length of Time in Queue", "fieldConfig": { @@ -491,12 +481,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "rate(workqueue_queue_duration_seconds_sum{job=\"$job\"}[5m]) / rate(workqueue_queue_duration_seconds_count{job=\"$job\"}[5m]) > 0", + "expr": "rate(workqueue_queue_duration_seconds_sum{job=\"$job\"}[$__rate_interval]) / rate(workqueue_queue_duration_seconds_count{job=\"$job\"}[$__rate_interval]) > 0", "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{name}}", "range": true, "refId": "A" @@ -509,7 +499,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -597,11 +587,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "sum_over_time(workqueue_depth{job=\"$job\"}[5m]) > 1 ", + "expr": "sum_over_time(workqueue_depth{job=\"$job\"}[$__rate_interval]) > 1 ", "instant": false, + "interval": "1m", "legendFormat": "__auto", "range": true, "refId": "A" @@ -613,9 +604,8 @@ }, { "datasource": { - "default": false, - "type": "datasource", - "uid": "-- Mixed --" + "type": "prometheus", + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -702,12 +692,13 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "sum(increase(workqueue_adds_total{job=\"$job\"}[5m]) > 1) by (name)", - "interval": "5m", - "legendFormat": "{{cluster}} {{instance}} {{name}}", + "expr": "sum(increase(workqueue_adds_total{job=\"$job\"}[$__rate_interval]) > 1) by (name)", + "instant": false, + "interval": "$__rate_interval", + "legendFormat": "__auto", "range": true, "refId": "A" } @@ -717,7 +708,7 @@ "type": "timeseries" } ], - "refresh": "5s", + "refresh": "5m", "schemaVersion": 39, "tags": [ "RiskyAdventure" @@ -727,8 +718,8 @@ { "current": { "selected": false, - "text": "prometheus", - "value": "be3cycu4o9340e" + "text": "default", + "value": "default" }, "hide": 0, "includeAll": false, @@ -747,11 +738,11 @@ "current": {}, "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "definition": "label_values(job)", "hide": 0, - "includeAll": true, + "includeAll": false, "label": "Job", "multi": false, "name": "job", @@ -761,10 +752,10 @@ "query": "label_values(job)", "refId": "PrometheusVariableQueryEditor-VariableQuery" }, - "refresh": 2, + "refresh": 1, "regex": "", "skipUrlSync": false, - "sort": 1, + "sort": 0, "type": "query" } ] @@ -780,7 +771,7 @@ }, "timezone": "utc", "title": "EKS Controller Manager", - "uid": "8675309", - "version": 5, + "uid": "86753092", + "version": 3, "weekStart": "" } \ No newline at end of file diff --git a/artifacts/grafana-dashboards/eks/infrastructure/ksh.json b/artifacts/grafana-dashboards/eks/infrastructure/ksh.json index 648205f..b9f5cd2 100644 --- a/artifacts/grafana-dashboards/eks/infrastructure/ksh.json +++ b/artifacts/grafana-dashboards/eks/infrastructure/ksh.json @@ -1,14 +1,5 @@ { - "__inputs": [ - { - "name": "DS_PROMETHEUS", - "label": "prometheus", - "description": "", - "type": "datasource", - "pluginId": "prometheus", - "pluginName": "Prometheus" - } - ], + "__inputs": [], "__elements": {}, "__requires": [ { @@ -63,7 +54,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -146,12 +137,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "sum(sum_over_time(scheduler_pending_pods[1m]))by (queue)", + "expr": "sum(sum_over_time(scheduler_pending_pods[$__rate_interval]))by (queue)", "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "__auto", "range": true, "refId": "A" @@ -164,7 +155,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -247,13 +238,13 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "increase(scheduler_pod_scheduling_attempts_sum[5m]) / increase(scheduler_pod_scheduling_attempts_count[5m])", + "expr": "increase(scheduler_pod_scheduling_attempts_sum[$__rate_interval]) / increase(scheduler_pod_scheduling_attempts_count[$__rate_interval])", "hide": false, "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{instance}}", "range": true, "refId": "A" @@ -266,7 +257,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -312,7 +303,8 @@ "mode": "absolute", "steps": [ { - "color": "purple" + "color": "purple", + "value": null }, { "color": "red", @@ -350,12 +342,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "rate(scheduler_scheduling_attempt_duration_seconds_sum{}[5m]) / rate(scheduler_scheduling_attempt_duration_seconds_count[5m])", + "expr": "rate(scheduler_scheduling_attempt_duration_seconds_sum{}[$__rate_interval]) / rate(scheduler_scheduling_attempt_duration_seconds_count[$__rate_interval])", "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{profile}} - {{instance}}", "range": true, "refId": "A" @@ -368,7 +360,7 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "fieldConfig": { "defaults": { @@ -414,7 +406,8 @@ "mode": "absolute", "steps": [ { - "color": "purple" + "color": "purple", + "value": null }, { "color": "red", @@ -451,12 +444,12 @@ { "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "editorMode": "code", - "expr": "increase(scheduler_preemption_attempts_total[5m])", + "expr": "rate(scheduler_preemption_attempts_total[$__rate_interval])", "instant": false, - "interval": "5m", + "interval": "1m", "legendFormat": "{{profile}} - {{instance}}", "range": true, "refId": "A" @@ -467,18 +460,18 @@ "type": "timeseries" } ], - "refresh": "5s", + "refresh": "5m", "schemaVersion": 39, "tags": [ - "RiskyAdventure" + "Infrastructure" ], "templating": { "list": [ { "current": { "selected": false, - "text": "No data sources found", - "value": "" + "text": "default", + "value": "default" }, "hide": 2, "includeAll": false, @@ -487,7 +480,7 @@ "options": [], "query": "prometheus", "refresh": 1, - "regex": "/Prom.*/", + "regex": "", "skipUrlSync": false, "type": "datasource" }, @@ -495,7 +488,7 @@ "current": {}, "datasource": { "type": "prometheus", - "uid": "${DS_PROMETHEUS}" + "uid": "$datasource" }, "definition": "label_values(scheduler_pending_pods,job)", "hide": 0, @@ -529,7 +522,7 @@ }, "timezone": "", "title": "EKS Scheduler", - "uid": "CqT0Jg-nz", - "version": 4, + "uid": "CqT0Jg-nz2", + "version": 5, "weekStart": "" } \ No newline at end of file