From 8c93fe2ecc4c985e9ac9a56d8059ca046625e926 Mon Sep 17 00:00:00 2001 From: Paul Cuzner Date: Wed, 13 Sep 2017 11:55:20 +1200 Subject: [PATCH] network/backend: dashboards updated to use the network interfaces by prefix Before hand the network stats uses *, and therefore included metrics for 'lo' etc, which was observed to skew reporting in some environments. --- dashboards/current/ceph-backend-storage.json | 2540 ++++++++--------- dashboards/current/network-usage-by-node.json | 59 +- 2 files changed, 1301 insertions(+), 1298 deletions(-) diff --git a/dashboards/current/ceph-backend-storage.json b/dashboards/current/ceph-backend-storage.json index 8cc216a..f0e7a81 100644 --- a/dashboards/current/ceph-backend-storage.json +++ b/dashboards/current/ceph-backend-storage.json @@ -1,1276 +1,1276 @@ { - "dashboard" : { - "links" : [ - { - "icon" : "external link", - "keepTime" : true, - "title" : "Shortcuts", - "tags" : [ - "overview" - ], - "asDropdown" : true, - "type" : "dashboards", - "targetBlank" : true, - "includeVars" : true - } - ], - "version" : 20, - "title" : "Ceph Backend Storage", - "timezone" : "browser", - "timepicker" : { - "refresh_intervals" : [ - "5s", - "10s", - "30s", - "1m", - "5m", - "15m", - "30m", - "1h", - "2h", - "1d" - ], - "time_options" : [ - "5m", - "15m", - "1h", - "6h", - "12h", - "24h", - "2d", - "7d", - "30d" - ] - }, - "time" : { - "from" : "now-1h", - "to" : "now" - }, - "templating" : { - "list" : [ + "dashboard": { + "annotations": { + "list": [] + }, + "editable": false, + "gnetId": null, + "graphTooltip": 0, + "hideControls": true, + "id": 13, + "links": [ { - "sort" : 0, - "tagsQuery" : "", - "regex" : "", - "name" : "cluster_name", - "tagValuesQuery" : "", - "label" : null, - "current" : { - "value" : "ceph", - "text" : "ceph", - "selected" : true - }, - "options" : [], - "multi" : false, - "datasource" : "Local", - "includeAll" : false, - "hide" : 2, - "query" : "collectd.*.$domain.cephmetrics.gauge.*", - "useTags" : false, - "refresh" : 1, - "allValue" : null, - "type" : "query", - "tags" : [] - }, - { - "options" : [ - { - "value" : "test.lab", - "text" : "test.lab", - "selected" : true - } - ], - "multi" : false, - "hide" : 2, - "includeAll" : false, - "query" : "test.lab", - "allValue" : null, - "type" : "custom", - "name" : "domain", - "label" : null, - "current" : { - "selected" : true, - "text" : "test.lab", - "value" : "test.lab" - } - }, + "asDropdown": true, + "icon": "external link", + "includeVars": true, + "keepTime": true, + "tags": [ + "overview" + ], + "targetBlank": true, + "title": "Shortcuts", + "type": "dashboards" + } + ], + "refresh": "10s", + "rows": [ { - "allValue" : null, - "type" : "custom", - "query" : "80,85,90,95,98", - "label" : "Percentile", - "current" : { - "value" : "95", - "text" : "95", - "selected" : true - }, - "name" : "percentile", - "options" : [ - { - "text" : "80", - "selected" : false, - "value" : "80" - }, - { - "selected" : false, - "text" : "85", - "value" : "85" - }, - { - "selected" : false, - "text" : "90", - "value" : "90" - }, - { - "text" : "95", - "selected" : true, - "value" : "95" - }, - { - "value" : "98", - "text" : "98", - "selected" : false - } - ], - "multi" : false, - "includeAll" : false, - "hide" : 0 - }, + "collapse": false, + "height": 250, + "panels": [ + { + "columns": [ + { + "text": "Current", + "value": "current" + } + ], + "fontSize": "100%", + "hideTimeOverride": true, + "id": 56, + "links": [], + "minSpan": 1, + "pageSize": 100, + "scroll": true, + "showHeader": true, + "sort": { + "col": null, + "desc": false + }, + "span": 1, + "styles": [ + { + "alias": "OSD Id", + "colorMode": null, + "colors": [ + "rgba(50, 172, 45, 0.97)", + "rgba(237, 129, 40, 0.89)", + "rgba(245, 54, 54, 0.9)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 0, + "pattern": "Metric", + "sanitize": false, + "thresholds": [ + "" + ], + "type": "number", + "unit": "short" + }, + { + "alias": "Status", + "colorMode": "cell", + "colors": [ + "rgba(50, 172, 45, 0.97)", + "rgba(237, 129, 40, 0.89)", + "rgba(245, 54, 54, 0.9)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 0, + "pattern": "Current", + "thresholds": [ + "0" + ], + "type": "hidden", + "unit": "short" + } + ], + "targets": [ + { + "hide": false, + "refId": "B", + "target": "currentBelow(groupByNode(transformNull(collectd.*.$domain.cephmetrics.gauge.$cluster_name.mon.osd_state.*.up,99),-2,\"maxSeries\"),0.5)", + "textEditor": true + } + ], + "timeFrom": "1m", + "timeShift": null, + "title": "OSD's Down", + "transform": "timeseries_aggregations", + "type": "table" + }, + { + "columns": [ + { + "text": "Current", + "value": "current" + } + ], + "fontSize": "100%", + "id": 57, + "links": [], + "minSpan": 2, + "pageSize": null, + "scroll": true, + "showHeader": true, + "sort": { + "col": 0, + "desc": true + }, + "span": 2, + "styles": [ + { + "alias": "Time", + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "pattern": "Time", + "type": "date" + }, + { + "alias": "% Full", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 0, + "pattern": "Host and Disk", + "thresholds": [], + "type": "number", + "unit": "short" + }, + { + "alias": "% Full", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 0, + "pattern": "Current", + "thresholds": [], + "type": "number", + "unit": "none" + }, + { + "alias": "Host.OSD Id", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 2, + "pattern": "Metric", + "thresholds": [], + "type": "number", + "unit": "short" + }, + { + "alias": "", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "decimals": 2, + "pattern": "/.*/", + "thresholds": [], + "type": "number", + "unit": "short" + } + ], + "targets": [ + { + "refId": "A", + "target": "aliasByNode(currentAbove(transformNull(keepLastValue(collectd.*.$domain.cephmetrics.gauge.$cluster_name.osd.*.osd_percent_used),0),$disk_full_threshold),1,-2)", + "textEditor": true + } + ], + "title": "Disks Near Full", + "transform": "timeseries_aggregations", + "type": "table" + }, + { + "columns": [ + { + "text": "Current", + "value": "current" + } + ], + "fontSize": "100%", + "id": 58, + "links": [], + "maxDataPoints": "1", + "minSpan": 3, + "pageSize": null, + "scroll": true, + "showHeader": true, + "sort": { + "col": 0, + "desc": false + }, + "span": 3, + "styles": [ + { + "alias": "Time", + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "pattern": "Time", + "type": "date" + }, + { + "alias": "Total OSD Capacity", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 2, + "pattern": "Current", + "thresholds": [], + "type": "number", + "unit": "decbytes" + }, + { + "alias": "Host Name", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 2, + "pattern": "Metric", + "thresholds": [], + "type": "number", + "unit": "short" + }, + { + "alias": "", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "decimals": 2, + "pattern": "/.*/", + "thresholds": [], + "type": "number", + "unit": "short" + } + ], + "targets": [ + { + "refId": "A", + "target": "groupByNode(keepLastValue(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.stat_bytes),1,\"sumSeries\")", + "textEditor": true + } + ], + "title": "Host OSD Raw Capacity", + "transform": "timeseries_aggregations", + "type": "table" + }, + { + "columns": [ + { + "text": "Current", + "value": "current" + } + ], + "fontSize": "100%", + "id": 59, + "links": [], + "maxDataPoints": "10", + "minSpan": 3, + "pageSize": null, + "scroll": true, + "showHeader": true, + "sort": { + "col": 0, + "desc": false + }, + "span": 3, + "styles": [ + { + "alias": "Time", + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "pattern": "Time", + "type": "date" + }, + { + "alias": "# OSDs", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 0, + "pattern": "Current", + "thresholds": [], + "type": "number", + "unit": "none" + }, + { + "alias": "Host Name", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "dateFormat": "YYYY-MM-DD HH:mm:ss", + "decimals": 2, + "pattern": "Metric", + "thresholds": [], + "type": "number", + "unit": "short" + }, + { + "alias": "", + "colorMode": null, + "colors": [ + "rgba(245, 54, 54, 0.9)", + "rgba(237, 129, 40, 0.89)", + "rgba(50, 172, 45, 0.97)" + ], + "decimals": 2, + "pattern": "/.*/", + "thresholds": [], + "type": "number", + "unit": "short" + } + ], + "targets": [ + { + "refId": "A", + "target": "aliasByNode(keepLastValue(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.num_osds,6),1)", + "textEditor": true + } + ], + "title": "Host OSD Capacity Breakdown", + "transform": "timeseries_aggregations", + "type": "table" + } + ], + "repeat": null, + "repeatIteration": null, + "repeatRowId": null, + "showTitle": true, + "title": "Disk/OSD Host Summary", + "titleSize": "h5" + }, { - "hide" : 0, - "includeAll" : true, - "options" : [ - { - "selected" : true, - "text" : "All", - "value" : "$__all" - }, - { - "value" : "ceph-1", - "selected" : false, - "text" : "ceph-1" - }, - { - "value" : "ceph-2", - "selected" : false, - "text" : "ceph-2" - }, - { - "selected" : false, - "text" : "ceph-3", - "value" : "ceph-3" - } - ], - "multi" : true, - "label" : "OSD Hostname", - "current" : { - "value" : "$__all", - "selected" : true, - "text" : "All" - }, - "name" : "osd_servers", - "allValue" : "", - "type" : "custom", - "query" : "ceph-1,ceph-2,ceph-3" - }, + "collapse": true, + "height": "300", + "panels": [ + { + "cards": { + "cardPadding": null, + "cardRound": null + }, + "color": { + "cardColor": "#b4ff00", + "colorScale": "sqrt", + "colorScheme": "interpolateRdYlGn", + "exponent": 0.5, + "mode": "spectrum" + }, + "dataFormat": "timeseries", + "description": "The heatmap categorizes disk utilization into discrete buckets (e.g util 0-5) and shows the frequency of the number of disks that fall within that range as a color. The color chosen depends on the number of disks in the 'bucket', ranging from green (low) to red (high). Hover over a colored block to show the count of disk utilization observations at that point.", + "heatmap": {}, + "highlightCards": true, + "id": 53, + "links": [], + "minSpan": 6, + "span": 6, + "targets": [ + { + "refId": "A", + "target": "collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.util", + "textEditor": true + } + ], + "title": "Disk Drive Utilization Heatmap - $osd_servers", + "tooltip": { + "show": true, + "showHistogram": false + }, + "type": "heatmap", + "xAxis": { + "show": true + }, + "xBucketNumber": 180, + "xBucketSize": "", + "yAxis": { + "decimals": null, + "format": "short", + "logBase": 1, + "max": "100", + "min": "0", + "show": true, + "splitFactor": null + }, + "yBucketNumber": null, + "yBucketSize": 5 + }, + { + "aliasColors": { + "Read Throughput": "#629E51", + "Write Throughput": "#E0752D" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 1, + "id": 24, + "legend": { + "avg": false, + "current": false, + "max": false, + "min": false, + "show": true, + "total": false, + "values": false + }, + "lines": true, + "linewidth": 1, + "links": [], + "minSpan": 6, + "nullPointMode": "null", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [], + "spaceLength": 10, + "span": 6, + "stack": true, + "steppedLine": false, + "targets": [ + { + "refId": "A", + "target": "alias(sumSeries(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.r_bytes_per_sec), \"Read Throughput\")", + "textEditor": true + }, + { + "refId": "B", + "target": "alias(sumSeries(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.w_bytes_per_sec), \"Write Throughput\")", + "textEditor": true + } + ], + "thresholds": [], + "timeFrom": null, + "timeShift": null, + "title": "Disk Throughput - $osd_servers", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "decbytes", + "label": null, + "logBase": 1, + "max": null, + "min": "0", + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": true + } + ] + }, + { + "aliasColors": { + "Read Latency": "#629E51", + "Write Latency": "#E0752D" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 1, + "id": 40, + "legend": { + "avg": false, + "current": false, + "max": false, + "min": false, + "show": true, + "total": false, + "values": false + }, + "lines": true, + "linewidth": 1, + "links": [ + { + "dashUri": "db/latency-by-server", + "dashboard": "Latency by Server", + "includeVars": true, + "keepTime": true, + "targetBlank": true, + "title": "Latency by Server", + "type": "dashboard" + } + ], + "minSpan": 6, + "nullPointMode": "null as zero", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [], + "spaceLength": 10, + "span": 6, + "stack": true, + "steppedLine": false, + "targets": [ + { + "refId": "A", + "target": "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.r_await),$percentile), \"Read Latency\")", + "textEditor": true + }, + { + "refId": "C", + "target": "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.w_await),$percentile), \"Write Latency\")", + "textEditor": true + } + ], + "thresholds": [ + { + "colorMode": "custom", + "fill": false, + "line": true, + "lineColor": "rgba(178, 0, 0, 0.29)", + "op": "gt", + "value": 50 + } + ], + "timeFrom": null, + "timeShift": null, + "title": "Disk Latency - $osd_servers OSDs @ $percentile%ile", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "ms", + "label": null, + "logBase": 1, + "max": null, + "min": "0", + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": true + } + ] + }, + { + "aliasColors": { + "disk busy %": "#3F6833" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 1, + "id": 43, + "legend": { + "avg": false, + "current": true, + "max": false, + "min": false, + "show": true, + "total": false, + "values": true + }, + "lines": true, + "linewidth": 1, + "links": [ + { + "dashUri": "db/disk-busy-by-server", + "dashboard": "Disk Busy by Server", + "includeVars": true, + "keepTime": true, + "targetBlank": true, + "title": "Disk Busy by Server", + "type": "dashboard" + } + ], + "minSpan": 6, + "nullPointMode": "null", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [], + "spaceLength": 10, + "span": 6, + "stack": false, + "steppedLine": false, + "targets": [ + { + "refId": "A", + "target": "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.util),$percentile), \"disk busy %\")", + "textEditor": true + } + ], + "thresholds": [ + { + "colorMode": "custom", + "fill": false, + "line": true, + "lineColor": "rgba(178, 0, 0, 0.29)", + "op": "gt", + "value": 80 + } + ], + "timeFrom": null, + "timeShift": null, + "title": "Disk Utilization - $osd_servers OSDs at $percentile%ile", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "short", + "label": "", + "logBase": 1, + "max": "100", + "min": "0", + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": false + } + ] + }, + { + "aliasColors": { + "IOPS/spindle": "#3F6833" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 1, + "id": 46, + "legend": { + "avg": false, + "current": false, + "max": false, + "min": false, + "show": false, + "total": false, + "values": false + }, + "lines": true, + "linewidth": 1, + "links": [], + "minSpan": 6, + "nullPointMode": "null", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [], + "spaceLength": 10, + "span": 6, + "stack": false, + "steppedLine": false, + "targets": [ + { + "refId": "A", + "target": "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.iops),$percentile),\"IOPS/spindle\")", + "textEditor": true + } + ], + "thresholds": [], + "timeFrom": null, + "timeShift": null, + "title": "IOPS per Disk @ $percentile%ile - $osd_servers OSDs", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": "0", + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": true + } + ] + }, + { + "aliasColors": { + "IOPS": "#3F6833" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 1, + "id": 47, + "legend": { + "avg": false, + "current": false, + "max": false, + "min": false, + "show": false, + "total": false, + "values": false + }, + "lines": true, + "linewidth": 1, + "links": [ + { + "dashUri": "db/iops-by-server", + "dashboard": "IOPS by Server", + "includeVars": true, + "keepTime": true, + "targetBlank": true, + "title": "IOPS by Server", + "type": "dashboard" + } + ], + "minSpan": 6, + "nullPointMode": "null", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [], + "spaceLength": 10, + "span": 6, + "stack": false, + "steppedLine": false, + "targets": [ + { + "refId": "A", + "target": "alias(sumSeries(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.iops),\"IOPS\")", + "textEditor": true + } + ], + "thresholds": [], + "timeFrom": null, + "timeShift": null, + "title": "Total Disk IOPS - $osd_servers OSDs", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": "0", + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": false + } + ] + } + ], + "repeat": null, + "repeatIteration": null, + "repeatRowId": null, + "showTitle": true, + "title": "Disk/OSD Load Summary", + "titleSize": "h5" + }, { - "type" : "custom", - "allValue" : null, - "query" : "85", - "current" : { - "text" : "85", - "selected" : true, - "value" : "85" - }, - "label" : null, - "name" : "disk_full_threshold", - "multi" : false, - "options" : [ - { - "text" : "85", - "selected" : true, - "value" : "85" - } - ], - "hide" : 2, - "includeAll" : false + "collapse": false, + "height": 250, + "panels": [ + { + "aliasColors": { + "CPU Busy": "#447EBC", + "CPU Busy @ 95%ile": "#890F02", + "Cluster-wide CPU Busy @ 95%ile": "#890F02", + "Max CPU Busy": "#BF1B00", + "Max CPU Busy - all OSD Hosts": "#BF1B00" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 3, + "id": 44, + "legend": { + "avg": false, + "current": false, + "max": false, + "min": false, + "show": true, + "total": false, + "values": false + }, + "lines": true, + "linewidth": 1, + "links": [], + "minSpan": 6, + "nullPointMode": "null as zero", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [ + { + "alias": "Cluster-wide CPU Busy @ 95%ile", + "fill": 0 + } + ], + "spaceLength": 10, + "span": 6, + "stack": false, + "steppedLine": false, + "targets": [ + { + "refId": "A", + "target": "alias(percentileOfSeries(groupByNode(collectd.*.$domain.cpu.percent.{wait,interrupt,nice,user,system}, 1, \"sumSeries\"),$percentile),\"Cluster-wide CPU Busy @ $percentile%ile\")", + "textEditor": true + }, + { + "hide": false, + "refId": "B", + "target": "alias(averageSeries(groupByNode(collectd.$osd_servers.$domain.cpu.percent.{wait,interrupt,nice,user,system}, 1, \"sumSeries\")),\"Average OSD Host(s) CPU Busy\")", + "textEditor": true + } + ], + "thresholds": [], + "timeFrom": null, + "timeShift": null, + "title": "CPU Utilization - $osd_servers", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "short", + "label": "", + "logBase": 1, + "max": "100", + "min": "0", + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": true + } + ] + }, + { + "aliasColors": { + "Network load (rx+tx)": "#3F6833" + }, + "bars": false, + "dashLength": 10, + "dashes": false, + "datasource": "Local", + "fill": 1, + "id": 49, + "legend": { + "avg": false, + "current": false, + "max": false, + "min": false, + "show": true, + "total": false, + "values": false + }, + "lines": true, + "linewidth": 1, + "links": [ + { + "dashUri": "db/network-usage-by-node", + "dashboard": "Network Usage by Node", + "includeVars": true, + "keepTime": true, + "targetBlank": true, + "title": "Network Load by Node", + "type": "dashboard" + } + ], + "minSpan": 6, + "nullPointMode": "null", + "percentage": false, + "pointradius": 5, + "points": false, + "renderer": "flot", + "seriesOverrides": [], + "spaceLength": 10, + "span": 6, + "stack": false, + "steppedLine": false, + "targets": [ + { + "hide": true, + "refId": "A", + "target": "collectd.$osd_servers.$domain.interface.{en,eth,bond}*.if_octets.rx" + }, + { + "hide": true, + "refId": "B", + "target": "collectd.$osd_servers.$domain.interface.{en,eth,bond}*.if_octets.tx" + }, + { + "refId": "C", + "target": "alias(sumSeries(#A,#B), \"Network load (rx+tx)\")", + "targetFull": "alias(sumSeries(collectd.$osd_servers.$domain.interface.{en,eth,bond}*.if_octets.rx,collectd.$osd_servers.$domain.interface.{en,eth,bond}*.if_octets.tx), \"Network load (rx+tx)\")", + "textEditor": true + } + ], + "thresholds": [], + "timeFrom": null, + "timeShift": null, + "title": "Network Load - $osd_servers", + "tooltip": { + "shared": true, + "sort": 0, + "value_type": "individual" + }, + "type": "graph", + "xaxis": { + "buckets": null, + "mode": "time", + "name": null, + "show": true, + "values": [] + }, + "yaxes": [ + { + "format": "bytes", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": true + }, + { + "format": "short", + "label": null, + "logBase": 1, + "max": null, + "min": null, + "show": true + } + ] + } + ], + "repeat": null, + "repeatIteration": null, + "repeatRowId": null, + "showTitle": true, + "title": "OSD Host CPU and Network Load", + "titleSize": "h5" } - ] - }, - "gnetId" : null, - "style" : "dark", - "hideControls" : true, - "editable" : false, - "annotations" : { - "list" : [] - }, - "graphTooltip" : 0, - "schemaVersion" : 14, - "id" : 13, - "tags" : [], - "rows" : [ - { - "panels" : [ - { - "type" : "table", - "minSpan" : 1, - "transform" : "timeseries_aggregations", - "id" : 56, - "targets" : [ - { - "target" : "currentBelow(groupByNode(transformNull(collectd.*.$domain.cephmetrics.gauge.$cluster_name.mon.osd_state.*.up,99),-2,\"maxSeries\"),0.5)", - "hide" : false, - "refId" : "B", - "textEditor" : true - } - ], - "columns" : [ - { - "value" : "current", - "text" : "Current" - } - ], - "showHeader" : true, - "fontSize" : "100%", - "styles" : [ - { - "thresholds" : [ - "" - ], - "type" : "number", - "sanitize" : false, - "pattern" : "Metric", - "colors" : [ - "rgba(50, 172, 45, 0.97)", - "rgba(237, 129, 40, 0.89)", - "rgba(245, 54, 54, 0.9)" - ], - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colorMode" : null, - "alias" : "OSD Id", - "decimals" : 0, - "unit" : "short" - }, - { - "thresholds" : [ - "0" - ], - "type" : "hidden", - "pattern" : "Current", - "colors" : [ - "rgba(50, 172, 45, 0.97)", - "rgba(237, 129, 40, 0.89)", - "rgba(245, 54, 54, 0.9)" - ], - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colorMode" : "cell", - "alias" : "Status", - "decimals" : 0, - "unit" : "short" - } - ], - "pageSize" : 100, - "scroll" : true, - "title" : "OSD's Down", - "timeFrom" : "1m", - "span" : 1, - "links" : [], - "sort" : { - "col" : null, - "desc" : false - }, - "timeShift" : null, - "hideTimeOverride" : true - }, - { - "type" : "table", - "transform" : "timeseries_aggregations", - "minSpan" : 2, - "styles" : [ - { - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "type" : "date", - "alias" : "Time", - "pattern" : "Time" - }, - { - "type" : "number", - "thresholds" : [], - "pattern" : "Host and Disk", - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null, - "decimals" : 0, - "alias" : "% Full", - "unit" : "short" - }, - { - "unit" : "none", - "alias" : "% Full", - "decimals" : 0, - "colorMode" : null, - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "pattern" : "Current", - "thresholds" : [], - "type" : "number" - }, - { - "type" : "number", - "thresholds" : [], - "pattern" : "Metric", - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null, - "decimals" : 2, - "alias" : "Host.OSD Id", - "unit" : "short" - }, - { - "pattern" : "/.*/", - "type" : "number", - "thresholds" : [], - "decimals" : 2, - "alias" : "", - "unit" : "short", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null - } - ], - "id" : 57, - "pageSize" : null, - "targets" : [ - { - "textEditor" : true, - "refId" : "A", - "target" : "aliasByNode(currentAbove(transformNull(keepLastValue(collectd.*.$domain.cephmetrics.gauge.$cluster_name.osd.*.osd_percent_used),0),$disk_full_threshold),1,-2)" - } - ], - "scroll" : true, - "title" : "Disks Near Full", - "columns" : [ - { - "value" : "current", - "text" : "Current" - } - ], - "sort" : { - "desc" : true, - "col" : 0 - }, - "span" : 2, - "links" : [], - "showHeader" : true, - "fontSize" : "100%" - }, - { - "scroll" : true, - "title" : "Host OSD Raw Capacity", - "links" : [], - "sort" : { - "col" : 0, - "desc" : false - }, - "span" : 3, - "styles" : [ - { - "type" : "date", - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "pattern" : "Time", - "alias" : "Time" - }, - { - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null, - "decimals" : 2, - "alias" : "Total OSD Capacity", - "unit" : "decbytes", - "type" : "number", - "thresholds" : [], - "pattern" : "Current" - }, - { - "thresholds" : [], - "type" : "number", - "pattern" : "Metric", - "colorMode" : null, - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "unit" : "short", - "alias" : "Host Name", - "decimals" : 2 - }, - { - "decimals" : 2, - "alias" : "", - "unit" : "short", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null, - "pattern" : "/.*/", - "thresholds" : [], - "type" : "number" - } - ], - "pageSize" : null, - "targets" : [ - { - "textEditor" : true, - "refId" : "A", - "target" : "groupByNode(keepLastValue(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.stat_bytes),1,\"sumSeries\")" - } - ], - "maxDataPoints" : "1", - "showHeader" : true, - "fontSize" : "100%", - "columns" : [ - { - "value" : "current", - "text" : "Current" - } - ], - "type" : "table", - "minSpan" : 3, - "transform" : "timeseries_aggregations", - "id" : 58 - }, - { - "styles" : [ - { - "pattern" : "Time", - "alias" : "Time", - "type" : "date", - "dateFormat" : "YYYY-MM-DD HH:mm:ss" - }, - { - "type" : "number", - "thresholds" : [], - "pattern" : "Current", - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null, - "alias" : "# OSDs", - "decimals" : 0, - "unit" : "none" - }, - { - "pattern" : "Metric", - "type" : "number", - "thresholds" : [], - "alias" : "Host Name", - "decimals" : 2, - "unit" : "short", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "dateFormat" : "YYYY-MM-DD HH:mm:ss", - "colorMode" : null - }, - { - "decimals" : 2, - "alias" : "", - "unit" : "short", - "colors" : [ - "rgba(245, 54, 54, 0.9)", - "rgba(237, 129, 40, 0.89)", - "rgba(50, 172, 45, 0.97)" - ], - "colorMode" : null, - "pattern" : "/.*/", - "thresholds" : [], - "type" : "number" - } - ], - "pageSize" : null, - "scroll" : true, - "title" : "Host OSD Capacity Breakdown", - "span" : 3, - "sort" : { - "col" : 0, - "desc" : false - }, - "links" : [], - "type" : "table", - "transform" : "timeseries_aggregations", - "minSpan" : 3, - "id" : 59, - "targets" : [ - { - "textEditor" : true, - "refId" : "A", - "target" : "aliasByNode(keepLastValue(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.num_osds,6),1)" - } - ], - "maxDataPoints" : "10", - "fontSize" : "100%", - "showHeader" : true, - "columns" : [ - { - "value" : "current", - "text" : "Current" - } - ] - } - ], - "showTitle" : true, - "repeatRowId" : null, - "titleSize" : "h5", - "collapse" : false, - "repeatIteration" : null, - "title" : "Disk/OSD Host Summary", - "repeat" : null, - "height" : 250 - }, - { - "height" : "300", - "repeatIteration" : null, - "title" : "Disk/OSD Load Summary", - "repeat" : null, - "collapse" : true, - "repeatRowId" : null, - "titleSize" : "h5", - "panels" : [ - { - "heatmap" : {}, - "span" : 6, - "links" : [], - "highlightCards" : true, - "xAxis" : { - "show" : true - }, - "yBucketSize" : 5, - "title" : "Disk Drive Utilization Heatmap - $osd_servers", - "xBucketSize" : "", - "description" : "The heatmap categorizes disk utilization into discrete buckets (e.g util 0-5) and shows the frequency of the number of disks that fall within that range as a color. The color chosen depends on the number of disks in the 'bucket', ranging from green (low) to red (high). Hover over a colored block to show the count of disk utilization observations at that point.", - "cards" : { - "cardRound" : null, - "cardPadding" : null - }, - "yBucketNumber" : null, - "xBucketNumber" : 180, - "dataFormat" : "timeseries", - "targets" : [ - { - "textEditor" : true, - "refId" : "A", - "target" : "collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.util" - } - ], - "yAxis" : { - "format" : "short", - "splitFactor" : null, - "min" : "0", - "decimals" : null, - "max" : "100", - "logBase" : 1, - "show" : true - }, - "id" : 53, - "color" : { - "exponent" : 0.5, - "colorScheme" : "interpolateRdYlGn", - "colorScale" : "sqrt", - "mode" : "spectrum", - "cardColor" : "#b4ff00" - }, - "type" : "heatmap", - "tooltip" : { - "show" : true, - "showHistogram" : false - }, - "minSpan" : 6 - }, - { - "title" : "Disk Throughput - $osd_servers", - "nullPointMode" : "null", - "dashes" : false, - "spaceLength" : 10, - "renderer" : "flot", - "aliasColors" : { - "Write Throughput" : "#E0752D", - "Read Throughput" : "#629E51" - }, - "points" : false, - "timeShift" : null, - "span" : 6, - "linewidth" : 1, - "links" : [], - "timeFrom" : null, - "yaxes" : [ - { - "min" : "0", - "label" : null, - "format" : "decbytes", - "logBase" : 1, - "show" : true, - "max" : null - }, - { - "min" : null, - "label" : null, - "format" : "short", - "show" : true, - "logBase" : 1, - "max" : null - } - ], - "thresholds" : [], - "percentage" : false, - "stack" : true, - "targets" : [ - { - "target" : "alias(sumSeries(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.r_bytes_per_sec), \"Read Throughput\")", - "textEditor" : true, - "refId" : "A" - }, - { - "target" : "alias(sumSeries(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.w_bytes_per_sec), \"Write Throughput\")", - "refId" : "B", - "textEditor" : true - } - ], - "dashLength" : 10, - "datasource" : "Local", - "bars" : false, - "steppedLine" : false, - "fill" : 1, - "pointradius" : 5, - "seriesOverrides" : [], - "legend" : { - "show" : true, - "avg" : false, - "max" : false, - "values" : false, - "total" : false, - "min" : false, - "current" : false - }, - "minSpan" : 6, - "type" : "graph", - "tooltip" : { - "value_type" : "individual", - "shared" : true, - "sort" : 0 - }, - "id" : 24, - "xaxis" : { - "mode" : "time", - "show" : true, - "name" : null, - "buckets" : null, - "values" : [] - }, - "lines" : true - }, - { - "yaxes" : [ - { - "max" : null, - "logBase" : 1, - "show" : true, - "min" : "0", - "format" : "ms", - "label" : null - }, - { - "max" : null, - "logBase" : 1, - "show" : true, - "format" : "short", - "min" : null, - "label" : null - } - ], - "timeFrom" : null, - "links" : [ - { - "keepTime" : true, - "targetBlank" : true, - "type" : "dashboard", - "title" : "Latency by Server", - "dashUri" : "db/latency-by-server", - "includeVars" : true, - "dashboard" : "Latency by Server" - } - ], - "span" : 6, - "linewidth" : 1, - "timeShift" : null, - "points" : false, - "aliasColors" : { - "Read Latency" : "#629E51", - "Write Latency" : "#E0752D" - }, - "dashes" : false, - "spaceLength" : 10, - "renderer" : "flot", - "nullPointMode" : "null as zero", - "title" : "Disk Latency - $osd_servers OSDs @ $percentile%ile", - "percentage" : false, - "thresholds" : [ - { - "value" : 50, - "op" : "gt", - "fill" : false, - "lineColor" : "rgba(178, 0, 0, 0.29)", - "line" : true, - "colorMode" : "custom" - } - ], - "fill" : 1, - "steppedLine" : false, - "datasource" : "Local", - "bars" : false, - "dashLength" : 10, - "targets" : [ - { - "target" : "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.r_await),$percentile), \"Read Latency\")", - "refId" : "A", - "textEditor" : true - }, - { - "textEditor" : true, - "refId" : "C", - "target" : "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.w_await),$percentile), \"Write Latency\")" - } - ], - "stack" : true, - "lines" : true, - "xaxis" : { - "mode" : "time", - "show" : true, - "buckets" : null, - "name" : null, - "values" : [] - }, - "id" : 40, - "tooltip" : { - "shared" : true, - "sort" : 0, - "value_type" : "individual" - }, - "type" : "graph", - "minSpan" : 6, - "legend" : { - "max" : false, - "show" : true, - "avg" : false, - "current" : false, - "total" : false, - "min" : false, - "values" : false - }, - "seriesOverrides" : [], - "pointradius" : 5 - }, - { - "yaxes" : [ - { - "max" : "100", - "show" : true, - "logBase" : 1, - "min" : "0", - "format" : "short", - "label" : "" - }, - { - "logBase" : 1, - "show" : false, - "max" : null, - "min" : null, - "label" : null, - "format" : "short" - } - ], - "timeFrom" : null, - "linewidth" : 1, - "span" : 6, - "links" : [ - { - "includeVars" : true, - "dashUri" : "db/disk-busy-by-server", - "dashboard" : "Disk Busy by Server", - "title" : "Disk Busy by Server", - "type" : "dashboard", - "targetBlank" : true, - "keepTime" : true - } - ], - "timeShift" : null, - "points" : false, - "aliasColors" : { - "disk busy %" : "#3F6833" - }, - "spaceLength" : 10, - "dashes" : false, - "renderer" : "flot", - "title" : "Disk Utilization - $osd_servers OSDs at $percentile%ile", - "nullPointMode" : "null", - "percentage" : false, - "thresholds" : [ - { - "line" : true, - "colorMode" : "custom", - "lineColor" : "rgba(178, 0, 0, 0.29)", - "op" : "gt", - "fill" : false, - "value" : 80 - } - ], - "fill" : 1, - "steppedLine" : false, - "datasource" : "Local", - "bars" : false, - "dashLength" : 10, - "targets" : [ - { - "refId" : "A", - "textEditor" : true, - "target" : "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.util),$percentile), \"disk busy %\")" - } - ], - "stack" : false, - "lines" : true, - "xaxis" : { - "show" : true, - "mode" : "time", - "values" : [], - "name" : null, - "buckets" : null - }, - "id" : 43, - "tooltip" : { - "value_type" : "individual", - "sort" : 0, - "shared" : true - }, - "type" : "graph", - "minSpan" : 6, - "legend" : { - "values" : true, - "total" : false, - "min" : false, - "current" : true, - "show" : true, - "avg" : false, - "max" : false - }, - "pointradius" : 5, - "seriesOverrides" : [] - }, - { - "spaceLength" : 10, - "dashes" : false, - "renderer" : "flot", - "title" : "IOPS per Disk @ $percentile%ile - $osd_servers OSDs", - "nullPointMode" : "null", - "points" : false, - "aliasColors" : { - "IOPS/spindle" : "#3F6833" - }, - "span" : 6, - "links" : [], - "linewidth" : 1, - "timeShift" : null, - "timeFrom" : null, - "yaxes" : [ - { - "max" : null, - "show" : true, - "logBase" : 1, - "format" : "short", - "min" : "0", - "label" : null - }, - { - "label" : null, - "min" : null, - "format" : "short", - "max" : null, - "logBase" : 1, - "show" : true - } - ], - "thresholds" : [], - "percentage" : false, - "dashLength" : 10, - "targets" : [ - { - "target" : "alias(percentileOfSeries(group(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.iops),$percentile),\"IOPS/spindle\")", - "textEditor" : true, - "refId" : "A" - } - ], - "stack" : false, - "steppedLine" : false, - "datasource" : "Local", - "bars" : false, - "fill" : 1, - "legend" : { - "min" : false, - "total" : false, - "current" : false, - "values" : false, - "max" : false, - "show" : false, - "avg" : false - }, - "seriesOverrides" : [], - "pointradius" : 5, - "type" : "graph", - "tooltip" : { - "shared" : true, - "sort" : 0, - "value_type" : "individual" - }, - "minSpan" : 6, - "id" : 46, - "lines" : true, - "xaxis" : { - "values" : [], - "name" : null, - "buckets" : null, - "show" : true, - "mode" : "time" - } - }, - { - "thresholds" : [], - "percentage" : false, - "spaceLength" : 10, - "renderer" : "flot", - "dashes" : false, - "nullPointMode" : "null", - "title" : "Total Disk IOPS - $osd_servers OSDs", - "points" : false, - "aliasColors" : { - "IOPS" : "#3F6833" - }, - "span" : 6, - "links" : [ - { - "dashboard" : "IOPS by Server", - "includeVars" : true, - "dashUri" : "db/iops-by-server", - "title" : "IOPS by Server", - "type" : "dashboard", - "keepTime" : true, - "targetBlank" : true - } - ], - "linewidth" : 1, - "timeShift" : null, - "yaxes" : [ - { - "label" : null, - "min" : "0", - "format" : "short", - "max" : null, - "logBase" : 1, - "show" : true - }, - { - "label" : null, - "min" : null, - "format" : "short", - "max" : null, - "show" : false, - "logBase" : 1 - } - ], - "timeFrom" : null, - "legend" : { - "min" : false, - "total" : false, - "current" : false, - "values" : false, - "max" : false, - "avg" : false, - "show" : false - }, - "seriesOverrides" : [], - "pointradius" : 5, - "type" : "graph", - "tooltip" : { - "shared" : true, - "sort" : 0, - "value_type" : "individual" - }, - "minSpan" : 6, - "id" : 47, - "lines" : true, - "xaxis" : { - "mode" : "time", - "show" : true, - "name" : null, - "buckets" : null, - "values" : [] - }, - "targets" : [ - { - "refId" : "A", - "textEditor" : true, - "target" : "alias(sumSeries(collectd.$osd_servers.$domain.cephmetrics.gauge.$cluster_name.osd.*.perf.iops),\"IOPS\")" - } - ], - "dashLength" : 10, - "stack" : false, - "steppedLine" : false, - "bars" : false, - "datasource" : "Local", - "fill" : 1 - } - ], - "showTitle" : true - }, - { - "panels" : [ - { - "thresholds" : [], - "percentage" : false, - "points" : false, - "aliasColors" : { - "Max CPU Busy - all OSD Hosts" : "#BF1B00", - "CPU Busy" : "#447EBC", - "CPU Busy @ 95%ile" : "#890F02", - "Max CPU Busy" : "#BF1B00", - "Cluster-wide CPU Busy @ 95%ile" : "#890F02" - }, - "dashes" : false, - "spaceLength" : 10, - "renderer" : "flot", - "title" : "CPU Utilization - $osd_servers", - "nullPointMode" : "null as zero", - "timeFrom" : null, - "yaxes" : [ - { - "min" : "0", - "label" : "", - "format" : "short", - "max" : "100", - "logBase" : 1, - "show" : true - }, - { - "format" : "short", - "min" : null, - "label" : null, - "max" : null, - "logBase" : 1, - "show" : true - } - ], - "span" : 6, - "links" : [], - "linewidth" : 1, - "timeShift" : null, - "type" : "graph", - "tooltip" : { - "value_type" : "individual", - "shared" : true, - "sort" : 0 - }, - "minSpan" : 6, - "legend" : { - "values" : false, - "current" : false, - "min" : false, - "total" : false, - "avg" : false, - "show" : true, - "max" : false - }, - "pointradius" : 5, - "seriesOverrides" : [ - { - "fill" : 0, - "alias" : "Cluster-wide CPU Busy @ 95%ile" - } - ], - "lines" : true, - "xaxis" : { - "mode" : "time", - "show" : true, - "buckets" : null, - "name" : null, - "values" : [] - }, - "id" : 44, - "dashLength" : 10, - "targets" : [ - { - "target" : "alias(percentileOfSeries(groupByNode(collectd.*.$domain.cpu.percent.{wait,interrupt,nice,user,system}, 1, \"sumSeries\"),$percentile),\"Cluster-wide CPU Busy @ $percentile%ile\")", - "refId" : "A", - "textEditor" : true - }, - { - "textEditor" : true, - "refId" : "B", - "hide" : false, - "target" : "alias(averageSeries(groupByNode(collectd.$osd_servers.$domain.cpu.percent.{wait,interrupt,nice,user,system}, 1, \"sumSeries\")),\"Average OSD Host(s) CPU Busy\")" - } - ], - "stack" : false, - "fill" : 3, - "steppedLine" : false, - "bars" : false, - "datasource" : "Local" - }, - { - "steppedLine" : false, - "bars" : false, - "datasource" : "Local", - "fill" : 1, - "dashLength" : 10, - "targets" : [ - { - "hide" : true, - "target" : "collectd.$osd_servers.$domain.interface.*.if_octets.rx", - "refId" : "A" - }, - { - "hide" : true, - "target" : "collectd.$osd_servers.$domain.interface.*.if_octets.tx", - "refId" : "B" - }, - { - "refId" : "C", - "textEditor" : true, - "targetFull" : "alias(sumSeries(collectd.$osd_servers.$domain.interface.*.if_octets.rx,collectd.$osd_servers.$domain.interface.*.if_octets.tx), \"Network load (rx+tx)\")", - "target" : "alias(sumSeries(#A,#B), \"Network load (rx+tx)\")" - } - ], - "stack" : false, - "id" : 49, - "lines" : true, - "xaxis" : { - "values" : [], - "name" : null, - "buckets" : null, - "show" : true, - "mode" : "time" - }, - "legend" : { - "avg" : false, - "show" : true, - "max" : false, - "values" : false, - "current" : false, - "total" : false, - "min" : false - }, - "pointradius" : 5, - "seriesOverrides" : [], - "type" : "graph", - "tooltip" : { - "value_type" : "individual", - "sort" : 0, - "shared" : true - }, - "minSpan" : 6, - "links" : [ - { - "dashUri" : "db/network-usage-by-node", - "includeVars" : true, - "dashboard" : "Network Usage by Node", - "title" : "Network Load by Node", - "type" : "dashboard", - "keepTime" : true, - "targetBlank" : true - } - ], - "span" : 6, - "linewidth" : 1, - "timeShift" : null, - "yaxes" : [ - { - "show" : true, - "logBase" : 1, - "max" : null, - "label" : null, - "min" : null, - "format" : "bytes" - }, - { - "show" : true, - "logBase" : 1, - "max" : null, - "label" : null, - "min" : null, - "format" : "short" - } - ], - "timeFrom" : null, - "renderer" : "flot", - "spaceLength" : 10, - "dashes" : false, - "nullPointMode" : "null", - "title" : "Network Load - $osd_servers", - "points" : false, - "aliasColors" : { - "Network load (rx+tx)" : "#3F6833" - }, - "percentage" : false, - "thresholds" : [] - } - ], - "showTitle" : true, - "collapse" : true, - "titleSize" : "h5", - "repeatRowId" : null, - "title" : "OSD Host CPU and Network Load", - "repeat" : null, - "repeatIteration" : null, - "height" : 250 - } - ], - "refresh" : "10s" - }, - "meta" : { - "expires" : "0001-01-01T00:00:00Z", - "updatedBy" : "admin", - "type" : "db", - "canSave" : true, - "canStar" : true, - "canEdit" : true, - "createdBy" : "admin", - "slug" : "ceph-backend-storage", - "updated" : "2017-08-21T03:31:38Z", - "created" : "2017-08-03T21:42:28Z", - "version" : 20 - } -} + ], + "schemaVersion": 14, + "style": "dark", + "tags": [], + "templating": { + "list": [ + { + "allValue": null, + "current": { + "selected": true, + "text": "ceph", + "value": "ceph" + }, + "datasource": "Local", + "hide": 2, + "includeAll": false, + "label": null, + "multi": false, + "name": "cluster_name", + "options": [], + "query": "collectd.*.$domain.cephmetrics.gauge.*", + "refresh": 1, + "regex": "", + "sort": 0, + "tagValuesQuery": "", + "tags": [], + "tagsQuery": "", + "type": "query", + "useTags": false + }, + { + "allValue": null, + "current": { + "selected": true, + "text": "test.lab", + "value": "test.lab" + }, + "hide": 2, + "includeAll": false, + "label": null, + "multi": false, + "name": "domain", + "options": [ + { + "selected": true, + "text": "test.lab", + "value": "test.lab" + } + ], + "query": "test.lab", + "type": "custom" + }, + { + "allValue": null, + "current": { + "selected": true, + "text": "95", + "value": "95" + }, + "hide": 0, + "includeAll": false, + "label": "Percentile", + "multi": false, + "name": "percentile", + "options": [ + { + "selected": false, + "text": "80", + "value": "80" + }, + { + "selected": false, + "text": "85", + "value": "85" + }, + { + "selected": false, + "text": "90", + "value": "90" + }, + { + "selected": true, + "text": "95", + "value": "95" + }, + { + "selected": false, + "text": "98", + "value": "98" + } + ], + "query": "80,85,90,95,98", + "type": "custom" + }, + { + "allValue": "", + "current": { + "selected": true, + "text": "All", + "value": "$__all" + }, + "hide": 0, + "includeAll": true, + "label": "OSD Hostname", + "multi": true, + "name": "osd_servers", + "options": [ + { + "selected": true, + "text": "All", + "value": "$__all" + }, + { + "selected": false, + "text": "ceph-1", + "value": "ceph-1" + }, + { + "selected": false, + "text": "ceph-2", + "value": "ceph-2" + }, + { + "selected": false, + "text": "ceph-3", + "value": "ceph-3" + } + ], + "query": "ceph-1,ceph-2,ceph-3", + "type": "custom" + }, + { + "allValue": null, + "current": { + "selected": true, + "text": "85", + "value": "85" + }, + "hide": 2, + "includeAll": false, + "label": null, + "multi": false, + "name": "disk_full_threshold", + "options": [ + { + "selected": true, + "text": "85", + "value": "85" + } + ], + "query": "85", + "type": "custom" + } + ] + }, + "time": { + "from": "now-1h", + "to": "now" + }, + "timepicker": { + "refresh_intervals": [ + "5s", + "10s", + "30s", + "1m", + "5m", + "15m", + "30m", + "1h", + "2h", + "1d" + ], + "time_options": [ + "5m", + "15m", + "1h", + "6h", + "12h", + "24h", + "2d", + "7d", + "30d" + ] + }, + "timezone": "browser", + "title": "Ceph Backend Storage", + "version": 26 + }, + "meta": { + "canEdit": true, + "canSave": true, + "canStar": true, + "created": "2017-08-03T21:42:28Z", + "createdBy": "admin", + "expires": "0001-01-01T00:00:00Z", + "slug": "ceph-backend-storage", + "type": "db", + "updated": "2017-09-12T23:43:51Z", + "updatedBy": "admin", + "version": 26 + } +} \ No newline at end of file diff --git a/dashboards/current/network-usage-by-node.json b/dashboards/current/network-usage-by-node.json index e63a0b7..d58d308 100644 --- a/dashboards/current/network-usage-by-node.json +++ b/dashboards/current/network-usage-by-node.json @@ -7,7 +7,7 @@ "gnetId": null, "graphTooltip": 0, "hideControls": true, - "id": 73, + "id": 20, "links": [ { "asDropdown": true, @@ -71,17 +71,17 @@ "targets": [ { "refId": "A", - "target": "alias(sumSeries(groupByNode(collectd.$osd_servers.$domain.interface.*.if_octets.*,1,\"sumSeries\")),\"OSD Hosts\")", + "target": "alias(sumSeries(groupByNode(collectd.$osd_servers.$domain.interface.{bond,en,eth}*.if_octets.*,1,\"sumSeries\")),\"OSD Hosts\")", "textEditor": true }, { "refId": "B", - "target": "alias(sumSeries(groupByNode(collectd.$rgw_servers.$domain.interface.*.if_octets.*,1,\"sumSeries\")),\"RadosGW Hosts\")", + "target": "alias(sumSeries(groupByNode(collectd.$rgw_servers.$domain.interface.{bond,en,eth}*.if_octets.*,1,\"sumSeries\")),\"RadosGW Hosts\")", "textEditor": true }, { "refId": "C", - "target": "alias(sumSeries(groupByNode(collectd.$mon_servers.$domain.interface.*.if_octets.*,1,\"sumSeries\")),\"MON Hosts\")", + "target": "alias(sumSeries(groupByNode(collectd.$mon_servers.$domain.interface.{bond,en,eth}*.if_octets.*,1,\"sumSeries\")),\"MON Hosts\")", "textEditor": true } ], @@ -166,7 +166,7 @@ "targets": [ { "refId": "A", - "target": "groupByNode(collectd.$mon_servers.$domain.interface.*.if_octets.*,1,\"sumSeries\")", + "target": "groupByNode(collectd.$mon_servers.$domain.interface.{bond,en,eth}*.if_octets.*,1,\"sumSeries\")", "textEditor": true } ], @@ -251,7 +251,7 @@ "targets": [ { "refId": "A", - "target": "groupByNode(collectd.$osd_servers.$domain.interface.*.if_octets.*,1,\"sumSeries\")", + "target": "groupByNode(collectd.$osd_servers.$domain.interface.{bond,en,eth}*.if_octets.*,1,\"sumSeries\")", "textEditor": true } ], @@ -336,7 +336,7 @@ "targets": [ { "refId": "A", - "target": "groupByNode(collectd.$rgw_servers.$domain.interface.*.if_octets.*,1,\"sumSeries\")", + "target": "groupByNode(collectd.$rgw_servers.$domain.interface.{bond,en,eth}*.if_octets.*,1,\"sumSeries\")", "textEditor": true } ], @@ -412,28 +412,29 @@ }, { "selected": false, - "text": "obj-osd-1", - "value": "obj-osd-1" + "text": "ceph-1", + "value": "ceph-1" }, { "selected": false, - "text": "obj-osd-2", - "value": "obj-osd-2" + "text": "ceph-2", + "value": "ceph-2" }, { "selected": false, - "text": "obj-osd-3", - "value": "obj-osd-3" + "text": "ceph-3", + "value": "ceph-3" } ], - "query": "obj-osd-1,obj-osd-2,obj-osd-3", + "query": "ceph-1,ceph-2,ceph-3", "type": "custom" }, { "allValue": null, "current": { - "text": "storage.lab", - "value": "storage.lab" + "selected": true, + "text": "test.lab", + "value": "test.lab" }, "hide": 2, "includeAll": false, @@ -443,16 +444,17 @@ "options": [ { "selected": true, - "text": "storage.lab", - "value": "storage.lab" + "text": "test.lab", + "value": "test.lab" } ], - "query": "storage.lab", + "query": "test.lab", "type": "custom" }, { "allValue": null, "current": { + "selected": true, "text": "ceph", "value": "ceph" }, @@ -476,6 +478,7 @@ { "allValue": null, "current": { + "selected": false, "text": "All", "value": "$__all" }, @@ -499,8 +502,8 @@ { "allValue": null, "current": { - "text": "obj-rgw-1", - "value": "obj-rgw-1" + "text": "All", + "value": "All" }, "hide": 0, "includeAll": true, @@ -509,7 +512,7 @@ "name": "rgw_servers", "options": [ { - "selected": true, + "selected": false, "text": "obj-rgw-1", "value": "obj-rgw-1" } @@ -550,19 +553,19 @@ }, "timezone": "browser", "title": "Network Usage by Node", - "version": 1 + "version": 5 }, "meta": { "canEdit": true, "canSave": true, "canStar": true, - "created": "2017-08-02T01:47:18Z", - "createdBy": "admin@localhost", + "created": "2017-08-03T21:42:28Z", + "createdBy": "admin", "expires": "0001-01-01T00:00:00Z", "slug": "network-usage-by-node", "type": "db", - "updated": "2017-08-03T04:16:09Z", - "updatedBy": "admin@localhost", - "version": 1 + "updated": "2017-09-12T23:49:40Z", + "updatedBy": "admin", + "version": 5 } } \ No newline at end of file -- 2.47.3