From 2df3ce1902b4c70846372ba271428d2e333fbbee Mon Sep 17 00:00:00 2001 From: cloudbehl Date: Fri, 29 Mar 2024 13:18:50 +0530 Subject: [PATCH] monitoring/ceph-mixin: Add cluster variable to ceph-cluster.json Fixes: https://tracker.ceph.com/issues/65218 Signed-off-by: cloudbehl --- .../dashboards_out/ceph-cluster.json | 80 ++++++++++++------- 1 file changed, 51 insertions(+), 29 deletions(-) diff --git a/monitoring/ceph-mixin/dashboards_out/ceph-cluster.json b/monitoring/ceph-mixin/dashboards_out/ceph-cluster.json index 95ff659e3977b..3bd150e9ae263 100644 --- a/monitoring/ceph-mixin/dashboards_out/ceph-cluster.json +++ b/monitoring/ceph-mixin/dashboards_out/ceph-cluster.json @@ -101,7 +101,7 @@ "tableColumn": "", "targets": [ { - "expr": "ceph_health_status", + "expr": "ceph_health_status{cluster=~'$cluster'}", "format": "time_series", "instant": true, "interval": "$interval", @@ -182,7 +182,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "count(ceph_osd_metadata)", + "expr": "count(ceph_osd_metadata{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "All", @@ -197,7 +197,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "sum(ceph_osd_in)", + "expr": "sum(ceph_osd_in{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "In", @@ -212,7 +212,7 @@ "displayAliasType": "Warning / Critical", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "sum(ceph_osd_in == bool 0)", + "expr": "sum(ceph_osd_in{cluster=~'$cluster'} == bool 0)", "format": "time_series", "interval": "", "intervalFactor": 1, @@ -229,7 +229,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "sum(ceph_osd_up)", + "expr": "sum(ceph_osd_up{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Up", @@ -245,7 +245,7 @@ "displayAliasType": "Warning / Critical", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "sum(ceph_osd_up == bool 0)", + "expr": "sum(ceph_osd_up{cluster=~'$cluster'} == bool 0)", "format": "time_series", "intervalFactor": 1, "legendFormat": "Down", @@ -392,7 +392,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "sum(ceph_mon_quorum_status)", + "expr": "sum(ceph_mon_quorum_status{cluster=~'$cluster'})", "format": "time_series", "interval": "", "intervalFactor": 1, @@ -409,7 +409,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "count(ceph_mon_quorum_status)", + "expr": "count(ceph_mon_quorum_status{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Total", @@ -426,7 +426,7 @@ "displayAliasType": "Warning / Critical", "displayType": "Annotation", "displayValueWithAlias": "Never", - "expr": "count(ceph_mon_quorum_status) - sum(ceph_mon_quorum_status)", + "expr": "count(ceph_mon_quorum_status{cluster=~'$cluster'}) - sum(ceph_mon_quorum_status{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "MONs out of Quorum", @@ -486,7 +486,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "count(ceph_mgr_status == 1) or vector(0)", + "expr": "count(ceph_mgr_status{cluster=~'$cluster'} == 1) or vector(0)", "format": "time_series", "intervalFactor": 1, "instant": true, @@ -502,7 +502,7 @@ "displayAliasType": "Always", "displayType": "Regular", "displayValueWithAlias": "When Alias Displayed", - "expr": "count(ceph_mgr_status == 0) or vector(0)", + "expr": "count(ceph_mgr_status{cluster=~'$cluster'} == 0) or vector(0)", "format": "time_series", "instant": true, "intervalFactor": 1, @@ -577,7 +577,7 @@ "tableColumn": "", "targets": [ { - "expr": "sum(ceph_osd_stat_bytes_used)/sum(ceph_osd_stat_bytes)", + "expr": "sum(ceph_osd_stat_bytes_used{cluster=~'$cluster'})/sum(ceph_osd_stat_bytes{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Used", @@ -677,49 +677,49 @@ "steppedLine": false, "targets": [ { - "expr": "sum(ceph_pg_total)", + "expr": "sum(ceph_pg_total{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Total", "refId": "A" }, { - "expr": "sum(ceph_pg_active)", + "expr": "sum(ceph_pg_active{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Active", "refId": "B" }, { - "expr": "sum(ceph_pg_total - ceph_pg_active)", + "expr": "sum(ceph_pg_total{cluster=~'$cluster'} - ceph_pg_active{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Inactive", "refId": "G" }, { - "expr": "sum(ceph_pg_undersized)", + "expr": "sum(ceph_pg_undersized{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Undersized", "refId": "F" }, { - "expr": "sum(ceph_pg_degraded)", + "expr": "sum(ceph_pg_degraded{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Degraded", "refId": "C" }, { - "expr": "sum(ceph_pg_inconsistent)", + "expr": "sum(ceph_pg_inconsistent{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Inconsistent", "refId": "D" }, { - "expr": "sum(ceph_pg_down)", + "expr": "sum(ceph_pg_down{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Down", @@ -813,28 +813,28 @@ "steppedLine": false, "targets": [ { - "expr": "quantile(0.95, ceph_osd_apply_latency_ms)", + "expr": "quantile(0.95, ceph_osd_apply_latency_ms{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Apply Latency P_95", "refId": "A" }, { - "expr": "quantile(0.95, ceph_osd_commit_latency_ms)", + "expr": "quantile(0.95, ceph_osd_commit_latency_ms{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Commit Latency P_95", "refId": "B" }, { - "expr": "avg(ceph_osd_apply_latency_ms)", + "expr": "avg(ceph_osd_apply_latency_ms{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Avg Apply Latency", "refId": "C" }, { - "expr": "avg(ceph_osd_commit_latency_ms)", + "expr": "avg(ceph_osd_commit_latency_ms{cluster=~'$cluster'})", "format": "time_series", "intervalFactor": 1, "legendFormat": "Avg Commit Latency", @@ -928,14 +928,14 @@ "steppedLine": false, "targets": [ { - "expr": "sum(irate(ceph_osd_op_w_in_bytes[1m]))", + "expr": "sum(irate(ceph_osd_op_w_in_bytes{cluster=~'$cluster'}[1m]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "Writes", "refId": "A" }, { - "expr": "sum(irate(ceph_osd_op_r_out_bytes[1m]))", + "expr": "sum(irate(ceph_osd_op_r_out_bytes{cluster=~'$cluster'}[1m]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "Reads", @@ -1024,7 +1024,7 @@ "steppedLine": false, "targets": [ { - "expr": "sum(deriv(ceph_pool_stored[1m]))", + "expr": "sum(deriv(ceph_pool_stored{cluster=~'$cluster'}[1m]))", "format": "time_series", "intervalFactor": 1, "refId": "A" @@ -1097,7 +1097,7 @@ "span": 12, "targets": [ { - "expr": "ceph_osd_stat_bytes_used / ceph_osd_stat_bytes", + "expr": "ceph_osd_stat_bytes_used{cluster=~'$cluster'} / ceph_osd_stat_bytes{cluster=~'$cluster'}", "format": "time_series", "interval": "1m", "intervalFactor": 1, @@ -1159,7 +1159,7 @@ "links": [], "targets": [ { - "expr": "ceph_osd_numpg", + "expr": "ceph_osd_numpg{cluster=~'$cluster'}", "format": "time_series", "intervalFactor": 1, "legendFormat": "#PGs", @@ -1235,7 +1235,7 @@ "steppedLine": false, "targets": [ { - "expr": "sum(irate(ceph_osd_recovery_ops[1m]))", + "expr": "sum(irate(ceph_osd_recovery_ops{cluster=~'$cluster'}[1m]))", "format": "time_series", "intervalFactor": 1, "legendFormat": "Op/s", @@ -1298,6 +1298,28 @@ "regex": "", "type": "datasource" }, + { + "datasource": { + "type": "prometheus", + "uid": "${datasource}" + }, + "definition": "label_values(ceph_health_status, cluster)", + "hide": 0, + "includeAll": false, + "label": "Cluster", + "multi": false, + "name": "cluster", + "options": [], + "query": { + "query": "label_values(ceph_health_status, cluster)", + "refId": "StandardVariableQuery" + }, + "refresh": 1, + "regex": "", + "skipUrlSync": false, + "sort": 0, + "type": "query" + }, { "auto": true, "auto_count": 10, -- 2.39.5