From: Boris Ranto <branto@redhat.com>
Date: Tue, 19 Nov 2019 22:17:47 +0000 (+0100)
Subject: dashboards: Ignore wal_device label
X-Git-Tag: v2.0.9^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fpull%2F249%2Fhead;p=cephmetrics.git

dashboards: Ignore wal_device label

The wal_device label was added to ceph_disk_occupation. We need to
ignore it in these queries to provide proper matching between values.
Otherwise, the query won't return any data. This is
backwards-compatible, if you ignore a non-existing label, nothing will
change.

Signed-off-by: Boris Ranto <branto@redhat.com>
---

diff --git a/dashboards/mgr-prometheus/ceph-at-a-glance.json b/dashboards/mgr-prometheus/ceph-at-a-glance.json
index bbf59c4..37edab6 100644
--- a/dashboards/mgr-prometheus/ceph-at-a-glance.json
+++ b/dashboards/mgr-prometheus/ceph-at-a-glance.json
@@ -2599,7 +2599,7 @@
       "tableColumn": "",
       "targets": [
         {
-          "expr": "sum(\n  sum(\n    rate(\n      node_disk_reads_completed[$__interval]\n    )\n  ) by (instance,device)\n   + ignoring(ceph_daemon,job) group_right(instance) ceph_disk_occupation\n)\n+\nsum(\n  sum(\n    rate(\n      node_disk_writes_completed[$__interval]\n    )\n  ) by (instance,device)\n   + ignoring(ceph_daemon,job) group_right(instance) ceph_disk_occupation\n)",
+          "expr": "sum(\n  sum(\n    rate(\n      node_disk_reads_completed[$__interval]\n    )\n  ) by (instance,device)\n   + ignoring(ceph_daemon,job,wal_device) group_right(instance) ceph_disk_occupation\n)\n+\nsum(\n  sum(\n    rate(\n      node_disk_writes_completed[$__interval]\n    )\n  ) by (instance,device)\n   + ignoring(ceph_daemon,job,wal_device) group_right(instance) ceph_disk_occupation\n)",
           "format": "time_series",
           "groupBy": [],
           "hide": false,
@@ -2708,7 +2708,7 @@
       "tableColumn": "",
       "targets": [
         {
-          "expr": "# should only include OSD hosts\nsum(\n  sum(\n    rate(\n      node_disk_bytes_read[$__interval]\n    )\n  ) by (instance,device)\n  + ignoring(ceph_daemon,job) group_right(instance) ceph_disk_occupation\n)\n+\nsum(\n  sum(\n    rate(\n      node_disk_bytes_written[$__interval]\n    )\n  ) by (instance,device)\n  + ignoring(ceph_daemon,job) group_right(instance) ceph_disk_occupation\n)",
+          "expr": "# should only include OSD hosts\nsum(\n  sum(\n    rate(\n      node_disk_bytes_read[$__interval]\n    )\n  ) by (instance,device)\n  + ignoring(ceph_daemon,job,wal_device) group_right(instance) ceph_disk_occupation\n)\n+\nsum(\n  sum(\n    rate(\n      node_disk_bytes_written[$__interval]\n    )\n  ) by (instance,device)\n  + ignoring(ceph_daemon,job,wal_device) group_right(instance) ceph_disk_occupation\n)",
           "format": "time_series",
           "groupBy": [],
           "hide": false,
@@ -2934,21 +2934,21 @@
       "tableColumn": "",
       "targets": [
         {
-          "expr": "quantile(\n  5 / 100,\n    (irate(node_disk_read_time_ms[1m]) + irate(node_disk_write_time_ms[1m]) / \n    (irate(node_disk_reads_completed[1m]) + irate(node_disk_writes_completed[1m])) \n    +\n    ignoring(ceph_daemon,job) ceph_disk_occupation))",
+          "expr": "quantile(\n  5 / 100,\n    (irate(node_disk_read_time_ms[1m]) + irate(node_disk_write_time_ms[1m]) / \n    (irate(node_disk_reads_completed[1m]) + irate(node_disk_writes_completed[1m])) \n    +\n    ignoring(ceph_daemon,job,wal_device) ceph_disk_occupation))",
           "format": "time_series",
           "hide": false,
           "intervalFactor": 1,
           "refId": "C"
         },
         {
-          "expr": "max(\n  irate(node_disk_read_time_ms[30s]) / irate(node_disk_reads_completed[30s])\n  +\n  irate(node_disk_write_time_ms[30s]) / irate(node_disk_writes_completed[30s])\n  + ignoring(ceph_daemon,job) ceph_disk_occupation\n)",
+          "expr": "max(\n  irate(node_disk_read_time_ms[30s]) / irate(node_disk_reads_completed[30s])\n  +\n  irate(node_disk_write_time_ms[30s]) / irate(node_disk_writes_completed[30s])\n  + ignoring(ceph_daemon,job,wal_device) ceph_disk_occupation\n)",
           "format": "time_series",
           "hide": true,
           "intervalFactor": 2,
           "refId": "B"
         },
         {
-          "expr": "sum(\n    (irate(node_disk_read_time_ms[1m]) + irate(node_disk_write_time_ms[1m]) / \n    (irate(node_disk_reads_completed[1m]) + irate(node_disk_writes_completed[1m])) \n    +\n    ignoring(ceph_daemon,job) ceph_disk_occupation)) / count(ceph_osd_up)",
+          "expr": "sum(\n    (irate(node_disk_read_time_ms[1m]) + irate(node_disk_write_time_ms[1m]) / \n    (irate(node_disk_reads_completed[1m]) + irate(node_disk_writes_completed[1m])) \n    +\n    ignoring(ceph_daemon,job,wal_device) ceph_disk_occupation)) / count(ceph_osd_up)",
           "format": "time_series",
           "hide": true,
           "intervalFactor": 1,
diff --git a/dashboards/mgr-prometheus/iops-by-server.json b/dashboards/mgr-prometheus/iops-by-server.json
index b4da67b..eb6db6b 100644
--- a/dashboards/mgr-prometheus/iops-by-server.json
+++ b/dashboards/mgr-prometheus/iops-by-server.json
@@ -93,7 +93,7 @@
               "refId": "B"
             },
             {
-              "expr": "sum(\n  sum(\n    irate(node_disk_reads_completed{job=\"node\" }[1m]) + \n    irate(node_disk_writes_completed{job=\"node\"}[1m]))\n  by(instance, device) + ignoring(ceph_daemon,job) group_right(instance) ceph_disk_occupation)\n  by(instance)",
+              "expr": "sum(\n  sum(\n    irate(node_disk_reads_completed{job=\"node\" }[1m]) + \n    irate(node_disk_writes_completed{job=\"node\"}[1m]))\n  by(instance, device) + ignoring(ceph_daemon,job,wal_device) group_right(instance) ceph_disk_occupation)\n  by(instance)",
               "format": "time_series",
               "intervalFactor": 1,
               "legendFormat": "{{instance}}",
@@ -181,7 +181,7 @@
           "steppedLine": false,
           "targets": [
             {
-              "expr": "sum(\n  sum(\n    irate(node_disk_reads_completed{instance=~\"[[osd_servers]].*\"}[1m]) + irate(node_disk_writes_completed[1m]))\n  by(instance,device) +\n  ignoring(ceph_daemon,job) group_right(instance) ceph_disk_occupation)\n    \n  \n",
+              "expr": "sum(\n  sum(\n    irate(node_disk_reads_completed{instance=~\"[[osd_servers]].*\"}[1m]) + irate(node_disk_writes_completed[1m]))\n  by(instance,device) +\n  ignoring(ceph_daemon,job,wal_device) group_right(instance) ceph_disk_occupation)\n    \n  \n",
               "format": "time_series",
               "intervalFactor": 2,
               "refId": "A"
diff --git a/dashboards/mgr-prometheus/latency-by-server.json b/dashboards/mgr-prometheus/latency-by-server.json
index 7e9510f..61260e5 100644
--- a/dashboards/mgr-prometheus/latency-by-server.json
+++ b/dashboards/mgr-prometheus/latency-by-server.json
@@ -90,7 +90,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "max(\n    (irate(node_disk_read_time_ms[30s]) + irate(node_disk_write_time_ms[30s]) / \n    (irate(node_disk_reads_completed[30s]) + irate(node_disk_writes_completed[30s])) +\n    ignoring(ceph_daemon,job) ceph_disk_occupation))\n    by(instance)",
+          "expr": "max(\n    (irate(node_disk_read_time_ms[30s]) + irate(node_disk_write_time_ms[30s]) / \n    (irate(node_disk_reads_completed[30s]) + irate(node_disk_writes_completed[30s])) +\n    ignoring(ceph_daemon,job,wal_device) ceph_disk_occupation))\n    by(instance)",
           "format": "time_series",
           "intervalFactor": 1,
           "legendFormat": "{{instance}}",