mgr/dashboard: Add ceph_daemon filter to rgw overview grafana panel

queries

Currently rgw_servers filtering is not working in RGW Overview garfana graphs.
It is showing data of all the RGW services, even though filter set to single service.
This PR intends to solve this issue

Fixes: https://tracker.ceph.com/issues/69074

Signed-off-by: Aashish Sharma <aasharma@redhat.com>

3 files changed, 18 insertions, 12 deletions

diff --git a/monitoring/ceph-mixin/dashboards/rgw.libsonnet b/monitoring/ceph-mixin/dashboards/rgw.libsonnet
index 79a4b7a14eb..c0c548b79c8 100644
--- a/monitoring/ceph-mixin/dashboards/rgw.libsonnet
+++ b/monitoring/ceph-mixin/dashboards/rgw.libsonnet
@@ -298,7 +298,7 @@ local g = import 'grafonnet/grafana.libsonnet';
           label_replace(
             rate(ceph_rgw_op_get_obj_lat_sum{%(matchers)s}[$__rate_interval]) /
               rate(ceph_rgw_op_get_obj_lat_count{%(matchers)s}[$__rate_interval]) *
-              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{%(matchers)s},
+              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~"$rgw_servers", %(matchers)s},
             "rgw_host", "$1", "ceph_daemon", "rgw.(.*)"
           )
         ||| % $.matchers(),
@@ -314,7 +314,7 @@ local g = import 'grafonnet/grafana.libsonnet';
               label_replace(
                 rate(ceph_rgw_op_put_obj_lat_sum{%(matchers)s}[$__rate_interval]) /
                   rate(ceph_rgw_op_put_obj_lat_count{%(matchers)s}[$__rate_interval]) *
-                  on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{%(matchers)s},
+                  on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~"$rgw_servers", %(matchers)s},
                 "rgw_host", "$1", "ceph_daemon", "rgw.(.*)"
               )
             ||| % $.matchers(),
@@ -331,7 +331,7 @@ local g = import 'grafonnet/grafana.libsonnet';
           sum by (rgw_host) (
             label_replace(
               rate(ceph_rgw_req{%(matchers)s}[$__rate_interval]) *
-                on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{%(matchers)s},
+                on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~"$rgw_servers", %(matchers)s},
               "rgw_host", "$1", "ceph_daemon", "rgw.(.*)"
             )
           )
@@ -351,7 +351,7 @@ local g = import 'grafonnet/grafana.libsonnet';
           label_replace(
             rate(ceph_rgw_op_get_obj_lat_sum{%(matchers)s}[$__rate_interval]) /
               rate(ceph_rgw_op_get_obj_lat_count{%(matchers)s}[$__rate_interval]) *
-              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{%(matchers)s},
+              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~"$rgw_servers", %(matchers)s},
             "rgw_host", "$1", "ceph_daemon", "rgw.(.*)"
           )
         ||| % $.matchers(),
@@ -385,7 +385,7 @@ local g = import 'grafonnet/grafana.libsonnet';
           label_replace(sum by (instance_id) (
             rate(ceph_rgw_op_get_obj_bytes{%(matchers)s}[$__rate_interval]) +
               rate(ceph_rgw_op_put_obj_bytes{%(matchers)s}[$__rate_interval])) *
-              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{%(matchers)s},
+              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~"$rgw_servers", %(matchers)s},
             "rgw_host", "$1", "ceph_daemon", "rgw.(.*)"
           )
         ||| % $.matchers(),
@@ -404,7 +404,7 @@ local g = import 'grafonnet/grafana.libsonnet';
           label_replace(
             rate(ceph_rgw_op_put_obj_lat_sum{%(matchers)s}[$__rate_interval]) /
               rate(ceph_rgw_op_put_obj_lat_count{%(matchers)s}[$__rate_interval]) *
-              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{%(matchers)s},
+              on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~"$rgw_servers", %(matchers)s},
             "rgw_host", "$1", "ceph_daemon", "rgw.(.*)"
           )
         ||| % $.matchers(),
diff --git a/monitoring/ceph-mixin/dashboards_out/radosgw-overview.json b/monitoring/ceph-mixin/dashboards_out/radosgw-overview.json
index 5e185b63b7f..5bf8279c27c 100644
--- a/monitoring/ceph-mixin/dashboards_out/radosgw-overview.json
+++ b/monitoring/ceph-mixin/dashboards_out/radosgw-overview.json
@@ -108,14 +108,14 @@
          "steppedLine": false,
          "targets": [
             {
-               "expr": "label_replace(\n  rate(ceph_rgw_op_get_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_get_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
+               "expr": "label_replace(\n  rate(ceph_rgw_op_get_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_get_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~\"$rgw_servers\", cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
                "format": "time_series",
                "intervalFactor": 1,
                "legendFormat": "GET {{rgw_host}}",
                "refId": "A"
             },
             {
-               "expr": "label_replace(\n  rate(ceph_rgw_op_put_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_put_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
+               "expr": "label_replace(\n  rate(ceph_rgw_op_put_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_put_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~\"$rgw_servers\", cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
                "format": "time_series",
                "intervalFactor": 1,
                "legendFormat": "PUT {{rgw_host}}",
@@ -210,7 +210,7 @@
          "steppedLine": false,
          "targets": [
             {
-               "expr": "sum by (rgw_host) (\n  label_replace(\n    rate(ceph_rgw_req{cluster=~\"$cluster\", }[$__rate_interval]) *\n      on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{cluster=~\"$cluster\", },\n    \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n  )\n)\n",
+               "expr": "sum by (rgw_host) (\n  label_replace(\n    rate(ceph_rgw_req{cluster=~\"$cluster\", }[$__rate_interval]) *\n      on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~\"$rgw_servers\", cluster=~\"$cluster\", },\n    \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n  )\n)\n",
                "format": "time_series",
                "intervalFactor": 1,
                "legendFormat": "{{rgw_host}}",
@@ -305,7 +305,7 @@
          "steppedLine": false,
          "targets": [
             {
-               "expr": "label_replace(\n  rate(ceph_rgw_op_get_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_get_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
+               "expr": "label_replace(\n  rate(ceph_rgw_op_get_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_get_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~\"$rgw_servers\", cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
                "format": "time_series",
                "intervalFactor": 1,
                "legendFormat": "{{rgw_host}}",
@@ -502,7 +502,7 @@
          "steppedLine": false,
          "targets": [
             {
-               "expr": "label_replace(sum by (instance_id) (\n  rate(ceph_rgw_op_get_obj_bytes{cluster=~\"$cluster\", }[$__rate_interval]) +\n    rate(ceph_rgw_op_put_obj_bytes{cluster=~\"$cluster\", }[$__rate_interval])) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
+               "expr": "label_replace(sum by (instance_id) (\n  rate(ceph_rgw_op_get_obj_bytes{cluster=~\"$cluster\", }[$__rate_interval]) +\n    rate(ceph_rgw_op_put_obj_bytes{cluster=~\"$cluster\", }[$__rate_interval])) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~\"$rgw_servers\", cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
                "format": "time_series",
                "intervalFactor": 1,
                "legendFormat": "{{rgw_host}}",
@@ -597,7 +597,7 @@
          "steppedLine": false,
          "targets": [
             {
-               "expr": "label_replace(\n  rate(ceph_rgw_op_put_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_put_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
+               "expr": "label_replace(\n  rate(ceph_rgw_op_put_obj_lat_sum{cluster=~\"$cluster\", }[$__rate_interval]) /\n    rate(ceph_rgw_op_put_obj_lat_count{cluster=~\"$cluster\", }[$__rate_interval]) *\n    on (instance_id) group_left (ceph_daemon) ceph_rgw_metadata{ceph_daemon=~\"$rgw_servers\", cluster=~\"$cluster\", },\n  \"rgw_host\", \"$1\", \"ceph_daemon\", \"rgw.(.*)\"\n)\n",
                "format": "time_series",
                "intervalFactor": 1,
                "legendFormat": "{{rgw_host}}",
diff --git a/monitoring/ceph-mixin/tests_dashboards/features/radosgw_overview.feature b/monitoring/ceph-mixin/tests_dashboards/features/radosgw_overview.feature
index 8d96dcdd610..a34d5759437 100644
--- a/monitoring/ceph-mixin/tests_dashboards/features/radosgw_overview.feature
+++ b/monitoring/ceph-mixin/tests_dashboards/features/radosgw_overview.feature
@@ -7,6 +7,7 @@ Scenario: "Test Average GET Latencies"
     | ceph_rgw_op_get_obj_lat_count{instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 20 60 80 |
     | ceph_rgw_metadata{ceph_daemon="rgw.foo", hostname="localhost", instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 1 1 1 |
   When interval is `30s`
+  And variable `rgw_servers` is `rgw.foo`
   Then Grafana panel `Average GET/PUT Latencies by RGW Instance` with legend `GET {{rgw_host}}` shows:
     | metrics | values |
     | {ceph_daemon="rgw.foo", instance="127.0.0.1", instance_id="58892247", job="ceph", rgw_host="foo", cluster="mycluster"} | 1.5 |
@@ -18,6 +19,7 @@ Scenario: "Test Average PUT Latencies"
     | ceph_rgw_op_put_obj_lat_count{instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 10 30 50 |
     | ceph_rgw_metadata{ceph_daemon="rgw.foo", hostname="localhost", instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 1 1 1 |
   When interval is `30s`
+  And variable `rgw_servers` is `rgw.foo`
   Then Grafana panel `Average GET/PUT Latencies by RGW Instance` with legend `PUT {{rgw_host}}` shows:
     | metrics | values |
     | {ceph_daemon="rgw.foo", instance="127.0.0.1", instance_id="58892247", job="ceph", rgw_host="foo", cluster="mycluster"} | 1 |
@@ -28,6 +30,7 @@ Scenario: "Test Total Requests/sec by RGW Instance"
     | ceph_rgw_req{instance="127.0.0.1", instance_id="92806566", job="ceph", cluster="mycluster"} | 10 50 100 |
     | ceph_rgw_metadata{ceph_daemon="rgw.1", hostname="localhost", instance="127.0.0.1", instance_id="92806566", job="ceph", cluster="mycluster"} | 1 1 1 |
   When interval is `30s`
+  And variable `rgw_servers` is `rgw.1`
   Then Grafana panel `Total Requests/sec by RGW Instance` with legend `{{rgw_host}}` shows:
     | metrics | values |
     | {rgw_host="1"} | 1.5 |
@@ -39,6 +42,7 @@ Scenario: "Test GET Latencies by RGW Instance"
     | ceph_rgw_op_get_obj_lat_count{instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 20 60 80 |
     | ceph_rgw_metadata{ceph_daemon="rgw.foo", hostname="localhost", instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 1 1 1 |
   When interval is `30s`
+  And variable `rgw_servers` is `rgw.foo`
   Then Grafana panel `GET Latencies by RGW Instance` with legend `{{rgw_host}}` shows:
     | metrics | values |
     | {ceph_daemon="rgw.foo", instance="127.0.0.1", instance_id="58892247", job="ceph", rgw_host="foo", cluster="mycluster"} | 1.5 |
@@ -71,6 +75,7 @@ Scenario: "Test Bandwidth by RGW Instance"
     | ceph_rgw_metadata{ceph_daemon="rgw.1", hostname="localhost", instance="127.0.0.1", instance_id="92806566", job="ceph", cluster="mycluster"} | 1 1 1 |
   When evaluation time is `1m`
   And interval is `30s`
+  And variable `rgw_servers` is `rgw.1`
   Then Grafana panel `Bandwidth by RGW Instance` with legend `{{rgw_host}}` shows:
     | metrics | values |
     | {ceph_daemon="rgw.1", instance_id="92806566", rgw_host="1"} | 2.25 |
@@ -83,6 +88,7 @@ Scenario: "Test PUT Latencies by RGW Instance"
     | ceph_rgw_metadata{ceph_daemon="rgw.foo", hostname="localhost", instance="127.0.0.1", instance_id="58892247", job="ceph", cluster="mycluster"} | 1 1 1 |
   When evaluation time is `1m`
   And interval is `30s`
+  And variable `rgw_servers` is `rgw.foo`
   Then Grafana panel `PUT Latencies by RGW Instance` with legend `{{rgw_host}}` shows:
     | metrics | values |
     | {ceph_daemon="rgw.foo", instance="127.0.0.1", instance_id="58892247", job="ceph", rgw_host="foo", cluster="mycluster"} | 1 |