1
0
Fork 0
mirror of https://github.com/monitoring-mixins/website.git synced 2024-12-14 11:37:31 +00:00

assets,site/content: daily assets regeneration

This commit is contained in:
github-actions[bot] 2022-10-07 04:13:56 +00:00
parent 7b627a85d2
commit 3548e267c8
10 changed files with 511 additions and 57 deletions

View file

@ -560,6 +560,7 @@ groups:
summary: Client certificate is about to expire.
expr: |
apiserver_client_certificate_expiration_seconds_count{job="kube-apiserver"} > 0 and on(job) histogram_quantile(0.01, sum by (job, le) (rate(apiserver_client_certificate_expiration_seconds_bucket{job="kube-apiserver"}[5m]))) < 604800
for: 5m
labels:
severity: warning
- alert: KubeClientCertificateExpiration
@ -570,6 +571,7 @@ groups:
summary: Client certificate is about to expire.
expr: |
apiserver_client_certificate_expiration_seconds_count{job="kube-apiserver"} > 0 and on(job) histogram_quantile(0.01, sum by (job, le) (rate(apiserver_client_certificate_expiration_seconds_bucket{job="kube-apiserver"}[5m]))) < 86400
for: 5m
labels:
severity: critical
- alert: KubeAggregatedAPIErrors

View file

@ -58,7 +58,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(loki_ingester_memory_chunks{cluster=\"$cluster\", job=~\"$namespace/ingester\"})",
"expr": "sum(loki_ingester_memory_chunks{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "series",
@ -134,7 +134,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(loki_ingester_memory_chunks{cluster=\"$cluster\", job=~\"$namespace/ingester\"}) / sum(loki_ingester_memory_streams{cluster=\"$cluster\", job=~\"$namespace/ingester\"})",
"expr": "sum(loki_ingester_memory_chunks{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}) / sum(loki_ingester_memory_streams{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "chunks",
@ -222,7 +222,7 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)) * 1",
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)) * 1",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "99th Percentile",
@ -230,7 +230,7 @@
"step": 10
},
{
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)) * 1",
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)) * 1",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "50th Percentile",
@ -238,7 +238,7 @@
"step": 10
},
{
"expr": "sum(rate(loki_ingester_chunk_utilization_sum{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) * 1 / sum(rate(loki_ingester_chunk_utilization_count{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]))",
"expr": "sum(rate(loki_ingester_chunk_utilization_sum{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) * 1 / sum(rate(loki_ingester_chunk_utilization_count{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "Average",
@ -314,7 +314,7 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_age_seconds_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)) * 1e3",
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_age_seconds_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)) * 1e3",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "99th Percentile",
@ -322,7 +322,7 @@
"step": 10
},
{
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_age_seconds_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)) * 1e3",
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_age_seconds_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)) * 1e3",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "50th Percentile",
@ -330,7 +330,7 @@
"step": 10
},
{
"expr": "sum(rate(loki_ingester_chunk_age_seconds_sum{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) * 1e3 / sum(rate(loki_ingester_chunk_age_seconds_count{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]))",
"expr": "sum(rate(loki_ingester_chunk_age_seconds_sum{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) * 1e3 / sum(rate(loki_ingester_chunk_age_seconds_count{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "Average",
@ -418,7 +418,7 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_entries_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)) * 1",
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_entries_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)) * 1",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "99th Percentile",
@ -426,7 +426,7 @@
"step": 10
},
{
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_entries_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)) * 1",
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_entries_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)) * 1",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "50th Percentile",
@ -434,7 +434,7 @@
"step": 10
},
{
"expr": "sum(rate(loki_ingester_chunk_entries_sum{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) * 1 / sum(rate(loki_ingester_chunk_entries_count{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]))",
"expr": "sum(rate(loki_ingester_chunk_entries_sum{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) * 1 / sum(rate(loki_ingester_chunk_entries_count{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "Average",
@ -510,7 +510,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_chunk_store_index_entries_per_chunk_sum{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[5m])) / sum(rate(loki_chunk_store_index_entries_per_chunk_count{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[5m]))",
"expr": "sum(rate(loki_chunk_store_index_entries_per_chunk_sum{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[5m])) / sum(rate(loki_chunk_store_index_entries_per_chunk_count{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[5m]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "Index Entries",
@ -598,7 +598,7 @@
"steppedLine": false,
"targets": [
{
"expr": "cortex_ingester_flush_queue_length{cluster=\"$cluster\", job=~\"$namespace/ingester\"}",
"expr": "cortex_ingester_flush_queue_length{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{pod}}",
@ -682,7 +682,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_ingester_chunk_age_seconds_count{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_ingester_chunk_age_seconds_count{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{status}}",
@ -770,7 +770,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]))",
"expr": "sum(rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{pod}}",
@ -846,7 +846,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by (reason) (rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) / ignoring(reason) group_left sum(rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]))",
"expr": "sum by (reason) (rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) / ignoring(reason) group_left sum(rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{reason}}",
@ -926,7 +926,7 @@
"span": 12,
"targets": [
{
"expr": "sum by (le) (rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval]))",
"expr": "sum by (le) (rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval]))",
"format": "heatmap",
"intervalFactor": 2,
"legendFormat": "{{le}}",
@ -988,7 +988,7 @@
"span": 12,
"targets": [
{
"expr": "sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[$__rate_interval])) by (le)",
"expr": "sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[$__rate_interval])) by (le)",
"format": "heatmap",
"intervalFactor": 2,
"legendFormat": "{{le}}",
@ -1058,7 +1058,7 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[1m])) by (le))",
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[1m])) by (le))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "p99",
@ -1066,7 +1066,7 @@
"step": 10
},
{
"expr": "histogram_quantile(0.90, sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[1m])) by (le))",
"expr": "histogram_quantile(0.90, sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[1m])) by (le))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "p90",
@ -1074,7 +1074,7 @@
"step": 10
},
{
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[1m])) by (le))",
"expr": "histogram_quantile(0.50, sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[1m])) by (le))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "p50",
@ -1162,7 +1162,7 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.5, sum(rate(loki_ingester_chunk_bounds_hours_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[5m])) by (le))",
"expr": "histogram_quantile(0.5, sum(rate(loki_ingester_chunk_bounds_hours_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[5m])) by (le))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "p50",
@ -1170,7 +1170,7 @@
"step": 10
},
{
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_bounds_hours_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[5m])) by (le))",
"expr": "histogram_quantile(0.99, sum(rate(loki_ingester_chunk_bounds_hours_bucket{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[5m])) by (le))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "p99",
@ -1178,7 +1178,7 @@
"step": 10
},
{
"expr": "sum(rate(loki_ingester_chunk_bounds_hours_sum{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[5m])) / sum(rate(loki_ingester_chunk_bounds_hours_count{cluster=\"$cluster\", job=~\"$namespace/ingester\"}[5m]))",
"expr": "sum(rate(loki_ingester_chunk_bounds_hours_sum{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[5m])) / sum(rate(loki_ingester_chunk_bounds_hours_count{cluster=\"$cluster\", job=~\"$namespace/ingester.*\"}[5m]))",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "avg",

View file

@ -6,7 +6,6 @@
"gnetId": null,
"graphTooltip": 0,
"hideControls": false,
"id": 8,
"iteration": 1583185057230,
"links": [
{

View file

@ -607,6 +607,82 @@
"refresh": 1,
"regex": "",
"type": "datasource"
},
{
"allValue": null,
"current": {
"text": "prod",
"value": "prod"
},
"datasource": "$datasource",
"hide": 0,
"includeAll": false,
"label": "cluster",
"multi": false,
"name": "cluster",
"options": [ ],
"query": "label_values(loki_build_info, cluster)",
"refresh": 1,
"regex": "",
"sort": 2,
"tagValuesQuery": "",
"tags": [ ],
"tagsQuery": "",
"type": "query",
"useTags": false
},
{
"allValue": null,
"current": {
"text": "prod",
"value": "prod"
},
"datasource": "$datasource",
"hide": 0,
"includeAll": false,
"label": "namespace",
"multi": false,
"name": "namespace",
"options": [ ],
"query": "label_values(loki_build_info{cluster=~\"$cluster\"}, namespace)",
"refresh": 1,
"regex": "",
"sort": 2,
"tagValuesQuery": "",
"tags": [ ],
"tagsQuery": "",
"type": "query",
"useTags": false
},
{
"hide": 0,
"label": null,
"name": "loki_datasource",
"options": [ ],
"query": "loki",
"refresh": 1,
"regex": "",
"type": "datasource"
},
{
"allValue": ".+",
"current": { },
"datasource": "$datasource",
"hide": 0,
"includeAll": true,
"label": null,
"multi": false,
"name": "tenant",
"options": [ ],
"query": "query_result(sum by (id) (grafanacloud_logs_instance_info) and sum(label_replace(loki_tenant:active_streams{cluster=\"$cluster\",namespace=\"$namespace\"},\"id\",\"$1\",\"tenant\",\"(.*)\")) by(id))",
"refresh": 0,
"regex": "/\"([^\"]+)\"/",
"sort": 1,
"tagValuesQuery": "",
"tags": [ ],
"tagsQuery": "",
"type": "query",
"useTags": false
}
]
},

View file

@ -6,7 +6,6 @@
"gnetId": null,
"graphTooltip": 0,
"hideControls": false,
"id": 68,
"iteration": 1588704280892,
"links": [
{
@ -973,18 +972,18 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"expr": "histogram_quantile(0.99, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"legendFormat": ".99",
"refId": "A"
},
{
"expr": "histogram_quantile(0.9, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"expr": "histogram_quantile(0.9, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"hide": false,
"legendFormat": ".9",
"refId": "B"
},
{
"expr": "histogram_quantile(0.5, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"expr": "histogram_quantile(0.5, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"hide": false,
"legendFormat": ".5",
"refId": "C"
@ -1081,7 +1080,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\", status_code!~\"5[0-9]{2}\", route=\"/logproto.Pusher/Push\"}[5m])) by (route)\n/\nsum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\"}[5m])) by (route)",
"expr": "sum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\", status_code!~\"5[0-9]{2}\", route=\"/logproto.Pusher/Push\"}[5m])) by (route)\n/\nsum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\", route=\"/logproto.Pusher/Push\"}[5m])) by (route)",
"interval": "",
"legendFormat": "{{route}}",
"refId": "A"
@ -1492,17 +1491,17 @@
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=\"$cluster\"})) * 1e3",
"expr": "histogram_quantile(0.99, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=\"$cluster\"})) * 1e3",
"legendFormat": ".99-{{route}}",
"refId": "A"
},
{
"expr": "histogram_quantile(0.9, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=\"$cluster\"})) * 1e3",
"expr": "histogram_quantile(0.9, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=\"$cluster\"})) * 1e3",
"legendFormat": ".9-{{route}}",
"refId": "B"
},
{
"expr": "histogram_quantile(0.5, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=\"$cluster\"})) * 1e3",
"expr": "histogram_quantile(0.5, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=\"$cluster\"})) * 1e3",
"legendFormat": ".5-{{route}}",
"refId": "C"
}
@ -1598,7 +1597,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\", status_code!~\"5[0-9]{2}\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\"}[5m])) by (route)\n/\nsum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\"}[5m])) by (route)",
"expr": "sum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\", status_code!~\"5[0-9]{2}\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\"}[5m])) by (route)\n/\nsum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\"}[5m])) by (route)",
"interval": "",
"legendFormat": "{{route}}",
"refId": "A"
@ -2286,9 +2285,9 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_distributor_ingester_append_failures_total{cluster=\"$cluster\", namespace=\"$namespace\"}[5m])) by (ingester)",
"expr": "sum(rate(loki_distributor_ingester_append_failures_total{cluster=\"$cluster\", namespace=\"$namespace\"}[5m])) by (pod)",
"intervalFactor": 1,
"legendFormat": "{{ingester}}",
"legendFormat": "{{pod}}",
"refId": "A"
}
],
@ -2755,7 +2754,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate({cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\"} | logfmt | level=\"error\"[1m]))",
"expr": "sum(rate({cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\"} | logfmt | level=\"error\"[1m]))",
"refId": "A"
}
],
@ -2818,7 +2817,7 @@
"panels": [ ],
"targets": [
{
"expr": "{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\"} | logfmt | level=\"error\"",
"expr": "{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\"} | logfmt | level=\"error\"",
"refId": "A"
}
],
@ -2869,7 +2868,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\", status_code!~\"5[0-9]{2}\"}[1m])) by (route)\n/\nsum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester\"}[1m])) by (route)",
"expr": "sum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\", status_code!~\"5[0-9]{2}\"}[1m])) by (route)\n/\nsum(rate(loki_request_duration_seconds_count{cluster=\"$cluster\", namespace=\"$namespace\", job=~\"($namespace)/ingester.*\"}[1m])) by (route)",
"interval": "",
"intervalFactor": 1,
"legendFormat": "{{route}}",
@ -2977,7 +2976,7 @@
"steppedLine": false,
"targets": [
{
"expr": "topk(10,sum by (tenant) (loki_ingester_memory_streams{cluster=\"$cluster\",job=~\"($namespace)/ingester\"}))",
"expr": "topk(10,sum by (tenant) (loki_ingester_memory_streams{cluster=\"$cluster\",job=~\"($namespace)/ingester.*\"}))",
"interval": "",
"legendFormat": "{{ tenant }}",
"refId": "A"
@ -3068,7 +3067,7 @@
"steppedLine": false,
"targets": [
{
"expr": "topk(10, sum by (tenant) (rate(loki_ingester_streams_created_total{cluster=\"$cluster\",job=~\"($namespace)/ingester\"}[1m]) > 0))",
"expr": "topk(10, sum by (tenant) (rate(loki_ingester_streams_created_total{cluster=\"$cluster\",job=~\"($namespace)/ingester.*\"}[1m]) > 0))",
"interval": "",
"legendFormat": "{{ tenant }}",
"refId": "A"
@ -3178,13 +3177,13 @@
"steppedLine": false,
"targets": [
{
"expr": "sum(rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\",job=~\"($namespace)/ingester\"}[1m]))",
"expr": "sum(rate(loki_ingester_chunks_flushed_total{cluster=\"$cluster\",job=~\"($namespace)/ingester.*\"}[1m]))",
"interval": "",
"legendFormat": "Chunks",
"refId": "A"
},
{
"expr": "sum(increase(loki_chunk_store_deduped_chunks_total{cluster=\"$cluster\", job=~\"($namespace)/ingester\"}[1m]))/sum(increase(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"($namespace)/ingester\"}[1m])) < 1",
"expr": "sum(increase(loki_chunk_store_deduped_chunks_total{cluster=\"$cluster\", job=~\"($namespace)/ingester.*\"}[1m]))/sum(increase(loki_ingester_chunks_flushed_total{cluster=\"$cluster\", job=~\"($namespace)/ingester.*\"}[1m])) < 1",
"interval": "",
"legendFormat": "De-Dupe Ratio",
"refId": "B"
@ -3262,7 +3261,7 @@
"reverseYBuckets": false,
"targets": [
{
"expr": "sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\",job=~\"($namespace)/ingester\"}[1m])) by (le)",
"expr": "sum(rate(loki_ingester_chunk_size_bytes_bucket{cluster=\"$cluster\",job=~\"($namespace)/ingester.*\"}[1m])) by (le)",
"format": "heatmap",
"instant": false,
"interval": "",
@ -3417,7 +3416,7 @@
"reverseYBuckets": false,
"targets": [
{
"expr": "sum by (le) (rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"($namespace)/ingester\"}[1m]))",
"expr": "sum by (le) (rate(loki_ingester_chunk_utilization_bucket{cluster=\"$cluster\", job=~\"($namespace)/ingester.*\"}[1m]))",
"format": "heatmap",
"instant": false,
"interval": "",

View file

@ -1829,7 +1829,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by(pod) (go_memstats_heap_inuse_bytes{cluster=~\"$cluster\", job=~\"($namespace)/ingester\"})",
"expr": "sum by(pod) (go_memstats_heap_inuse_bytes{cluster=~\"$cluster\", job=~\"($namespace)/ingester.+\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{pod}}",

View file

@ -630,7 +630,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_index_request_duration_seconds_count{cluster=~\"$cluster\",job=~\"($namespace)/querier\", operation!=\"index_chunk\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_request_duration_seconds_count{cluster=~\"$cluster\",job=~\"($namespace)/ingester-zone.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{status}}",
@ -704,6 +704,194 @@
"span": 6,
"stack": false,
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=~\"$cluster\"})) * 1e3",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{ route }} 99th Percentile",
"refId": "A",
"step": 10
},
{
"expr": "histogram_quantile(0.50, sum by (le,route) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=~\"$cluster\"})) * 1e3",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{ route }} 50th Percentile",
"refId": "B",
"step": 10
},
{
"expr": "1e3 * sum(job_route:loki_request_duration_seconds_sum:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=~\"$cluster\"}) by (route) / sum(job_route:loki_request_duration_seconds_count:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=~\"/logproto.Querier/Query|/logproto.Querier/Label|/logproto.Querier/Series|/logproto.Querier/QuerySample|/logproto.Querier/GetChunkIDs\", cluster=~\"$cluster\"}) by (route) ",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{ route }} Average",
"refId": "C",
"step": 10
}
],
"thresholds": [ ],
"timeFrom": null,
"timeShift": null,
"title": "Latency",
"tooltip": {
"shared": true,
"sort": 2,
"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": false
}
]
}
],
"repeat": null,
"repeatIteration": null,
"repeatRowId": null,
"showTitle": true,
"title": "Ingester - Zone Aware",
"titleSize": "h6"
},
{
"collapse": false,
"height": "250px",
"panels": [
{
"aliasColors": {
"1xx": "#EAB839",
"2xx": "#7EB26D",
"3xx": "#6ED0E0",
"4xx": "#EF843C",
"5xx": "#E24D42",
"error": "#E24D42",
"success": "#7EB26D"
},
"bars": false,
"dashLength": 10,
"dashes": false,
"datasource": "$datasource",
"fill": 10,
"id": 9,
"legend": {
"avg": false,
"current": false,
"max": false,
"min": false,
"show": true,
"total": false,
"values": false
},
"lines": true,
"linewidth": 0,
"links": [ ],
"nullPointMode": "null as zero",
"percentage": false,
"pointradius": 5,
"points": false,
"renderer": "flot",
"seriesOverrides": [ ],
"spaceLength": 10,
"span": 6,
"stack": true,
"steppedLine": false,
"targets": [
{
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_index_request_duration_seconds_count{cluster=~\"$cluster\",job=~\"($namespace)/querier\", operation!=\"index_chunk\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{status}}",
"refId": "A",
"step": 10
}
],
"thresholds": [ ],
"timeFrom": null,
"timeShift": null,
"title": "QPS",
"tooltip": {
"shared": true,
"sort": 2,
"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
}
]
},
{
"aliasColors": { },
"bars": false,
"dashLength": 10,
"dashes": false,
"datasource": "$datasource",
"fill": 1,
"id": 10,
"legend": {
"avg": false,
"current": false,
"max": false,
"min": false,
"show": true,
"total": false,
"values": false
},
"lines": true,
"linewidth": 1,
"links": [ ],
"nullPointMode": "null as zero",
"percentage": false,
"pointradius": 5,
"points": false,
"renderer": "flot",
"seriesOverrides": [ ],
"spaceLength": 10,
"span": 6,
"stack": false,
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum(rate(loki_index_request_duration_seconds_bucket{cluster=~\"$cluster\",job=~\"($namespace)/querier\", operation!=\"index_chunk\"}[$__rate_interval])) by (le)) * 1e3",
@ -793,7 +981,7 @@
"dashes": false,
"datasource": "$datasource",
"fill": 10,
"id": 9,
"id": 11,
"legend": {
"avg": false,
"current": false,
@ -869,7 +1057,7 @@
"dashes": false,
"datasource": "$datasource",
"fill": 1,
"id": 10,
"id": 12,
"legend": {
"avg": false,
"current": false,

View file

@ -321,7 +321,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by(pod) (loki_ingester_memory_streams{cluster=~\"$cluster\", job=~\"($namespace)/ingester\"})",
"expr": "sum by(pod) (loki_ingester_memory_streams{cluster=~\"$cluster\", job=~\"($namespace)/ingester.*\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{pod}}",
@ -574,7 +574,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by(pod) (go_memstats_heap_inuse_bytes{cluster=~\"$cluster\", job=~\"($namespace)/ingester\"})",
"expr": "sum by(pod) (go_memstats_heap_inuse_bytes{cluster=~\"$cluster\", job=~\"($namespace)/ingester.*\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{pod}}",
@ -803,7 +803,7 @@
"steppedLine": false,
"targets": [
{
"expr": "max by(persistentvolumeclaim) (kubelet_volume_stats_used_bytes{cluster=~\"$cluster\", namespace=~\"$namespace\"} / kubelet_volume_stats_capacity_bytes{cluster=~\"$cluster\", namespace=~\"$namespace\"}) and count by(persistentvolumeclaim) (kube_persistentvolumeclaim_labels{cluster=~\"$cluster\", namespace=~\"$namespace\",label_name=~\"ingester.*\"})",
"expr": "max by(persistentvolumeclaim) (kubelet_volume_stats_used_bytes{cluster=~\"$cluster\", namespace=~\"$namespace\"} / kubelet_volume_stats_capacity_bytes{cluster=~\"$cluster\", namespace=~\"$namespace\"}) and count by(persistentvolumeclaim) (kube_persistentvolumeclaim_labels{cluster=~\"$cluster\", namespace=~\"$namespace\",label_name=~\"ingester.*.*\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{persistentvolumeclaim}}",

View file

@ -254,7 +254,7 @@
"steppedLine": false,
"targets": [
{
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_request_duration_seconds_count{cluster=~\"$cluster\",job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_request_duration_seconds_count{cluster=~\"$cluster\",job=~\"($namespace)/ingester-zone.*\", route=\"/logproto.Pusher/Push\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{status}}",
@ -328,6 +328,194 @@
"span": 6,
"stack": false,
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "99th Percentile",
"refId": "A",
"step": 10
},
{
"expr": "histogram_quantile(0.50, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "50th Percentile",
"refId": "B",
"step": 10
},
{
"expr": "1e3 * sum(job_route:loki_request_duration_seconds_sum:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"}) / sum(job_route:loki_request_duration_seconds_count:sum_rate{job=~\"($namespace)/ingester-zone.*\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "Average",
"refId": "C",
"step": 10
}
],
"thresholds": [ ],
"timeFrom": null,
"timeShift": null,
"title": "Latency",
"tooltip": {
"shared": true,
"sort": 2,
"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": false
}
]
}
],
"repeat": null,
"repeatIteration": null,
"repeatRowId": null,
"showTitle": true,
"title": "Ingester - Zone Aware",
"titleSize": "h6"
},
{
"collapse": false,
"height": "250px",
"panels": [
{
"aliasColors": {
"1xx": "#EAB839",
"2xx": "#7EB26D",
"3xx": "#6ED0E0",
"4xx": "#EF843C",
"5xx": "#E24D42",
"error": "#E24D42",
"success": "#7EB26D"
},
"bars": false,
"dashLength": 10,
"dashes": false,
"datasource": "$datasource",
"fill": 10,
"id": 5,
"legend": {
"avg": false,
"current": false,
"max": false,
"min": false,
"show": true,
"total": false,
"values": false
},
"lines": true,
"linewidth": 0,
"links": [ ],
"nullPointMode": "null as zero",
"percentage": false,
"pointradius": 5,
"points": false,
"renderer": "flot",
"seriesOverrides": [ ],
"spaceLength": 10,
"span": 6,
"stack": true,
"steppedLine": false,
"targets": [
{
"expr": "sum by (status) (\n label_replace(label_replace(rate(loki_request_duration_seconds_count{cluster=~\"$cluster\",job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\"}[$__rate_interval]),\n \"status\", \"${1}xx\", \"status_code\", \"([0-9])..\"),\n \"status\", \"${1}\", \"status_code\", \"([a-z]+)\"))\n",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "{{status}}",
"refId": "A",
"step": 10
}
],
"thresholds": [ ],
"timeFrom": null,
"timeShift": null,
"title": "QPS",
"tooltip": {
"shared": true,
"sort": 2,
"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
}
]
},
{
"aliasColors": { },
"bars": false,
"dashLength": 10,
"dashes": false,
"datasource": "$datasource",
"fill": 1,
"id": 6,
"legend": {
"avg": false,
"current": false,
"max": false,
"min": false,
"show": true,
"total": false,
"values": false
},
"lines": true,
"linewidth": 1,
"links": [ ],
"nullPointMode": "null as zero",
"percentage": false,
"pointradius": 5,
"points": false,
"renderer": "flot",
"seriesOverrides": [ ],
"spaceLength": 10,
"span": 6,
"stack": false,
"steppedLine": false,
"targets": [
{
"expr": "histogram_quantile(0.99, sum by (le) (job_route:loki_request_duration_seconds_bucket:sum_rate{job=~\"($namespace)/ingester\", route=\"/logproto.Pusher/Push\", cluster=~\"$cluster\"})) * 1e3",
@ -417,7 +605,7 @@
"dashes": false,
"datasource": "$datasource",
"fill": 10,
"id": 5,
"id": 7,
"legend": {
"avg": false,
"current": false,
@ -493,7 +681,7 @@
"dashes": false,
"datasource": "$datasource",
"fill": 1,
"id": 6,
"id": 8,
"legend": {
"avg": false,
"current": false,
@ -605,7 +793,7 @@
"dashes": false,
"datasource": "$datasource",
"fill": 10,
"id": 7,
"id": 9,
"legend": {
"avg": false,
"current": false,
@ -681,7 +869,7 @@
"dashes": false,
"datasource": "$datasource",
"fill": 1,
"id": 8,
"id": 10,
"legend": {
"avg": false,
"current": false,

View file

@ -786,6 +786,7 @@ annotations:
summary: Client certificate is about to expire.
expr: |
apiserver_client_certificate_expiration_seconds_count{job="kube-apiserver"} > 0 and on(job) histogram_quantile(0.01, sum by (job, le) (rate(apiserver_client_certificate_expiration_seconds_bucket{job="kube-apiserver"}[5m]))) < 604800
for: 5m
labels:
severity: warning
{{< /code >}}
@ -802,6 +803,7 @@ annotations:
summary: Client certificate is about to expire.
expr: |
apiserver_client_certificate_expiration_seconds_count{job="kube-apiserver"} > 0 and on(job) histogram_quantile(0.01, sum by (job, le) (rate(apiserver_client_certificate_expiration_seconds_bucket{job="kube-apiserver"}[5m]))) < 86400
for: 5m
labels:
severity: critical
{{< /code >}}