Skip to content

Commit

Permalink
INF production Grafana snapshot (#3890)
Browse files Browse the repository at this point in the history
  • Loading branch information
@joaquincasares
joaquincasares authored Sep 19, 2022
1 parent 64fb281 commit fb87b2e
Show file tree
Hide file tree
Showing 20 changed files with 11,474 additions and 1,459 deletions.
186 changes: 186 additions & 0 deletions monitoring/grafana/alerts/audius-content-node-bull-queue_waiting-queue-jobs-prod.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,186 @@
[
{
"annotations": {
"__alertId__": "3VPUdqZVz_039_P3_prod",
"__dashboardUid__": "3VPUdqZVz",
"__panelId__": "39",
"description": "If the number of waiting jobs is skyrocketing, it might mean:\n\n* Too many are being created.\n* The jobs for that specific queue are taking too long to execute.",
"runbook_url": "http://grafana.audius.co/d/3VPUdqZVz?viewPanel=39",
"summary": "The number of bull jobs waiting to be executed."
},
"condition": "B",
"data": [
{
"datasourceUid": "r2_nnDL7z",
"model": {
"datasource": {
"type": "prometheus",
"uid": "r2_nnDL7z"
},
"expr": "audius_cn_jobs_waiting_total{environment=~\"prod\", queue_name=~\".*\", host=~\".*\"}",
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "A"
},
"queryType": "",
"refId": "A",
"relativeTimeRange": {
"from": 600,
"to": 0
}
},
{
"datasourceUid": "-100",
"model": {
"conditions": [
{
"evaluator": {
"params": [
1100,
0
],
"type": "gt"
},
"operator": {
"type": "or"
},
"query": {
"params": [
"A"
]
},
"reducer": {
"params": [],
"type": "last"
},
"type": "query"
}
],
"datasource": {
"name": "Expression",
"type": "__expr__",
"uid": "__expr__"
},
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "B",
"type": "classic_conditions"
},
"queryType": "",
"refId": "B",
"relativeTimeRange": {
"from": 0,
"to": 0
}
}
],
"execErrState": "Alerting",
"folderUID": "gtl7D0W4z",
"for": "0s",
"id": 3303931,
"labels": {
"alert": "p3",
"env": "prod",
"team": "content"
},
"noDataState": "OK",
"orgID": 1,
"ruleGroup": "group",
"title": "PROD P3 | Waiting Queue Jobs",
"uid": "3VPUdqZVz_039_P3_prod"
},
{
"annotations": {
"__alertId__": "3VPUdqZVz_039_P2_prod",
"__dashboardUid__": "3VPUdqZVz",
"__panelId__": "39",
"description": "If the number of waiting jobs is skyrocketing, it might mean:\n\n* Too many are being created.\n* The jobs for that specific queue are taking too long to execute.",
"runbook_url": "http://grafana.audius.co/d/3VPUdqZVz?viewPanel=39",
"summary": "The number of bull jobs waiting to be executed."
},
"condition": "B",
"data": [
{
"datasourceUid": "r2_nnDL7z",
"model": {
"datasource": {
"type": "prometheus",
"uid": "r2_nnDL7z"
},
"expr": "audius_cn_jobs_waiting_total{environment=~\"prod\", queue_name=~\".*\", host=~\".*\"}",
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "A"
},
"queryType": "",
"refId": "A",
"relativeTimeRange": {
"from": 600,
"to": 0
}
},
{
"datasourceUid": "-100",
"model": {
"conditions": [
{
"evaluator": {
"params": [
5000,
0
],
"type": "gt"
},
"operator": {
"type": "or"
},
"query": {
"params": [
"A"
]
},
"reducer": {
"params": [],
"type": "last"
},
"type": "query"
}
],
"datasource": {
"name": "Expression",
"type": "__expr__",
"uid": "__expr__"
},
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "B",
"type": "classic_conditions"
},
"queryType": "",
"refId": "B",
"relativeTimeRange": {
"from": 0,
"to": 0
}
}
],
"execErrState": "Alerting",
"folderUID": "gtl7D0W4z",
"for": "0s",
"id": 3303921,
"labels": {
"alert": "p2",
"env": "prod",
"team": "content"
},
"noDataState": "OK",
"orgID": 1,
"ruleGroup": "group",
"title": "PROD P2 | Waiting Queue Jobs",
"uid": "3VPUdqZVz_039_P2_prod"
}
]
186 changes: 186 additions & 0 deletions monitoring/grafana/alerts/audius-content-node-bull-queue_waiting-queue-jobs-stage.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,186 @@
[
{
"annotations": {
"__alertId__": "3VPUdqZVz_039_P3_stage",
"__dashboardUid__": "3VPUdqZVz",
"__panelId__": "39",
"description": "If the number of waiting jobs is skyrocketing, it might mean:\n\n* Too many are being created.\n* The jobs for that specific queue are taking too long to execute.",
"runbook_url": "http://grafana.audius.co/d/3VPUdqZVz?viewPanel=39",
"summary": "The number of bull jobs waiting to be executed."
},
"condition": "B",
"data": [
{
"datasourceUid": "r2_nnDL7z",
"model": {
"datasource": {
"type": "prometheus",
"uid": "r2_nnDL7z"
},
"expr": "audius_cn_jobs_waiting_total{environment=~\"stage\", queue_name=~\".*\", host=~\".*\"}",
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "A"
},
"queryType": "",
"refId": "A",
"relativeTimeRange": {
"from": 600,
"to": 0
}
},
{
"datasourceUid": "-100",
"model": {
"conditions": [
{
"evaluator": {
"params": [
1100,
0
],
"type": "gt"
},
"operator": {
"type": "or"
},
"query": {
"params": [
"A"
]
},
"reducer": {
"params": [],
"type": "last"
},
"type": "query"
}
],
"datasource": {
"name": "Expression",
"type": "__expr__",
"uid": "__expr__"
},
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "B",
"type": "classic_conditions"
},
"queryType": "",
"refId": "B",
"relativeTimeRange": {
"from": 0,
"to": 0
}
}
],
"execErrState": "Alerting",
"folderUID": "gtl7D0W4z",
"for": "0s",
"id": 3303930,
"labels": {
"alert": "p3",
"env": "stage",
"team": "content"
},
"noDataState": "OK",
"orgID": 1,
"ruleGroup": "group",
"title": "STAGE P3 | Waiting Queue Jobs",
"uid": "3VPUdqZVz_039_P3_stage"
},
{
"annotations": {
"__alertId__": "3VPUdqZVz_039_P2_stage",
"__dashboardUid__": "3VPUdqZVz",
"__panelId__": "39",
"description": "If the number of waiting jobs is skyrocketing, it might mean:\n\n* Too many are being created.\n* The jobs for that specific queue are taking too long to execute.",
"runbook_url": "http://grafana.audius.co/d/3VPUdqZVz?viewPanel=39",
"summary": "The number of bull jobs waiting to be executed."
},
"condition": "B",
"data": [
{
"datasourceUid": "r2_nnDL7z",
"model": {
"datasource": {
"type": "prometheus",
"uid": "r2_nnDL7z"
},
"expr": "audius_cn_jobs_waiting_total{environment=~\"stage\", queue_name=~\".*\", host=~\".*\"}",
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "A"
},
"queryType": "",
"refId": "A",
"relativeTimeRange": {
"from": 600,
"to": 0
}
},
{
"datasourceUid": "-100",
"model": {
"conditions": [
{
"evaluator": {
"params": [
5000,
0
],
"type": "gt"
},
"operator": {
"type": "or"
},
"query": {
"params": [
"A"
]
},
"reducer": {
"params": [],
"type": "last"
},
"type": "query"
}
],
"datasource": {
"name": "Expression",
"type": "__expr__",
"uid": "__expr__"
},
"hide": false,
"intervalMs": 1000,
"maxDataPoints": 43200,
"refId": "B",
"type": "classic_conditions"
},
"queryType": "",
"refId": "B",
"relativeTimeRange": {
"from": 0,
"to": 0
}
}
],
"execErrState": "Alerting",
"folderUID": "gtl7D0W4z",
"for": "0s",
"id": 3303920,
"labels": {
"alert": "p2",
"env": "stage",
"team": "content"
},
"noDataState": "OK",
"orgID": 1,
"ruleGroup": "group",
"title": "STAGE P2 | Waiting Queue Jobs",
"uid": "3VPUdqZVz_039_P2_stage"
}
]
Loading

0 comments on commit fb87b2e

Please sign in to comment.