add metrics.status.im Grafana dashboard (#2817)

and simplify queries for the old one
This commit is contained in:
Ștefan Talpalaru 2021-08-29 00:51:10 +02:00 committed by GitHub
parent 0418fbada2
commit b1ed4d6ef4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 2777 additions and 32 deletions

17
grafana/README.md Normal file
View File

@ -0,0 +1,17 @@
These are Grafana dashboards exported to JSON and a sample Prometheus
configuration, to get you started.
## beacon_nodes_Grafana_dashboard.json
Can be loaded in a local Grafana instance directly.
## metrics.status.im.json
Exported from [https://metrics.status.im/d/pgeNfj2Wz23/nimbus-fleet-testnets?orgId=1](https://metrics.status.im/d/pgeNfj2Wz23/nimbus-fleet-testnets?orgId=1). It diverged a little from "beacon\_nodes\_Grafana\_dashboard.json" by adding Netdata metrics and a few extra panels.
In order to use it locally, you would have to make some changes:
- remove `beacon_current_epoch{job=\"beacon-node-metrics\"},` from the "instance" variable query
- disable the "container" variable by removing `,container=\"${container}\"` from all panel queries

View File

@ -81,27 +81,27 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (rate(process_cpu_seconds_total{instance=\"${instance}\"}[$__rate_interval])) * 100",
"expr": "rate(process_cpu_seconds_total{instance=\"${instance}\"}[$__rate_interval]) * 100",
"legendFormat": "CPU usage %",
"refId": "A"
},
{
"expr": "min without (node) (process_open_fds{instance=\"${instance}\"})",
"expr": "process_open_fds{instance=\"${instance}\"}",
"legendFormat": "open file descriptors",
"refId": "C"
},
{
"expr": "min without (node) (process_resident_memory_bytes{instance=\"${instance}\"})",
"expr": "process_resident_memory_bytes{instance=\"${instance}\"}",
"legendFormat": "RSS",
"refId": "D"
},
{
"expr": "min without (node) (nim_gc_mem_bytes{instance=\"${instance}\"})",
"expr": "nim_gc_mem_bytes{instance=\"${instance}\"}",
"legendFormat": "Nim GC mem total",
"refId": "F"
},
{
"expr": "min without (node) (nim_gc_mem_occupied_bytes{instance=\"${instance}\"})",
"expr": "nim_gc_mem_occupied_bytes{instance=\"${instance}\"}",
"legendFormat": "Nim GC mem used",
"refId": "G"
}
@ -198,7 +198,7 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (libp2p_open_streams{instance=\"${instance}\"})",
"expr": "libp2p_open_streams{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "{{type}} ({{dir}})",
"refId": "A"
@ -296,19 +296,19 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (libp2p_peers{instance=\"${instance}\"})",
"expr": "libp2p_peers{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "libp2p peers",
"refId": "A"
},
{
"expr": "min without (node) (libp2p_pubsub_peers{instance=\"${instance}\"})",
"expr": "libp2p_pubsub_peers{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "libp2p pubsub peers",
"refId": "C"
},
{
"expr": "min without (node) (nbc_peers{instance=\"${instance}\"})",
"expr": "nbc_peers{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "beacon_node peers",
"refId": "B"
@ -411,7 +411,7 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (nim_gc_heap_instance_occupied_bytes{instance=\"${instance}\"})",
"expr": "nim_gc_heap_instance_occupied_bytes{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "{{type_name}}",
"refId": "A"
@ -509,7 +509,7 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (beacon_active_validators{instance=\"${instance}\"})",
"expr": "beacon_active_validators{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "current validators",
"refId": "A"
@ -613,13 +613,13 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (rate(beacon_blocks_received_total{instance=\"${instance}\"}[$__rate_interval])) * 12",
"expr": "rate(beacon_blocks_received_total{instance=\"${instance}\"}[$__rate_interval]) * 12",
"interval": "",
"legendFormat": "received",
"refId": "B"
},
{
"expr": "min without (node) (rate(beacon_blocks_proposed_total{instance=\"${instance}\"}[$__rate_interval])) * 12",
"expr": "rate(beacon_blocks_proposed_total{instance=\"${instance}\"}[$__rate_interval]) * 12",
"interval": "",
"legendFormat": "proposed",
"refId": "A"
@ -715,7 +715,7 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (beacon_state_data_cache_hits_total{instance=\"${instance}\"}) * 100 / (min without (node) (beacon_state_data_cache_hits_total{instance=\"${instance}\"}) + min without (node) (beacon_state_data_cache_misses_total{instance=\"${instance}\"}))",
"expr": "beacon_state_data_cache_hits_total{instance=\"${instance}\"} * 100 / (beacon_state_data_cache_hits_total{instance=\"${instance}\"} + beacon_state_data_cache_misses_total{instance=\"${instance}\"})",
"interval": "",
"legendFormat": "cache hit rate",
"refId": "A"
@ -817,13 +817,13 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (rate(beacon_attestations_received_total{instance=\"${instance}\"}[$__rate_interval])) * 12",
"expr": "rate(beacon_attestations_received_total{instance=\"${instance}\"}[$__rate_interval]) * 12",
"interval": "",
"legendFormat": "received",
"refId": "A"
},
{
"expr": "min without (node) (rate(beacon_attestations_sent_total{instance=\"${instance}\"}[$__rate_interval])) * 12",
"expr": "rate(beacon_attestations_sent_total{instance=\"${instance}\"}[$__rate_interval]) * 12",
"interval": "",
"legendFormat": "sent",
"refId": "B"
@ -941,7 +941,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (process_resident_memory_bytes{instance=\"${instance}\"})",
"expr": "process_resident_memory_bytes{instance=\"${instance}\"}",
"refId": "A"
}
],
@ -1031,7 +1031,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (rate(process_cpu_seconds_total{instance=\"${instance}\"}[$__rate_interval])) * 100",
"expr": "rate(process_cpu_seconds_total{instance=\"${instance}\"}[$__rate_interval]) * 100",
"refId": "A"
}
],
@ -1121,7 +1121,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (beacon_slot{instance=\"${instance}\"})",
"expr": "beacon_slot{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1212,7 +1212,7 @@
"tableColumn": "",
"targets": [
{
"expr": "sum(min without (node) (beacon_attestations_sent_total))",
"expr": "sum(beacon_attestations_sent_total)",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1304,7 +1304,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (beacon_attestations_received_total{instance=\"${instance}\"})",
"expr": "beacon_attestations_received_total{instance=\"${instance}\"}",
"refId": "A"
}
],
@ -1362,7 +1362,7 @@
"reverseYBuckets": false,
"targets": [
{
"expr": "min without (node) (rate(beacon_attestation_delay_bucket{instance=\"${instance}\"}[$__rate_interval]))",
"expr": "rate(beacon_attestation_delay_bucket{instance=\"${instance}\"}[$__rate_interval])",
"format": "heatmap",
"instant": false,
"interval": "",
@ -1467,7 +1467,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (beacon_current_epoch{instance=\"${instance}\"})",
"expr": "beacon_current_epoch{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1558,7 +1558,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (beacon_current_justified_epoch{instance=\"${instance}\"})",
"expr": "beacon_current_justified_epoch{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1650,7 +1650,7 @@
"tableColumn": "",
"targets": [
{
"expr": "time() - min without (node) (process_start_time_seconds{instance=\"${instance}\"})",
"expr": "time() - process_start_time_seconds{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1742,7 +1742,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (nbc_peers{instance=\"${instance}\"})",
"expr": "nbc_peers{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1833,7 +1833,7 @@
"tableColumn": "",
"targets": [
{
"expr": "min without (node) (beacon_finalized_epoch{instance=\"${instance}\"})",
"expr": "beacon_finalized_epoch{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "",
"refId": "A"
@ -1894,7 +1894,7 @@
"reverseYBuckets": false,
"targets": [
{
"expr": "min without (node) (rate(beacon_aggregate_delay_bucket{instance=\"${instance}\"}[$__rate_interval]))",
"expr": "rate(beacon_aggregate_delay_bucket{instance=\"${instance}\"}[$__rate_interval])",
"format": "heatmap",
"instant": false,
"interval": "",
@ -1969,7 +1969,7 @@
"reverseYBuckets": false,
"targets": [
{
"expr": "min without (node) (rate(beacon_block_delay_bucket{instance=\"${instance}\"}[$__rate_interval]))",
"expr": "rate(beacon_block_delay_bucket{instance=\"${instance}\"}[$__rate_interval])",
"format": "heatmap",
"instant": false,
"interval": "",
@ -2043,7 +2043,7 @@
"reverseYBuckets": false,
"targets": [
{
"expr": "min without (node) (rate(beacon_store_block_duration_seconds_bucket{instance=\"${instance}\"}[$__rate_interval]))",
"expr": "rate(beacon_store_block_duration_seconds_bucket{instance=\"${instance}\"}[$__rate_interval])",
"format": "heatmap",
"instant": false,
"interval": "",
@ -2131,7 +2131,7 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (rate(attached_validator_balance_total{instance=\"${instance}\"}[$__rate_interval]) * 384 / 1000000000)",
"expr": "rate(attached_validator_balance_total{instance=\"${instance}\"}[$__rate_interval]) * 384 / 1000000000",
"interval": "",
"legendFormat": "GWei",
"refId": "A"
@ -2232,7 +2232,7 @@
"steppedLine": false,
"targets": [
{
"expr": "min without (node) (sqlite3_memory_used_bytes{instance=\"${instance}\"})",
"expr": "sqlite3_memory_used_bytes{instance=\"${instance}\"}",
"interval": "",
"legendFormat": "Memory used",
"refId": "A"

File diff suppressed because it is too large Load Diff