bstein/titan-iac
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

monitoring: fix net/io legend labels

Browse Source
This commit is contained in:
Brad Stein 2025-11-17 20:19:20 -03:00
parent b8998a3c6a
commit 53427cc8fa
3 changed files with 15 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
scripts/render_dashboards.py
View File

@ -144,6 +144,10 @@ def astreae_free_expr(mount):
return f"sum(node_filesystem_avail_bytes{{mountpoint=\"{mount}\",fstype!~\"tmpfs|overlay\"}})" return f"sum(node_filesystem_avail_bytes{{mountpoint=\"{mount}\",fstype!~\"tmpfs|overlay\"}})"
def topk_with_node(expr):
return f'label_replace(topk(1, {expr}), "__name__", "$1", "node", "(.*)")'
def node_net_expr(scope=""): def node_net_expr(scope=""):
base = ( base = (
'sum by (instance) (' 'sum by (instance) ('
@ -469,15 +473,15 @@ def build_overview():
hottest = [ hottest = [
(7, "Hottest node: CPU", node_cpu_expr(), "percent"), (7, "Hottest node: CPU", node_cpu_expr(), "percent"),
(8, "Hottest node: RAM", node_mem_expr(), "percent"), (8, "Hottest node: RAM", node_mem_expr(), "percent"),
(9, "Hottest node: NET (rx+tx)", node_net_expr(), "Bps"), (9, "Hottest node: NET (rx+tx)", topk_with_node(node_net_expr()), "Bps"),
(10, "Hottest node: I/O (r+w)", node_io_expr(), "Bps"), (10, "Hottest node: I/O (r+w)", topk_with_node(node_io_expr()), "Bps"),
] ]
for idx, (panel_id, title, expr, unit) in enumerate(hottest): for idx, (panel_id, title, expr, unit) in enumerate(hottest):
panels.append( panels.append(
stat_panel( stat_panel(
panel_id, panel_id,
title, title,
f"topk(1, {expr})", f"{expr}",
{"h": 5, "w": 6, "x": 6 * idx, "y": 5}, {"h": 5, "w": 6, "x": 6 * idx, "y": 5},
unit=unit, unit=unit,
thresholds=PERCENT_THRESHOLDS if unit == "percent" else None, thresholds=PERCENT_THRESHOLDS if unit == "percent" else None,

8
services/monitoring/dashboards/atlas-overview.json
View File

@ -438,7 +438,7 @@
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{mode=\"idle\"}[5m]))) * 100) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{mode=\"idle\"}[5m]))) * 100) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true
@ -511,7 +511,7 @@
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) ((avg by (instance) ((node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) / node_memory_MemTotal_bytes * 100)) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "avg by (node) ((avg by (instance) ((node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) / node_memory_MemTotal_bytes * 100)) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true
@ -584,7 +584,7 @@
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) ((sum by (instance) (rate(container_network_receive_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]) + rate(container_network_transmit_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "label_replace(topk(1, avg by (node) ((sum by (instance) (rate(container_network_receive_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]) + rate(container_network_transmit_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))), \"__name__\", \"$1\", \"node\", \"(.*)\")",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true
@ -653,7 +653,7 @@
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[5m]) + rate(node_disk_written_bytes_total[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "label_replace(topk(1, avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[5m]) + rate(node_disk_written_bytes_total[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))), \"__name__\", \"$1\", \"node\", \"(.*)\")",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true

8
services/monitoring/grafana-dashboard-overview.yaml
View File

@ -447,7 +447,7 @@ data:
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{mode=\"idle\"}[5m]))) * 100) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{mode=\"idle\"}[5m]))) * 100) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true
@ -520,7 +520,7 @@ data:
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) ((avg by (instance) ((node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) / node_memory_MemTotal_bytes * 100)) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "avg by (node) ((avg by (instance) ((node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes) / node_memory_MemTotal_bytes * 100)) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true
@ -593,7 +593,7 @@ data:
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) ((sum by (instance) (rate(container_network_receive_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]) + rate(container_network_transmit_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "label_replace(topk(1, avg by (node) ((sum by (instance) (rate(container_network_receive_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]) + rate(container_network_transmit_bytes_total{namespace!=\"\" ,pod!=\"\"}[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))), \"__name__\", \"$1\", \"node\", \"(.*)\")",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true
@ -662,7 +662,7 @@ data:
}, },
"targets": [ "targets": [
{ {
"expr": "topk(1, avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[5m]) + rate(node_disk_written_bytes_total[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\")))", "expr": "label_replace(topk(1, avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[5m]) + rate(node_disk_written_bytes_total[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=\"\"}, \"node\", \"$1\", \"nodename\", \"(.*)\"))), \"__name__\", \"$1\", \"node\", \"(.*)\")",
"refId": "A", "refId": "A",
"legendFormat": "{{node}}", "legendFormat": "{{node}}",
"instant": true "instant": true