feat: restructure Grafana configuration, migrate alert rules to new format and add VictoriaMetrics datasource
This commit is contained in:
69
hosts/web-arm/modules/grafana/alerting/ram_usage.nix
Normal file
69
hosts/web-arm/modules/grafana/alerting/ram_usage.nix
Normal file
@@ -0,0 +1,69 @@
|
||||
{ lib, pkgs, config, ... }:
|
||||
{
|
||||
services.grafana.provision.alerting.rules.settings.groups = [
|
||||
{
|
||||
name = "RAMUsageAlerts";
|
||||
folder = "System Alerts";
|
||||
interval = "1m";
|
||||
|
||||
rules = [
|
||||
{
|
||||
uid = "high-ram-usage-alert-uid";
|
||||
title = "HighRAMUsage";
|
||||
condition = "D"; # Condition is now D
|
||||
|
||||
data = [
|
||||
# Query A: Calculate RAM usage percentage
|
||||
{
|
||||
refId = "A";
|
||||
datasourceUid = "vm-datasource-uid";
|
||||
queryType = "prometheus";
|
||||
relativeTimeRange = { from = 60; to = 0; };
|
||||
model = {
|
||||
expr = ''
|
||||
(1 - node_memory_MemAvailable_bytes / (node_memory_MemTotal_bytes > 0)) * 100
|
||||
and node_memory_MemAvailable_bytes
|
||||
and node_memory_MemTotal_bytes
|
||||
'';
|
||||
legendFormat = "RAM usage on {{instance}} ({{job}})";
|
||||
instant = false;
|
||||
};
|
||||
}
|
||||
# Expression C: Reduce Query A to its last value, preserving labels
|
||||
{
|
||||
refId = "C";
|
||||
datasourceUid = "__expr__";
|
||||
model = {
|
||||
type = "reduce";
|
||||
expression = "A"; # Input is Query A
|
||||
reducer = "last"; # Get the last value of each series in A
|
||||
};
|
||||
}
|
||||
# Expression D: Apply math condition to the reduced values from C
|
||||
{
|
||||
refId = "D";
|
||||
datasourceUid = "__expr__";
|
||||
model = {
|
||||
type = "math";
|
||||
expression = "$C > 90"; # Alert if RAM usage from C is > 90%
|
||||
};
|
||||
}
|
||||
];
|
||||
|
||||
for = "10m"; # Duration the condition must be met
|
||||
noDataState = "NoData";
|
||||
execErrState = "Error";
|
||||
|
||||
annotations = {
|
||||
summary = "High RAM usage on {{ $labels.instance }}";
|
||||
description = ''RAM usage on {{ $labels.instance }} (job: {{ $labels.job }}) has been above 90% for more than 10 minutes. Current value: {{ if $values.C }}{{ $values.C | humanizePercentage }}{{ else }}N/A{{ end }}%.'';
|
||||
};
|
||||
labels = {
|
||||
severity = "warning";
|
||||
category = "performance";
|
||||
};
|
||||
}
|
||||
];
|
||||
}
|
||||
];
|
||||
}
|
||||
Reference in New Issue
Block a user