feat: restructure Grafana configuration, migrate alert rules to new format and add VictoriaMetrics datasource
This commit is contained in:
71
hosts/web-arm/modules/grafana/alerting/inode_usage.nix
Normal file
71
hosts/web-arm/modules/grafana/alerting/inode_usage.nix
Normal file
@@ -0,0 +1,71 @@
|
||||
{ lib, pkgs, config, ... }:
|
||||
{
|
||||
services.grafana.provision.alerting.rules.settings.groups = [
|
||||
{
|
||||
name = "InodeUsageAlerts";
|
||||
folder = "System Alerts";
|
||||
interval = "1m";
|
||||
|
||||
rules = [
|
||||
{
|
||||
uid = "high-inode-usage-alert-uid";
|
||||
title = "HighInodeUsage";
|
||||
condition = "D"; # Condition is now D
|
||||
|
||||
data = [
|
||||
# Query A: Calculate inode usage percentage
|
||||
{
|
||||
refId = "A";
|
||||
datasourceUid = "vm-datasource-uid";
|
||||
queryType = "prometheus";
|
||||
relativeTimeRange = { from = 60; to = 0; };
|
||||
model = {
|
||||
expr = ''
|
||||
(
|
||||
node_filesystem_files{fstype!~"tmpfs|rootfs",mountpoint!=""} - node_filesystem_files_free{fstype!~"tmpfs|rootfs",mountpoint!=""}
|
||||
) / (node_filesystem_files{fstype!~"tmpfs|rootfs",mountpoint!=""} > 0) * 100
|
||||
and node_filesystem_files{fstype!~"tmpfs|rootfs",mountpoint!=""}
|
||||
and node_filesystem_files_free{fstype!~"tmpfs|rootfs",mountpoint!=""}
|
||||
'';
|
||||
legendFormat = "{{mountpoint}} on {{instance}}";
|
||||
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 > 80"; # Alert if inode usage from C is > 80%
|
||||
};
|
||||
}
|
||||
];
|
||||
|
||||
for = "30m"; # Duration the condition must be met
|
||||
noDataState = "NoData";
|
||||
execErrState = "Error";
|
||||
|
||||
annotations = {
|
||||
summary = "High inode usage on {{ $labels.instance }} at {{ $labels.mountpoint }}";
|
||||
description = ''Inode usage on {{ $labels.instance }} for mount point {{ $labels.mountpoint }} (fstype: {{ $labels.fstype }}) has been above 80% for more than 30 minutes. Current value: {{ if $values.C }}{{ $values.C | humanizePercentage }}{{ else }}N/A{{ end }}%.'';
|
||||
};
|
||||
labels = {
|
||||
severity = "warning";
|
||||
category = "capacity";
|
||||
};
|
||||
}
|
||||
];
|
||||
}
|
||||
];
|
||||
}
|
||||
Reference in New Issue
Block a user