77 lines
3.1 KiB
Nix
77 lines
3.1 KiB
Nix
{ lib, pkgs, config, ... }:
|
|
{
|
|
grafanaAlertRuleDefinitions = [
|
|
{
|
|
uid = "high-disk-usage-alert-uid"; # Optional: provide a stable UID for the rule itself
|
|
title = "HighDiskUsage"; # Name of the alert rule (was 'alert' in vmalert)
|
|
|
|
# Condition for the alert to fire. 'D' refers to the refId of the threshold expression.
|
|
condition = "D"; # Condition is now D
|
|
# Removed rule-level relativeTimeRange
|
|
|
|
# Data queries and expressions
|
|
data = [
|
|
# Query A: Calculate disk usage percentage
|
|
{
|
|
refId = "A";
|
|
datasourceUid = "vm-datasource-uid"; # UID of the VictoriaMetrics datasource
|
|
queryType = "prometheus"; # Explicitly set, though often inferred
|
|
relativeTimeRange = { from = 60; to = 0; }; # Query-level, integer seconds
|
|
model = {
|
|
expr = ''
|
|
(
|
|
node_filesystem_size_bytes{fstype!~"tmpfs|rootfs",mountpoint!=""} - node_filesystem_avail_bytes{fstype!~"tmpfs|rootfs",mountpoint!=""}
|
|
) / (node_filesystem_size_bytes{fstype!~"tmpfs|rootfs",mountpoint!=""} > 0) * 100
|
|
and node_filesystem_size_bytes{fstype!~"tmpfs|rootfs",mountpoint!=""}
|
|
and node_filesystem_avail_bytes{fstype!~"tmpfs|rootfs",mountpoint!=""}
|
|
'';
|
|
legendFormat = "{{mountpoint}} on {{instance}}"; # Example legend
|
|
instant = false; # For range queries, default is 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 > 85"; # Check if the last value from each series in C is > 85
|
|
};
|
|
}
|
|
];
|
|
|
|
for = "15m"; # Duration the condition must be met (same as vmalert)
|
|
|
|
# How to handle states where data is missing or query errors
|
|
noDataState = "NoData"; # Options: NoData, Alerting, OK
|
|
execErrState = "Error"; # Options: Error, Alerting, OK
|
|
|
|
annotations = {
|
|
summary = "High disk usage on {{ $labels.instance }} at {{ $labels.mountpoint }}";
|
|
description = ''
|
|
Disk usage on {{ $labels.instance }} for mount point {{ $labels.mountpoint }}
|
|
(fstype: {{ $labels.fstype }}) has been above 85% for more than 15 minutes.
|
|
Current value: {{ if $values.C }}{{ $values.C | humanizePercentage }}{{ else }}N/A{{ end }}%.
|
|
''; # Using $values.C as it's the input to the math condition D
|
|
};
|
|
labels = {
|
|
severity = "warning";
|
|
category = "capacity";
|
|
# Grafana automatically adds labels from the query result (instance, mountpoint, etc.)
|
|
# and labels from the rule group/folder.
|
|
};
|
|
# isPaused = false; # Default is not paused
|
|
}
|
|
];
|
|
}
|