stockholm/krebs/2configs/shack/prometheus/server.nix
2019-06-20 22:25:46 +02:00

196 lines
6.3 KiB
Nix

{ pkgs, lib, config, ... }:
# from https://gist.github.com/globin/02496fd10a96a36f092a8e7ea0e6c7dd
{
networking = {
firewall.allowedTCPPorts = [
3000 # grafana
9090 # prometheus
9093 # alertmanager
];
useDHCP = true;
};
services = {
prometheus = {
enable = true;
extraFlags = [
"-storage.local.retention 8760h"
"-storage.local.series-file-shrink-ratio 0.3"
"-storage.local.memory-chunks 2097152"
"-storage.local.max-chunks-to-persist 1048576"
"-storage.local.index-cache-size.fingerprint-to-metric 2097152"
"-storage.local.index-cache-size.fingerprint-to-timerange 1048576"
"-storage.local.index-cache-size.label-name-to-label-values 2097152"
"-storage.local.index-cache-size.label-pair-to-fingerprints 41943040"
];
alertmanagerURL = [ "http://localhost:9093" ];
rules = [
''
ALERT node_down
IF up == 0
FOR 5m
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}}: Node is down.",
description = "{{$labels.alias}} has been down for more than 5 minutes."
}
ALERT node_systemd_service_failed
IF node_systemd_unit_state{state="failed"} == 1
FOR 4m
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}}: Service {{$labels.name}} failed to start.",
description = "{{$labels.alias}} failed to (re)start service {{$labels.name}}."
}
ALERT node_filesystem_full_90percent
IF sort(node_filesystem_free{device!="ramfs"} < node_filesystem_size{device!="ramfs"} * 0.1) / 1024^3
FOR 5m
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}}: Filesystem is running out of space soon.",
description = "{{$labels.alias}} device {{$labels.device}} on {{$labels.mountpoint}} got less than 10% space left on its filesystem."
}
ALERT node_filesystem_full_in_4h
IF predict_linear(node_filesystem_free{device!="ramfs"}[1h], 4*3600) <= 0
FOR 5m
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}}: Filesystem is running out of space in 4 hours.",
description = "{{$labels.alias}} device {{$labels.device}} on {{$labels.mountpoint}} is running out of space of in approx. 4 hours"
}
ALERT node_filedescriptors_full_in_3h
IF predict_linear(node_filefd_allocated[1h], 3*3600) >= node_filefd_maximum
FOR 20m
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}} is running out of available file descriptors in 3 hours.",
description = "{{$labels.alias}} is running out of available file descriptors in approx. 3 hours"
}
ALERT node_load1_90percent
IF node_load1 / on(alias) count(node_cpu{mode="system"}) by (alias) >= 0.9
FOR 1h
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}}: Running on high load.",
description = "{{$labels.alias}} is running with > 90% total load for at least 1h."
}
ALERT node_cpu_util_90percent
IF 100 - (avg by (alias) (irate(node_cpu{mode="idle"}[5m])) * 100) >= 90
FOR 1h
LABELS {
severity="page"
}
ANNOTATIONS {
summary = "{{$labels.alias}}: High CPU utilization.",
description = "{{$labels.alias}} has total CPU utilization over 90% for at least 1h."
}
ALERT node_ram_using_90percent
IF node_memory_MemFree + node_memory_Buffers + node_memory_Cached < node_memory_MemTotal * 0.1
FOR 30m
LABELS {
severity="page"
}
ANNOTATIONS {
summary="{{$labels.alias}}: Using lots of RAM.",
description="{{$labels.alias}} is using at least 90% of its RAM for at least 30 minutes now.",
}
''
];
scrapeConfigs = [
{
job_name = "node";
scrape_interval = "10s";
static_configs = [
{
targets = [
"localhost:9100"
];
labels = {
alias = "wolf.shack";
};
}
{
targets = [
"localhost:9130"
];
labels = {
alias = "unifi.shack";
};
}
{
targets = [
"10.42.22.184:9100" # puyak.shack
];
labels = {
alias = "puyak.shack";
};
}
{
targets = [
"phenylbutazon.shack:9100"
];
labels = {
alias = "phenylbutazon.shack";
};
}
{
targets = [
"ibuprofen.shack:9100"
];
labels = {
alias = "ibuprofen.shack";
};
}
];
}
];
alertmanager = {
enable = true;
listenAddress = "0.0.0.0";
configuration = {
"global" = {
"smtp_smarthost" = "smtp.example.com:587";
"smtp_from" = "alertmanager@example.com";
};
"route" = {
"group_by" = [ "alertname" "alias" ];
"group_wait" = "30s";
"group_interval" = "2m";
"repeat_interval" = "4h";
"receiver" = "team-admins";
};
"receivers" = [
{
"name" = "team-admins";
"email_configs" = [
{
"to" = "devnull@example.com";
"send_resolved" = true;
}
];
"webhook_configs" = [
{
"url" = "https://example.com/prometheus-alerts";
"send_resolved" = true;
}
];
}
];
};
};
};
};
}