infrastructure/machines/nixos/storage01/prometheus.nix
Tom Hubrecht ecbad0a638
All checks were successful
Check workflows / check_workflows (push) Successful in 21s
Run pre-commit on all files / check (push) Successful in 24s
Check meta / check_dns (pull_request) Successful in 19s
Check meta / check_meta (pull_request) Successful in 18s
Check workflows / check_workflows (pull_request) Successful in 19s
Build all the nodes / bridge01 (pull_request) Successful in 1m13s
Build all the nodes / geo01 (pull_request) Successful in 1m14s
Build all the nodes / compute01 (pull_request) Successful in 1m44s
Build all the nodes / geo02 (pull_request) Successful in 1m12s
Build all the nodes / rescue01 (pull_request) Successful in 1m30s
Build all the nodes / storage01 (pull_request) Successful in 1m29s
Build all the nodes / vault01 (pull_request) Successful in 1m26s
Build all the nodes / web02 (pull_request) Successful in 1m19s
Run pre-commit on all files / check (pull_request) Successful in 24s
Build all the nodes / web01 (pull_request) Successful in 1m56s
Build all the nodes / web03 (pull_request) Successful in 1m25s
chore: Abstract machines and modules
This adds subdirectories for the different types of systems, for the
modules and the machines
2024-12-08 13:39:10 +01:00

86 lines
1.9 KiB
Nix

{
config,
nodes,
lib,
...
}:
let
host = "prometheus.dgnum.eu";
port = 9091;
nodeExporterConfigs = lib.flatten (
lib.mapAttrsToList (
node:
{ config, ... }:
lib.optional config.dgn-node-monitoring.enable {
targets = [ "${node}.dgnum:${builtins.toString config.dgn-node-monitoring.port}" ];
labels = {
host = node;
};
}
) nodes
);
in
{
services.prometheus = {
enable = true;
inherit port;
checkConfig = "syntax-only";
enableReload = true;
listenAddress = "127.0.0.1";
webConfigFile = config.age.secrets."prometheus-web_config_file".path;
webExternalUrl = "https://${host}";
retentionTime = "1y";
extraFlags = [ "--storage.tsdb.retention.size=20GB" ];
globalConfig = {
scrape_interval = "15s"; # if you change this settings, please do it in grafana also
};
scrapeConfigs = [
{
job_name = "prometheus";
static_configs = [ { targets = [ "localhost:9090" ]; } ];
}
{
job_name = "node_exporter";
static_configs = nodeExporterConfigs;
}
{
job_name = "uptime_kuma";
scheme = "https";
static_configs = [ { targets = [ "status.dgnum.eu" ]; } ];
basic_auth = {
username = "prometheus";
password_file = config.age.secrets."prometheus-uptime-kuma-apikey".path;
};
}
{
job_name = "hyp01_ups";
metrics_path = "/ups_metrics";
static_configs = [ { targets = [ "100.80.255.180:9199" ]; } ];
}
{
job_name = "garage";
static_configs = [ { targets = [ "localhost:3903" ]; } ];
bearer_token_file = config.age.secrets."prometheus-garage_api".path;
}
];
};
dgn-web.simpleProxies.prometheus = {
inherit host port;
proxyWebsockets = true;
};
age-secrets.autoMatch = [ "prometheus" ];
}