infrastructure/machines/nixos/storage01/prometheus.nix
Tom Hubrecht 3678c24ed4
Some checks failed
Check workflows / check_workflows (pull_request) Successful in 16s
Check meta / check_dns (pull_request) Successful in 18s
Check meta / check_meta (pull_request) Successful in 19s
Build all the nodes / ap01 (pull_request) Successful in 1m3s
Build all the nodes / build01 (pull_request) Successful in 1m50s
Build all the nodes / bridge01 (pull_request) Successful in 2m6s
Build all the nodes / geo02 (pull_request) Successful in 2m2s
Build all the nodes / geo01 (pull_request) Successful in 2m4s
Build all the nodes / compute01 (pull_request) Failing after 2m13s
Build all the nodes / hypervisor01 (pull_request) Successful in 1m38s
Build all the nodes / netaccess01 (pull_request) Successful in 29s
Build all the nodes / netcore01 (pull_request) Successful in 30s
Build all the nodes / netcore02 (pull_request) Successful in 26s
Build all the nodes / hypervisor02 (pull_request) Successful in 1m45s
Build all the nodes / hypervisor03 (pull_request) Successful in 1m49s
Build all the nodes / tower01 (pull_request) Successful in 1m55s
Build all the nodes / vault01 (pull_request) Successful in 2m29s
Build all the nodes / web02 (pull_request) Successful in 1m36s
Build all the nodes / web01 (pull_request) Successful in 1m55s
Run pre-commit on all files / pre-commit (pull_request) Successful in 38s
Build all the nodes / web03 (pull_request) Successful in 1m42s
Build all the nodes / rescue01 (pull_request) Successful in 4m54s
Build the shell / build-shell (pull_request) Successful in 2m8s
Build all the nodes / storage01 (pull_request) Successful in 4m43s
Build all the nodes / ap01 (push) Successful in 31s
Build all the nodes / netaccess01 (push) Successful in 18s
Build all the nodes / netcore01 (push) Successful in 19s
Build all the nodes / netcore02 (push) Successful in 18s
Build all the nodes / bridge01 (push) Successful in 1m34s
Build all the nodes / hypervisor02 (push) Successful in 1m39s
Build all the nodes / geo02 (push) Successful in 1m40s
Build all the nodes / hypervisor01 (push) Successful in 1m53s
Build all the nodes / hypervisor03 (push) Successful in 2m0s
Build all the nodes / geo01 (push) Successful in 2m1s
Build all the nodes / build01 (push) Successful in 2m8s
Build all the nodes / compute01 (push) Failing after 2m8s
Build the shell / build-shell (push) Successful in 40s
Build all the nodes / rescue01 (push) Successful in 1m24s
Run pre-commit on all files / pre-commit (push) Successful in 49s
Build all the nodes / storage01 (push) Successful in 1m38s
Build all the nodes / tower01 (push) Successful in 1m37s
Build all the nodes / vault01 (push) Successful in 1m41s
Build all the nodes / web02 (push) Successful in 1m38s
Build all the nodes / web03 (push) Successful in 1m42s
Build all the nodes / web01 (push) Successful in 3m1s
feat(modules/dgn-monitoring): Replace dgn-node-monitoring
2025-02-09 00:30:46 +01:00

92 lines
2.1 KiB
Nix

# SPDX-FileCopyrightText: 2024 Maurice Debray <maurice.debray@dgnum.eu>
#
# SPDX-License-Identifier: EUPL-1.2
{
config,
serverNodes,
lib,
...
}:
let
host = "prometheus.dgnum.eu";
port = 9091;
nodeExporterConfigs = lib.flatten (
lib.mapAttrsToList (
node:
{ config, ... }:
lib.optional config.dgn-monitoring.exporters.enable {
targets = map (p: "${node}.dgnum:${builtins.toString p}") (
builtins.attrValues config.dgn-monitoring.exporters.ports
);
labels = {
host = node;
};
}
) serverNodes
);
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" ];
}