stockholm/makefu/2configs/logging/central-stats-server.nix

73 lines
2.0 KiB
Nix
Raw Normal View History

2017-01-09 15:02:02 +00:00
{pkgs, config, ...}:
with import <stockholm/lib>;
let
2017-01-09 15:03:36 +00:00
collectd-port = 25826;
influx-port = 8086;
2017-01-09 15:02:02 +00:00
grafana-port = 3000; # TODO nginx forward
in {
2017-02-03 19:45:27 +00:00
imports = [
../../lass/3modules/kapacitor.nix
];
2017-01-09 15:03:36 +00:00
services.grafana.enable = true;
2017-01-09 15:02:02 +00:00
services.grafana.addr = "0.0.0.0";
services.influxdb.enable = true;
# forward these via nginx
services.influxdb.extraConfig = {
meta.hostname = config.krebs.build.host.name;
# meta.logging-enabled = true;
http.bind-address = ":${toString influx-port}";
admin.bind-address = ":8083";
monitoring = {
enabled = false;
# write-interval = "24h";
};
2017-01-09 15:03:36 +00:00
collectd = [{
2017-01-09 15:02:02 +00:00
enabled = true;
typesdb = "${pkgs.collectd}/share/collectd/types.db";
database = "collectd_db";
port = collectd-port;
}];
};
2017-02-03 19:45:27 +00:00
lass.kapacitor =
let
echoToIrc = pkgs.writeDash "echo_irc" ''
set -euf
data="$(${pkgs.jq}/bin/jq -r .message)"
export LOGNAME=malarm
${pkgs.irc-announce}/bin/irc-announce \
irc.freenode.org 6667 malarm \#krebs-bots "$data" >/dev/null
'';
in {
enable = true;
alarms = {
cpu_deadman = ''
var data = batch
|query('''
SELECT mean("value") AS mean
FROM "collectd_db"."default"."cpu_value"
WHERE "type_instance" = 'idle' AND "type" = 'percent' fill(0)
''')
.period(10m)
.every(1m)
.groupBy('host')
data |alert()
.crit(lambda: "mean" < 50)
.stateChangesOnly()
.exec('${echoToIrc}')
data |deadman(1.0,5m)
.stateChangesOnly()
.exec('${echoToIrc}')
'';
};
};
2017-01-09 15:02:02 +00:00
networking.firewall.extraCommands = ''
iptables -A INPUT -i retiolum -p udp --dport ${toString collectd-port} -j ACCEPT
iptables -A INPUT -i retiolum -p tcp --dport ${toString influx-port} -j ACCEPT
iptables -A INPUT -i retiolum -p tcp --dport ${toString grafana-port} -j ACCEPT
'';
}