summaryrefslogtreecommitdiffstats
path: root/lass/2configs/monitoring
diff options
context:
space:
mode:
Diffstat (limited to 'lass/2configs/monitoring')
-rw-r--r--lass/2configs/monitoring/client.nix26
-rw-r--r--lass/2configs/monitoring/monit-alarms.nix44
-rw-r--r--lass/2configs/monitoring/node-exporter.nix6
-rw-r--r--lass/2configs/monitoring/prometheus-server.nix50
-rw-r--r--lass/2configs/monitoring/server.nix87
5 files changed, 48 insertions, 165 deletions
diff --git a/lass/2configs/monitoring/client.nix b/lass/2configs/monitoring/client.nix
deleted file mode 100644
index b8c245215..000000000
--- a/lass/2configs/monitoring/client.nix
+++ /dev/null
@@ -1,26 +0,0 @@
-{pkgs, config, ...}:
-with import <stockholm/lib>;
-{
- services.telegraf = {
- enable = true;
-
- extraConfig = {
- agent.interval = "1s";
- outputs = {
- influxdb = {
- urls = ["http://prism:8086"];
- database = "telegraf_db";
- user_agent = "telegraf";
- };
- };
- inputs = {
- cpu = {
- percpu = false;
- totalcpu = true;
- };
- mem = {};
- net = {};
- };
- };
- };
-}
diff --git a/lass/2configs/monitoring/monit-alarms.nix b/lass/2configs/monitoring/monit-alarms.nix
deleted file mode 100644
index 2cfc292e5..000000000
--- a/lass/2configs/monitoring/monit-alarms.nix
+++ /dev/null
@@ -1,44 +0,0 @@
-{pkgs, config, ...}:
-with import <stockholm/lib>;
-let
- echoToIrc = msg:
- pkgs.writeDash "echo_irc" ''
- set -euf
- export LOGNAME=prism-alarm
- ${pkgs.irc-announce}/bin/irc-announce \
- irc.r 6667 ${config.networking.hostName}-alarm \#noise "${msg}" >/dev/null
- '';
-
-in {
- krebs.monit = {
- enable = true;
- http.enable = true;
- alarms = {
- nirwanabluete = {
- test = "${pkgs.curl}/bin/curl -sf 'https://nirwanabluete.de/'";
- alarm = echoToIrc "test nirwanabluete failed";
- };
- ubik = {
- test = "${pkgs.curl}/bin/curl -sf 'https://ubikmedia.de'";
- alarm = echoToIrc "test ubik failed";
- };
- cac-panel = {
- test = "${pkgs.curl}/bin/curl -sf 'https://panel.cloudatcost.com/login.php'";
- alarm = echoToIrc "test cac-panel failed";
- };
- radio = {
- test = pkgs.writeBash "check_stream" ''
- ${pkgs.curl}/bin/curl -sif http://lassul.us:8000/radio.ogg \
- | ${pkgs.gawk}/bin/awk '/^\r$/{exit}{print $0}' \
- | ${pkgs.gnugrep}/bin/grep -q "200 OK" || exit "''${PIPESTATUS[0]}"
- '';
- alarm = echoToIrc "test radio failed";
- };
- };
- };
-
- krebs.iptables.tables.filter.INPUT.rules = [
- { predicate = "-p tcp -i retiolum --dport 9093"; target = "ACCEPT"; }
- ];
-}
-
diff --git a/lass/2configs/monitoring/node-exporter.nix b/lass/2configs/monitoring/node-exporter.nix
index 8c27e90d4..561e3a25c 100644
--- a/lass/2configs/monitoring/node-exporter.nix
+++ b/lass/2configs/monitoring/node-exporter.nix
@@ -1,7 +1,9 @@
{ config, lib, pkgs, ... }:
{
- networking.firewall.allowedTCPPorts = [ 9100 ];
-
+ krebs.iptables.tables.filter.INPUT.rules = [
+ { predicate = "-i retiolum -p tcp --dport 9100 -s ${config.krebs.hosts.prism.nets.retiolum.ip4.addr}"; target = "ACCEPT"; v6 = false; }
+ { predicate = "-i retiolum -p tcp --dport 9100 -s ${config.krebs.hosts.prism.nets.retiolum.ip6.addr}"; target = "ACCEPT"; v4 = false; }
+ ];
services.prometheus.exporters = {
node = {
enable = true;
diff --git a/lass/2configs/monitoring/prometheus-server.nix b/lass/2configs/monitoring/prometheus-server.nix
index d56d7e552..e16d421a0 100644
--- a/lass/2configs/monitoring/prometheus-server.nix
+++ b/lass/2configs/monitoring/prometheus-server.nix
@@ -9,6 +9,12 @@
# useDHCP = true;
#};
+ krebs.iptables.tables.filter.INPUT.rules = [
+ { predicate = "-i retiolum -p tcp --dport 3000"; target = "ACCEPT"; }
+ { predicate = "-i retiolum -p tcp --dport 9090"; target = "ACCEPT"; }
+ { predicate = "-i retiolum -p tcp --dport 9093"; target = "ACCEPT"; }
+ ];
+
services = {
prometheus = {
enable = true;
@@ -124,11 +130,10 @@
static_configs = [
{
targets = [
- "localhost:9100"
- ];
- labels = {
- alias = "prometheus.example.com";
- };
+ ] ++ map (host: "${host}:9100") (lib.attrNames (lib.filterAttrs (_: host: host.owner.name == "lass" && host.monitoring) config.krebs.hosts));
+ #labels = {
+ # alias = "prometheus.example.com";
+ #};
}
];
}
@@ -159,7 +164,7 @@
];
"webhook_configs" = [
{
- "url" = "https://example.com/prometheus-alerts";
+ "url" = "http://127.0.0.1:14813/prometheus-alerts";
"send_resolved" = true;
}
];
@@ -176,4 +181,37 @@
security = import <secrets/grafana_security.nix>; # { AdminUser = ""; adminPassword = ""}
};
};
+ services.logstash = {
+ enable = true;
+ inputConfig = ''
+ http {
+ port => 14813
+ host => "127.0.0.1"
+ }
+ '';
+ filterConfig = ''
+ if ([alerts]) {
+ ruby {
+ code => '
+ lines = []
+ event["alerts"].each {|p|
+ lines << "#{p["labels"]["instance"]}#{p["annotations"]["summary"]} #{p["status"]}"
+ }
+ event["output"] = lines.join("\n")
+ '
+ }
+ }
+ '';
+ outputConfig = ''
+ file { path => "/tmp/logs.json" codec => "json_lines" }
+ irc {
+ channels => [ "#noise" ]
+ host => "irc.r"
+ nick => "alarm"
+ codec => "json_lines"
+ format => "%{output}"
+ }
+ '';
+ #plugins = [ ];
+ };
}
diff --git a/lass/2configs/monitoring/server.nix b/lass/2configs/monitoring/server.nix
deleted file mode 100644
index adaecde2c..000000000
--- a/lass/2configs/monitoring/server.nix
+++ /dev/null
@@ -1,87 +0,0 @@
-{pkgs, config, ...}:
-with import <stockholm/lib>;
-{
- services.influxdb.enable = true;
-
- services.influxdb.extraConfig = {
- meta.hostname = config.krebs.build.host.name;
- # meta.logging-enabled = true;
- http.bind-address = ":8086";
- admin.bind-address = ":8083";
- http.log-enabled = false;
- monitoring = {
- enabled = false;
- # write-interval = "24h";
- };
- collectd = [{
- enabled = true;
- typesdb = "${pkgs.collectd}/share/collectd/types.db";
- database = "collectd_db";
- port = 25826;
- }];
- };
-
- krebs.kapacitor =
- let
- db = "telegraf_db";
- echoToIrc = pkgs.writeDash "echo_irc" ''
- set -euf
- data="$(${pkgs.jq}/bin/jq -r .message)"
- export LOGNAME=prism-alarm
- ${pkgs.irc-announce}/bin/irc-announce \
- irc.r 6667 prism-alarm \#noise "$data" >/dev/null
- '';
- in {
- enable = true;
- alarms = {
- cpu = {
- database = db;
- text = ''
- var data = batch
- |query(${"'''"}
- SELECT mean("usage_user") AS mean
- FROM "${db}"."default"."cpu"
- ${"'''"})
- .period(10m)
- .every(1m)
- .groupBy('host')
- data |alert()
- .crit(lambda: "mean" > 90)
- .exec('${echoToIrc}')
- data |deadman(1.0,5m)
- .stateChangesOnly()
- .exec('${echoToIrc}')
- '';
- };
- ram = {
- database = db;
- text = ''
- var data = batch
- |query(${"'''"}
- SELECT mean("used_percent") AS mean
- FROM "${db}"."default"."mem"
- ${"'''"})
- .period(10m)
- .every(1m)
- .groupBy('host')
- data |alert()
- .crit(lambda: "mean" > 90)
- .exec('${echoToIrc}')
- '';
- };
- };
- };
-
- services.grafana = {
- enable = true;
- addr = "0.0.0.0";
- auth.anonymous.enable = true;
- security = import <secrets/grafana_security.nix>; # { AdminUser = ""; adminPassword = ""}
- };
-
- krebs.iptables.tables.filter.INPUT.rules = [
- { predicate = "-p tcp -i retiolum --dport 8086"; target = "ACCEPT"; }
- { predicate = "-p tcp -i retiolum --dport 3000"; target = "ACCEPT"; }
- { predicate = "-p udp -i retiolum --dport 25826"; target = "ACCEPT"; }
- ];
-}
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-18 13:23:11 +0000