summaryrefslogtreecommitdiffstats
path: root/lass/2configs/monitoring
diff options
context:
space:
mode:
Diffstat (limited to 'lass/2configs/monitoring')
-rw-r--r--lass/2configs/monitoring/client.nix106
-rw-r--r--lass/2configs/monitoring/server.nix37
2 files changed, 44 insertions, 99 deletions
diff --git a/lass/2configs/monitoring/client.nix b/lass/2configs/monitoring/client.nix
index e879d6960..b8c245215 100644
--- a/lass/2configs/monitoring/client.nix
+++ b/lass/2configs/monitoring/client.nix
@@ -1,94 +1,26 @@
{pkgs, config, ...}:
with import <stockholm/lib>;
{
- lass.telegraf = {
+ services.telegraf = {
enable = true;
- interval = "1s";
-
- outputs = ''
- [outputs.influxdb]
- urls = ["http://prism:8086"]
- database = "telegraf_db"
- user_agent = "telegraf"
- '';
- inputs = [
- ''
- [cpu]
- percpu = false
- totalcpu = true
- drop = ["cpu_time"]
- ''
- ''
- [[inputs.mem]]
- ''
- ''
- [[inputs.ping]]
- urls = ["8.8.8.8"]
- ''
- ''
- [[inputs.net]]
- ''
- ''
- [[inputs.dns_query]]
- servers = ["8.8.8.8"]
- ''
- ];
- };
- systemd.services.telegraf.path = with pkgs; [
- iputils
- lm_sensors
- ];
-
- services.collectd = {
- enable = true;
- autoLoadPlugin = true;
- extraConfig = ''
- Hostname ${config.krebs.build.host.name}
- LoadPlugin load
- LoadPlugin disk
- LoadPlugin memory
- Interval 30.0
-
- LoadPlugin interface
- <Plugin "interface">
- Interface "*Link"
- Interface "lo"
- Interface "vboxnet*"
- Interface "virbr*"
- IgnoreSelected true
- </Plugin>
-
- LoadPlugin df
- <Plugin "df">
- MountPoint "/nix/store"
- FSType "tmpfs"
- FSType "binfmt_misc"
- FSType "debugfs"
- FSType "mqueue"
- FSType "hugetlbfs"
- FSType "systemd-1"
- FSType "cgroup"
- FSType "securityfs"
- FSType "ramfs"
- FSType "proc"
- FSType "devpts"
- FSType "devtmpfs"
- MountPoint "/var/lib/docker/devicemapper"
- IgnoreSelected true
- </Plugin>
-
- LoadPlugin cpu
- <Plugin cpu>
- ReportByCpu true
- ReportByState true
- ValuesPercentage true
- </Plugin>
-
- LoadPlugin network
- <Plugin "network">
- Server "prism" "25826"
- </Plugin>
- '';
+ extraConfig = {
+ agent.interval = "1s";
+ outputs = {
+ influxdb = {
+ urls = ["http://prism:8086"];
+ database = "telegraf_db";
+ user_agent = "telegraf";
+ };
+ };
+ inputs = {
+ cpu = {
+ percpu = false;
+ totalcpu = true;
+ };
+ mem = {};
+ net = {};
+ };
+ };
};
}
diff --git a/lass/2configs/monitoring/server.nix b/lass/2configs/monitoring/server.nix
index 2e1c15ca1..505cb7a17 100644
--- a/lass/2configs/monitoring/server.nix
+++ b/lass/2configs/monitoring/server.nix
@@ -1,9 +1,7 @@
{pkgs, config, ...}:
with import <stockholm/lib>;
{
- services.influxdb = {
- enable = true;
- };
+ services.influxdb.enable = true;
services.influxdb.extraConfig = {
meta.hostname = config.krebs.build.host.name;
@@ -29,24 +27,39 @@ with import <stockholm/lib>;
data="$(${pkgs.jq}/bin/jq -r .message)"
export LOGNAME=prism-alarm
${pkgs.irc-announce}/bin/irc-announce \
- irc.freenode.org 6667 prism-alarm \#krebs-bots "$data" >/dev/null
+ ni.r 6667 prism-alarm \#retiolum "$data" >/dev/null
'';
in {
enable = true;
+ check_db = "telegraf_db";
alarms = {
- test2 = ''
- batch
+ cpu = ''
+ var data = batch
|query(${"'''"}
SELECT mean("usage_user") AS mean
FROM "${config.lass.kapacitor.check_db}"."default"."cpu"
${"'''"})
- .every(3m)
- .period(1m)
+ .period(10m)
+ .every(1m)
+ .groupBy('host')
+ data |alert()
+ .crit(lambda: "mean" > 90)
+ .exec('${echoToIrc}')
+ data |deadman(1.0,5m)
+ .stateChangesOnly()
+ .exec('${echoToIrc}')
+ '';
+ ram = ''
+ var data = batch
+ |query(${"'''"}
+ SELECT mean("used_percent") AS mean
+ FROM "${config.lass.kapacitor.check_db}"."default"."mem"
+ ${"'''"})
+ .period(10m)
+ .every(1m)
.groupBy('host')
- |alert()
- .crit(lambda: "mean" > 90)
- // Whenever we get an alert write it to a file.
- .log('/tmp/alerts.log')
+ data |alert()
+ .crit(lambda: "mean" > 90)
.exec('${echoToIrc}')
'';
};