]> git.immae.eu Git - perso/Immae/Config/Nix.git/blob - systems/dilion/monitoring.nix
Add monitoring script with smartctl
[perso/Immae/Config/Nix.git] / systems / dilion / monitoring.nix
1 { lib, pkgs, config, name, monitoring, ... }:
2 let
3 hostFQDN = config.hostEnv.fqdn;
4 zfs_snapshot = name: {
5 service_description = "ZFS snapshot ${name} happened not too long ago";
6 use = "local-service";
7 check_command = ["check_zfs_snapshot" name];
8 __passive_servicegroups = "webstatus-resources";
9 };
10 in
11 {
12 config.myServices.monitoring.activatedPlugins = [ "memory" "command" "bandwidth" "mdadm" "zfs" "notify-secondary" "smartctl" ];
13 config.myServices.monitoring.objects = lib.mkMerge [
14 (monitoring.lib.objectsCommon {
15 inherit hostFQDN;
16 hostName = name;
17 master = false;
18 processWarn = "250"; processAlert = "400";
19 loadWarn = "1.0"; loadAlert = "1.2";
20 interface = builtins.head (builtins.attrNames config.networking.interfaces);
21 })
22
23 {
24 service = [
25 {
26 service_description = "Disk /dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A4B is sane";
27 use = "local-service";
28 check_command = [ "check_smartctl" "/dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A4B" ];
29 __passive_servicegroups = "webstatus-resources";
30
31 check_interval = 60;
32 }
33 {
34 service_description = "Disk /dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A70 is sane";
35 use = "local-service";
36 check_command = [ "check_smartctl" "/dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A70" ];
37 __passive_servicegroups = "webstatus-resources";
38
39 check_interval = 60;
40 }
41 {
42 service_description = "Disk /dev/disk/by-id/ata-TOSHIBA_MG04ACA400N_96K1K87YFVLC is sane";
43 use = "local-service";
44 check_command = [ "check_smartctl" "/dev/disk/by-id/ata-TOSHIBA_MG04ACA400N_96K1K87YFVLC" ];
45 __passive_servicegroups = "webstatus-resources";
46
47 check_interval = 60;
48 }
49 {
50 service_description = "Disk /dev/disk/by-id/ata-WDC_WD3000FYYZ-01UL1B2_WD-WMC1F0E4X6WP is sane";
51 use = "local-service";
52 check_command = [ "check_smartctl" "/dev/disk/by-id/ata-WDC_WD3000FYYZ-01UL1B2_WD-WMC1F0E4X6WP" ];
53 __passive_servicegroups = "webstatus-resources";
54
55 check_interval = 60;
56 }
57 {
58 service_description = "No mdadm array is degraded";
59 use = "local-service";
60 check_command = ["check_mdadm"];
61 __passive_servicegroups = "webstatus-resources";
62 }
63 {
64 service_description = "No ZFS pool is degraded";
65 use = "local-service";
66 check_command = ["check_zfs"];
67 __passive_servicegroups = "webstatus-resources";
68 }
69 (zfs_snapshot "zpool/backup/eldiron/zpool/root")
70 (zfs_snapshot "zpool/backup/eldiron/zpool/root/etc")
71 (zfs_snapshot "zpool/backup/eldiron/zpool/root/var")
72 ];
73 }
74 ];
75 }