comparison host/admin/src/private/lib/monitor.luan @ 1995:301a6561fb6b

add host/admin
author Franklin Schmidt <fschmidt@gmail.com>
date Fri, 04 Jul 2025 10:25:38 -0600
parents
children
comparison
equal deleted inserted replaced
1994:035996323891 1995:301a6561fb6b
1 local Luan = require "luan:Luan.luan"
2 local error = Luan.error
3 local Io = require "luan:Io.luan"
4 local uri = Io.uri or error()
5 local String = require "luan:String.luan"
6 local trim = String.trim or error()
7 local Time = require "luan:Time.luan"
8 local Thread = require "luan:Thread.luan"
9 local Http = require "luan:http/Http.luan"
10 local Config = require "site:/private/Config.luan"
11 local Utils = require "site:/private/lib/Utils.luan"
12 local Logging = require "luan:logging/Logging.luan"
13 local logger = Logging.logger "monitor"
14
15
16 local who_monitors_who = {
17 -- ["admin.me.luan.software"] = "admin.me.luan.software"
18 ["admin.s1.luan.software"] = "admin.s2.luan.software"
19 ["admin.s2.luan.software"] = "admin.s1.luan.software"
20 }
21 local frequency = Time.period{minutes=2}
22
23 --[[ -- for development
24 --who_monitors_who["admin.me.luan.software"] = "admin.me.luan.software"
25 frequency = Time.period{seconds=20}
26 Config.email_to = "fschmidt@gmail.com"
27 --]]
28
29
30
31 local domain = who_monitors_who[Utils.domain]
32
33 if domain == nil then
34 logger.info "nothing to monitor"
35 return true
36 end
37
38 local url = "https://"..domain.."/hi"
39 local options = { time_out = Time.period{seconds=20} }
40
41 local function init_check()
42 local fails = 0
43
44 return function()
45 try
46 local page = uri(url,options).read_text()
47 fails = 0
48 logger.info(domain.." is okay")
49 catch e
50 logger.error("Error connecting to "..domain..": "..e.get_message())
51 fails = fails + 1
52 if fails < 2 then return end
53 try
54 local s = Utils.ssh(domain,"/Users/administrator/luan/host/restart.sh monitoring")
55 if trim(s) == "stopped with stop script" then
56 logger.info("stopped with stop script")
57 else
58 logger.error("restart successful\n"..s)
59 Utils.send_mail {
60 Subject = domain.." restarted"
61 body = s
62 }
63 end
64 fails = 0
65 catch e
66 logger.error("restart failed: "..e.get_message())
67 if fails < 5 then
68 Utils.send_mail {
69 Subject = domain.." restart failed"
70 body = e.get_message()
71 }
72 end
73 end
74 end
75 end
76 end
77
78 Thread.schedule_closure(init_check,{repeating_delay=frequency})
79
80 logger.info("monitoring "..domain)
81
82 return true