contrib/statusd/statusd_netmon.lua

   1 -- Authors: Sadrul Habib Chowdhury <imadil@gmail.com>
   2 -- License: Public domain
   3 -- Last Changed: Unknown
   4 --
   5 -- statusd_netmon.lua: monitor the speed of a network interface
   6 --
   7 -- Thanx to Tuomo for pointing out a problem in the previous script.
   8 --
   9 -- In case this doesn't work for someone, do let me know :)
  10 --
  11 -- Author
  12 -- Sadrul Habib Chowdhury (Adil)
  13 -- imadil at gmail dot com
  14 --
  15 -- Support for per-stat monitors and thresholds added by Jeremy
  16 -- Hankins.
  17 --
  18 --
  19 -- Monitor values available with this monitor:
  20 --
  21 -- netmon
  22 -- netmon_kbsin
  23 -- netmon_kbsout
  24 -- netmon_avgin
  25 -- netmon_avgout
  26 -- netmon_count
  27 --
  28 -- To use the average values or the count you need show_avg and
  29 -- show_count turned on, respectively.  If you want the default format
  30 -- (which you get with %netmon) but with colors for important and
  31 -- critical thresholds, try:
  32 --
  33 -- %netmon_kbsin/%netmon_kbsout (%netmon_avgin/%netmon_avgout)
  34
  35 if not statusd_netmon then
  36   statusd_netmon = {
  37       device = "eth0",
  38       show_avg = 1,       -- show average stat?
  39       avg_sec = 60,       -- default, shows average of 1 minute
  40       show_count = 0,     -- show tcp connection count?
  41       interval = 1*1000,  -- update every second
  42
  43       -- Threshold information.  These values should likely be tweaked to
  44       -- suit local conditions.
  45       important = {
  46         kbsin = 1/10,
  47         kbsout = 1/10,
  48         avgin = 1/10,
  49         avgout = 1/10,
  50         count = 4,
  51       },
  52
  53       critical = {
  54         kbsin = 30,
  55         kbsout = 30,
  56         avgin = 5,
  57         avgout = 5,
  58         count = 50,
  59     }
  60 }
  61 end
  62
  63 local timer = nil       -- the timer
  64 local positions = {}    -- positions where the entries will be
  65 local last = {}         -- the last readings
  66 local history_in = {}   -- history to calculate the average
  67 local history_out = {}
  68 local total_in, total_out = 0, 0
  69 local counter = 0       --
  70
  71 local settings = table.join(statusd.get_config("netmon"), statusd_netmon)
  72
  73 --
  74 -- tokenize the string
  75 --
  76 local function tokenize(str)
  77     local ret = {}
  78     local i = 0
  79     local k = nil
  80
  81     for k in string.gfind(str, '(%w+)') do
  82         ret[i] = k
  83         i = i + 1
  84     end
  85     return ret
  86 end
  87
  88 --
  89 -- get the connection count
  90 --
  91 local function get_connection_count()
  92     local f = io.popen('netstat -st', 'r')
  93     if not f then return nil end
  94
  95     local output = f:read('*a')
  96     if string.len(output) == 0 then return nil end
  97
  98     local s, e, connections =
  99         string.find(output, '%s+(%d+)%s+connections established%s')
 100     f:close()
 101     return tonumber(connections)
 102 end
 103
 104 --
 105 -- calculate the average
 106 --
 107 local function calc_avg(lin, lout)
 108     if counter == settings.avg_sec then
 109         counter = 0
 110     end
 111
 112     total_in = total_in - history_in[counter] + lin
 113     history_in[counter] = lin
 114
 115     total_out = total_out - history_out[counter] + lout
 116     history_out[counter] = lout
 117
 118     counter = counter + 1
 119
 120     return total_in/settings.avg_sec, total_out/settings.avg_sec
 121 end
 122
 123 --
 124 -- parse the information
 125 --
 126 local function parse_netmon_info()
 127     local s
 128     local lin, lout
 129
 130     for s in io.lines('/proc/net/dev') do
 131         local f = string.find(s, settings.device)
 132         if f then
 133             local t = tokenize(s)
 134             return t[positions[0]], t[positions[1]]
 135        end
 136     end
 137     return nil, nil
 138 end
 139
 140 --
 141 -- Return a hint value for the given meter
 142 --
 143 local function get_hint(meter, val)
 144     local hint = "normal"
 145     local crit = settings.critical[meter]
 146     local imp = settings.important[meter]
 147     if crit and val > crit then
 148         hint = "critical"
 149     elseif imp and val > imp then
 150         hint = "important"
 151     end
 152     return hint
 153 end
 154
 155 --
 156 -- update the netmon monitor
 157 --
 158 local function update_netmon_info()
 159     local s
 160     local lin, lout
 161
 162     local function fmt(num)
 163         return(string.format("%.1fK", num))
 164     end
 165
 166     lin, lout = parse_netmon_info()
 167     if not lin or not lout then
 168         -- you should never reach here
 169         statusd.inform("netmon", "oops")
 170         statusd.inform("netmon_hint", "critical")
 171         return
 172     end
 173
 174     last[0], lin = lin, lin - last[0]
 175     last[1], lout = lout, lout - last[1]
 176
 177     local kbsin = lin/1024
 178     local kbsout = lout/1024
 179
 180     local output = string.format("%.1fK/%.1fK", kbsin, kbsout)
 181
 182     if settings.show_avg == 1 then
 183         local avgin, avgout = calc_avg(lin/1024, lout/1024)
 184         output = output .. string.format(" (%.1fK/%.1fK)", avgin, avgout)
 185
 186         statusd.inform("netmon_avgin", fmt(avgin))
 187         statusd.inform("netmon_avgin_hint",
 188                        get_hint("avgin", avgin))
 189         statusd.inform("netmon_avgout", fmt(avgout))
 190         statusd.inform("netmon_avgout_hint",
 191                        get_hint("avgout", avgout))
 192     end
 193     if settings.show_count == 1 then
 194         local count = get_connection_count()
 195         if count then
 196             output = "[" .. tostring(count) .. "] " .. output
 197
 198             statusd.inform("netmon_count", tostring(count))
 199             statusd.inform("netmon_count_hint",
 200                            get_hint("count", count))
 201         else
 202             output = "[?] " .. output
 203             statusd.inform("netmon_count", "?")
 204             statusd.inform("netmon_count_hint", "critical")
 205         end
 206     end
 207
 208     statusd.inform("netmon_kbsin", fmt(kbsin))
 209     statusd.inform("netmon_kbsin_hint",
 210                    get_hint("kbsin", kbsin))
 211     statusd.inform("netmon_kbsout", fmt(kbsout))
 212     statusd.inform("netmon_kbsout_hint",
 213                    get_hint("kbsout", kbsout))
 214     statusd.inform("netmon", output)
 215
 216     timer:set(settings.interval, update_netmon_info)
 217 end
 218
 219 --
 220 -- is everything ok to begin with?
 221 --
 222 local function sanity_check()
 223     local f = io.open('/proc/net/dev', 'r')
 224     local e
 225
 226     if not f then
 227         return false
 228     end
 229
 230     local s = f:read('*line')
 231     s = f:read('*line')         -- the second line, which should give
 232                                 -- us the positions of the info we seek
 233
 234     local t = tokenize(s)
 235     local n = #(t)
 236     local i = 0
 237
 238     for i = 0,n do
 239         if t[i] == "bytes" then
 240             positions[0] = i
 241             break
 242         end
 243     end
 244
 245     i = positions[0] + 1
 246
 247     for i=i,n do
 248         if t[i] == "bytes" then
 249             positions[1] = i
 250             break
 251         end
 252     end
 253
 254     if not positions[0] or not positions[1] then
 255         return false
 256     end
 257
 258     s = f:read('*a')        -- read the whole file
 259     if not string.find(s, settings.device) then
 260         return false        -- the device does not exist
 261     end
 262
 263     return true
 264 end
 265
 266 --
 267 -- start the timer
 268 --
 269 local function init_netmon_monitor()
 270     if sanity_check() then
 271         timer = statusd.create_timer()
 272         last[0], last[1] = parse_netmon_info()
 273
 274         if settings.show_avg == 1 then
 275             for i=0,settings.avg_sec-1 do
 276                 history_in[i], history_out[i] = 0, 0
 277             end
 278         end
 279
 280         statusd.inform("netmon_template", "xxxxxxxxxxxxxxxxxxxxxxx")
 281         update_netmon_info()
 282     else
 283         statusd.inform("netmon", "oops")
 284         statusd.inform("netmon_hint", "critical")
 285     end
 286 end
 287
 288 init_netmon_monitor()
 289