Files
packages/utils/prometheus-node-exporter-lua/files/usr/bin/prometheus-node-exporter-lua
Etienne Champetier 7b1b69ca6c prometheus-node-exporter-lua: rewrite wifi scraper
On my bullet m2, scrape duration goes from between 0.2 and 0.5 to a stable 0.025
We also don't depend on luci anymore

This remove wifi_network_up metric, but this metric was buggy
wifi_network_up{ifname="wlan0-1",ssid="test1",channel="11",mode="Master",bssid="12:34:56:78:9A:BC",country="FR",frequency="2.462"} 1
wifi_network_up{ifname="radio0.network2",ssid="test1",channel="11",mode="Master",country="US",frequency="2.462"} 0

Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
2017-12-08 21:22:16 -08:00

376 lines
11 KiB
Lua
Executable File

#!/usr/bin/lua
-- Metrics web server
-- Copyright (c) 2016 Jeff Schornick <jeff@schornick.org>
-- Copyright (c) 2015 Kevin Lyda
-- Licensed under the Apache License, Version 2.0
socket = require("socket")
-- Allow us to call unpack under both lua5.1 and lua5.2+
local unpack = unpack or table.unpack
-- This table defines the scrapers to run.
-- Each corresponds directly to a scraper_<name> function.
scrapers = { "cpu", "load_averages", "memory", "file_handles", "network",
"network_devices", "time", "uname", "nat", "wifi"}
-- Parsing
function space_split(s)
elements = {}
for element in s:gmatch("%S+") do
table.insert(elements, element)
end
return elements
end
function line_split(s)
elements = {}
for element in s:gmatch("[^\n]+") do
table.insert(elements, element)
end
return elements
end
function get_contents(filename)
local f = io.open(filename, "rb")
local contents = ""
if f then
contents = f:read "*a"
f:close()
end
return contents
end
-- Metric printing
function print_metric(metric, labels, value)
local label_string = ""
if labels then
for label,value in pairs(labels) do
label_string = label_string .. label .. '="' .. value .. '",'
end
label_string = "{" .. string.sub(label_string, 1, -2) .. "}"
end
output(string.format("%s%s %s", metric, label_string, value))
end
function metric(name, mtype, labels, value)
output("# TYPE " .. name .. " " .. mtype)
local outputter = function(labels, value)
print_metric(name, labels, value)
end
if value then
outputter(labels, value)
end
return outputter
end
local ubus = require "ubus"
local iwinfo = require "iwinfo"
function scraper_wifi()
local metric_wifi_network_quality = metric("wifi_network_quality","gauge")
local metric_wifi_network_bitrate = metric("wifi_network_bitrate","gauge")
local metric_wifi_network_noise = metric("wifi_network_noise","gauge")
local metric_wifi_network_signal = metric("wifi_network_signal","gauge")
local metric_wifi_station_signal = metric("wifi_station_signal","gauge")
local metric_wifi_station_tx_packets = metric("wifi_station_tx_packets","gauge")
local metric_wifi_station_rx_packets = metric("wifi_station_rx_packets","gauge")
local u = ubus.connect()
local status = u:call("network.wireless", "status", {})
for dev, dev_table in pairs(status) do
for _, intf in ipairs(dev_table['interfaces']) do
local ifname = intf['ifname']
local iw = iwinfo[iwinfo.type(ifname)]
local labels = {
channel = iw.channel(ifname),
ssid = iw.ssid(ifname),
bssid = iw.bssid(ifname),
mode = iw.mode(ifname),
ifname = ifname,
country = iw.country(ifname),
frequency = iw.frequency(ifname),
device = dev,
}
local qc = iw.quality(ifname) or 0
local qm = iw.quality_max(ifname) or 0
local quality = 0
if qc > 0 and qm > 0 then
quality = math.floor((100 / qm) * qc)
end
metric_wifi_network_quality(labels, quality)
metric_wifi_network_noise(labels, iw.noise(ifname) or 0)
metric_wifi_network_bitrate(labels, iw.bitrate(ifname) or 0)
metric_wifi_network_signal(labels, iw.signal(ifname) or -255)
local assoclist = iw.assoclist(ifname)
for mac, station in pairs(assoclist) do
local labels = {
ifname = ifname,
mac = mac,
}
metric_wifi_station_signal(labels, station.signal)
metric_wifi_station_tx_packets(labels, station.tx_packets)
metric_wifi_station_rx_packets(labels, station.rx_packets)
end
end
end
end
function scraper_cpu()
local stat = get_contents("/proc/stat")
-- system boot time, seconds since epoch
metric("node_boot_time", "gauge", nil, string.match(stat, "btime ([0-9]+)"))
-- context switches since boot (all CPUs)
metric("node_context_switches", "counter", nil, string.match(stat, "ctxt ([0-9]+)"))
-- cpu times, per CPU, per mode
local cpu_mode = {"user", "nice", "system", "idle", "iowait", "irq",
"softirq", "steal", "guest", "guest_nice"}
local i = 0
local cpu_metric = metric("node_cpu", "counter")
while string.match(stat, string.format("cpu%d ", i)) do
local cpu = space_split(string.match(stat, string.format("cpu%d ([0-9 ]+)", i)))
local labels = {cpu = "cpu" .. i}
for ii, mode in ipairs(cpu_mode) do
labels['mode'] = mode
cpu_metric(labels, cpu[ii] / 100)
end
i = i + 1
end
-- interrupts served
metric("node_intr", "counter", nil, string.match(stat, "intr ([0-9]+)"))
-- processes forked
metric("node_forks", "counter", nil, string.match(stat, "processes ([0-9]+)"))
-- processes running
metric("node_procs_running", "gauge", nil, string.match(stat, "procs_running ([0-9]+)"))
-- processes blocked for I/O
metric("node_procs_blocked", "gauge", nil, string.match(stat, "procs_blocked ([0-9]+)"))
end
function scraper_load_averages()
local loadavg = space_split(get_contents("/proc/loadavg"))
metric("node_load1", "gauge", nil, loadavg[1])
metric("node_load5", "gauge", nil, loadavg[2])
metric("node_load15", "gauge", nil, loadavg[3])
end
function scraper_memory()
local meminfo = line_split(get_contents("/proc/meminfo"):gsub("[):]", ""):gsub("[(]", "_"))
for i, mi in ipairs(meminfo) do
local name, size, unit = unpack(space_split(mi))
if unit == 'kB' then
size = size * 1024
end
metric("node_memory_" .. name, "gauge", nil, size)
end
end
function scraper_file_handles()
local file_nr = space_split(get_contents("/proc/sys/fs/file-nr"))
metric("node_filefd_allocated", "gauge", nil, file_nr[1])
metric("node_filefd_maximum", "gauge", nil, file_nr[3])
end
function scraper_network()
-- NOTE: Both of these are missing in OpenWRT kernels.
-- See: https://dev.openwrt.org/ticket/15781
local netstat = get_contents("/proc/net/netstat") .. get_contents("/proc/net/snmp")
-- all devices
local netsubstat = {"IcmpMsg", "Icmp", "IpExt", "Ip", "TcpExt", "Tcp", "UdpLite", "Udp"}
for i, nss in ipairs(netsubstat) do
local substat_s = string.match(netstat, nss .. ": ([A-Z][A-Za-z0-9 ]+)")
if substat_s then
local substat = space_split(substat_s)
local substatv = space_split(string.match(netstat, nss .. ": ([0-9 -]+)"))
for ii, ss in ipairs(substat) do
metric("node_netstat_" .. nss .. "_" .. ss, "gauge", nil, substatv[ii])
end
end
end
end
function scraper_network_devices()
local netdevstat = line_split(get_contents("/proc/net/dev"))
local netdevsubstat = {"receive_bytes", "receive_packets", "receive_errs",
"receive_drop", "receive_fifo", "receive_frame", "receive_compressed",
"receive_multicast", "transmit_bytes", "transmit_packets",
"transmit_errs", "transmit_drop", "transmit_fifo", "transmit_colls",
"transmit_carrier", "transmit_compressed"}
for i, line in ipairs(netdevstat) do
netdevstat[i] = string.match(netdevstat[i], "%S.*")
end
local nds_table = {}
local devs = {}
for i, nds in ipairs(netdevstat) do
local dev, stat_s = string.match(netdevstat[i], "([^:]+): (.*)")
if dev then
nds_table[dev] = space_split(stat_s)
table.insert(devs, dev)
end
end
for i, ndss in ipairs(netdevsubstat) do
netdev_metric = metric("node_network_" .. ndss, "gauge")
for ii, d in ipairs(devs) do
netdev_metric({device=d}, nds_table[d][i])
end
end
end
function scraper_time()
-- current time
metric("node_time", "counter", nil, os.time())
end
uname_labels = {
domainname = "",
nodename = "",
release = string.sub(get_contents("/proc/sys/kernel/osrelease"), 1, -2),
sysname = string.sub(get_contents("/proc/sys/kernel/ostype"), 1, -2),
version = string.sub(get_contents("/proc/sys/kernel/version"), 1, -2),
machine = string.sub(io.popen("uname -m"):read("*a"), 1, -2)
}
function scraper_uname()
uname_labels["domainname"] = string.sub(get_contents("/proc/sys/kernel/domainname"), 1, -2)
uname_labels["nodename"] = string.sub(get_contents("/proc/sys/kernel/hostname"), 1, -2)
metric("node_uname_info", "gauge", uname_labels, 1)
end
function scraper_nat()
-- documetation about nf_conntrack:
-- https://www.frozentux.net/iptables-tutorial/chunkyhtml/x1309.html
local natstat = line_split(get_contents("/proc/net/nf_conntrack"))
nat_metric = metric("node_nat_traffic", "gauge" )
for i, e in ipairs(natstat) do
-- output(string.format("%s\n",e ))
local fields = space_split(e)
local src, dest, bytes;
bytes = 0;
for ii, field in ipairs(fields) do
if src == nil and string.match(field, '^src') then
src = string.match(field,"src=([^ ]+)");
elseif dest == nil and string.match(field, '^dst') then
dest = string.match(field,"dst=([^ ]+)");
elseif string.match(field, '^bytes') then
local b = string.match(field, "bytes=([^ ]+)");
bytes = bytes + b;
-- output(string.format("\t%d %s",ii,field ));
end
end
-- local src, dest, bytes = string.match(natstat[i], "src=([^ ]+) dst=([^ ]+) .- bytes=([^ ]+)");
-- local src, dest, bytes = string.match(natstat[i], "src=([^ ]+) dst=([^ ]+) sport=[^ ]+ dport=[^ ]+ packets=[^ ]+ bytes=([^ ]+)")
local labels = { src = src, dest = dest }
-- output(string.format("src=|%s| dest=|%s| bytes=|%s|", src, dest, bytes ))
nat_metric(labels, bytes )
end
end
function timed_scrape(scraper)
local start_time = socket.gettime()
-- build the function name and call it from global variable table
_G["scraper_"..scraper]()
local duration = socket.gettime() - start_time
return duration
end
function run_all_scrapers()
times = {}
for i,scraper in ipairs(scrapers) do
runtime = timed_scrape(scraper)
times[scraper] = runtime
scrape_time_sums[scraper] = scrape_time_sums[scraper] + runtime
scrape_counts[scraper] = scrape_counts[scraper] + 1
end
local name = "node_exporter_scrape_duration_seconds"
local duration_metric = metric(name, "summary")
for i,scraper in ipairs(scrapers) do
local labels = {collector=scraper, result="success"}
duration_metric(labels, times[scraper])
print_metric(name.."_sum", labels, scrape_time_sums[scraper])
print_metric(name.."_count", labels, scrape_counts[scraper])
end
end
-- Web server-specific functions
function http_ok_header()
output("HTTP/1.0 200 OK\r\nServer: lua-metrics\r\nContent-Type: text/plain; version=0.0.4\r\n\r")
end
function http_not_found()
output("HTTP/1.0 404 Not Found\r\nServer: lua-metrics\r\nContent-Type: text/plain\r\n\r\nERROR: File Not Found.")
end
function serve(request)
if not string.match(request, "GET /metrics.*") then
http_not_found()
else
http_ok_header()
run_all_scrapers()
end
client:close()
return true
end
-- Main program
for k,v in ipairs(arg) do
if (v == "-p") or (v == "--port") then
port = arg[k+1]
end
if (v == "-b") or (v == "--bind") then
bind = arg[k+1]
end
end
scrape_counts = {}
scrape_time_sums = {}
for i,scraper in ipairs(scrapers) do
scrape_counts[scraper] = 0
scrape_time_sums[scraper] = 0
end
if port then
server = assert(socket.bind(bind, port))
while 1 do
client = server:accept()
client:settimeout(60)
local request, err = client:receive()
if not err then
output = function (str) client:send(str.."\n") end
if not serve(request) then
break
end
end
end
else
output = print
run_all_scrapers()
end