Telstra honesty report - monitor connection drops

telstra-monitor.plist

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>Label</key>
    <string>telstra-monitor</string>

    <key>KeepAlive</key>
    <true/>

    <key>EnvironmentVariables</key>
    <dict>
        <key>THRESHOLD</key>
        <string>20000</string>

        <key>SUMMARY_LOG</key>
        <string>/var/log/telstra-monitor-summary.log</string>
    </dict>

    <key>ProgramArguments</key>
    <array>
        <string>/Users/benkitzelman/.nvm/versions/node/v8.6.0/bin/node</string>
        <string>/Users/benkitzelman/dev/projects/telstra-monitor/index.js</string>
    </array>

    <key>StandardErrorPath</key>
    <string>/var/log/telstra-monitor.err</string>

    <key>StandardOutPath</key>
    <string>/var/log/telstra-monitor.log</string>
</dict>
</plist>

telstra-summary.sh

# will signal the telstra-monitor daemon to write the summary of outages so far to the configured file
# run this as sudo if querying as a daemon
pgrep -lf telstra-monitor | awk '{print $1}' | sudo xargs kill -s HUP && cat /var/log/telstra-monitor-summary.log

telstra.js

var spawn = require('child_process').spawn;

const ADDR                    = process.env.PING_HOST || "8.8.8.8"
const THRESHOLD_FOR_REPORT_MS = process.env.THRESHOLD || 2000
const SUMMARY_LOG_FILE        = process.env.SUMMARY_LOG || `${__dirname}/telstra-summary.log`

var serviceStart    = new Date
var pingCount       = 0
var failedPingCount = 0
var downtimeCount   = 0
var totalDowntimeMs = 0
var longestOutage   = { value: 0, start: '', end: '' }
var start;

var seconds = (ms) => {
  return `${ ms / 1000 }s`
}

var mins = (ms) => {
  return `${ ms / 1000 / 60 }m`
}

var hours = (ms) => {
  return `${ ms / 1000 / 60 / 60 }h`
}

var durations = (ms) => {
  return `${ seconds(ms) } (${ mins(ms) } or ${ hours(ms) })`
}

var log = (msg) => {
  console.log.apply( console, [ (new Date).toString(), msg ] )
}

var logSummary = () => {
  var pingFailureRate = failedPingCount / pingCount * 100
  var averageDowntime = ( totalDowntimeMs / downtimeCount ) || 0

  return `
    FROM ${ serviceStart.toString() } TO ${ (new Date).toString() }:
    ===========
    Total Dropouts: ${ downtimeCount }
    Ping failure rate: ${ pingFailureRate }% (${ failedPingCount } of ${ pingCount } pings failed)
    avg downtime: ${ durations( averageDowntime ) }
    Total downtime: ${ durations( totalDowntimeMs ) }
    Longest outage:
      downtime: ${ durations( longestOutage.value ) }
      started: ${ longestOutage.start }
      ended: ${ longestOutage.end }
    ===========
  `
}

var writeLogSummary = () => {
  require('fs').writeFileSync(SUMMARY_LOG_FILE, logSummary())
}

log(`Started Connectivity Monitor - logging drops longer than ${ seconds( THRESHOLD_FOR_REPORT_MS ) }\n`)
var child = spawn("ping", [ ADDR ])


child.stdout.on('data', (data) => {
  var now = new Date

  pingCount++
  var str = data.toString()

  if( new RegExp(`.*from ${ ADDR.replace(/\./, '\.') }.+time=`).test( str ) ) {
    var downtime = now - (start || now)
    if(downtime >= THRESHOLD_FOR_REPORT_MS ) {

      if(downtime > longestOutage.value) {
        longestOutage.value = downtime
        longestOutage.start = start
        longestOutage.end   = now
      }

      totalDowntimeMs += downtime
      log(`Reconnected. Outage lasted ${ seconds( downtime ) }\n`)
      start = null
    }

    return
  }

  failedPingCount++

  if(start) return

  start = new Date();
  downtimeCount++
  log(`DISCONNECTION STARTED: ${ str.replace(/\n/g, '') }`)
});

child.stderr.on('data', (data) => { log(`Error: ${data}`) });

child.on('close', (code) => { log(`Closing ${code}`) });

[ "SIGINT", "SIGTERM" ].forEach( (signal) => {
  process.on(signal, () => {
    log( logSummary() )
    process.exit()
  })
})

process.on('SIGHUP', writeLogSummary)