ArghWebWorks

Every morning, I had a client calling me and asking “Is our server abnormally unresponsive?” They were browsing their website and it seemed to them that the pages were taking longer than usual to render. Once I figured out why they were asking and what they were measuring with, I realized I could graph this amount of time, the important thing was figuring out how to get this information.

Here’s what I came up with: I was already running mrtg-rrd to do much of their graphing, so I wanted to create a MRTG Target line to correspond to “how long does our page take to render.” I wrote a script to run “/path/towget –delete-after -pq (page) -O /tmp/garbage” and display the amount of time it took. I wrote it in php because it was easiest.


$wget="/usr/bin/wget -pq --delete-after -O /tmp/garbage";
$toget = "$site";
$start=microtime(true);
exec( "$wget $toget", $out, $e );
$end=microtime(true);
$dur=round( $end-$start, 3 );
print "$dur\n";


The problem with this is that I was running it on their gateway; their gateway is also the monitor and grapher. It was taking less than a second to render the page ( because of local network speeds ). Instead, I wanted it to more closely correspond to what the client would be seeing. Their development server is at their office; so I put the script there and wrote the MRTG target to point at a local ( on the gateway ) script that got the value from the development server, printed it twice, printed the date and the name of the site. It looks like this:

$user='foo';
$password='bar';
$sitename='example.com';
exec("/usr/bin/lynx -auth=$user:$password --dump http://developmentserver.com/scriptname.php ", $out, $e );
$time=0;
while ( ($time==0) && (count( $out ) > 0 ) ) {
$time=(float) array_shift( $out );
}
print "$time\n";
print "$time\n";
print date("r\n");
print "$sitename\n";


A couple of important caveats. The day after I started running this ( it averages around 4.6 seconds, btw, much too long imho ), the displayed time doubled. It could have been a change to the page we’re looking for, but it turned out to be a DNS issue on the development server. So DNS is important. This doesn’t take into account the amount of time a browser takes to render a page; it’s just the time to download all the pieces. So processor intensive stuff, like super heavy javascript or super heavy tables, won’t show up here. However, if you keep in mind what the value is- how long to resolve and download all the pieces of a web page, this is a useful metric.