Archive for the ‘Software’ Category

A coworker and I came across this block of code which was causing us numerous headaches over the past few weeks. For a while, there was absolutely no time delay which was causing an inordinate amount of CPU cycles but thats besides the point. After that was fixed with a short delay we were left with what you see below. Can you tell what’s wrong?

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

   while ( true ) {
        TriggeredSend ts = queue.peek();
        if ( ts != null ) {
            switch( doPost( ts ) ) {
                case SUCCESS:
                    queue.take();
                    ts.complete( true );
                    break;
                case FAILURE:
                    queue.take();
                    ts.complete( false );
                    break;
                case SYSTEM_UNAVAILABLE:
                    systemIsAvailable = false;
                    break;
            }
        }
        Thread.sleep( delay );
   }

As with most concurrency bugs, its hard to determine what’s actually happening without experience. You can turn on your debugger but then you are changing the way the code behaves by slowing it down, essentially taking out any race conditions. Turns out, the same task was getting invoked multiple times well as some other strange conditions. That led us to the block of code you see above.

Figure it out yet? Well, what’s happening is that when the code is peeking into the queue, its assigning the result to a variable which is then being called. According to the java.util.concurrency spec, the peek method is used to see if the queue has tasks in it and not used to take any tasks off the queue. Not only that, but after the task was invoked, another task was being pulled off the queue. So what happens when multiple threads come through that block of code and peek? Well, they can both get the same task and invoke it. Bad bad bad.

For the past day or two I had been struggling at work to figure out why Internet Explore 7 would not pay attention to the response headers stating not to cache the response. First, I tried setting the date header to expire instantly, with a value of -1. QA confirmed that this solved the problem in some revisions of IE 7 but not all. After digging around the web it turns out that you have to set a few more headers, one of which I had never even heard of. Here’s what solved the problem for me. This snippet is in Java but could apply to any language:

response.setDateHeader( "Expires", -1 );
response.setHeader( "Cache-Control", "private" );
response.setHeader( "Last-Modified", new Date().toString() );
response.setHeader( "Pragma", "no-cache" );
response.setHeader( "Cache-Control", "no-store, 
                                 no-cache, must-revalidate" );

Today I came across an issue at work regarding the made up MIME type of ‘image/pjpeg’ while uploading images from IE 7. Java’s ImageIO was choking while trying to upload this type of file because it is not in the list of supported MIME types. The supported MIME type list can be viewed by invoking ImageIO.getReaderMIMETypes(). After reading many blogs of others pounding their head, making fun of Microsoft, and using other solutions like ImageMagick, I found a solution. Add this to your list of supported types and move on. This format is made up and will not cause any problems. Pass the input stream onto your File IO and be done with it. Microsoft owes me half an hour of my life back.

Up until recently I never had to write any code that dealt with reflection. Before now I worked for CNET which is basically a series of differnent CMS’ with one request, one response. My first week at the new job required me to build a JMS system that would accept method invocations across the wire seamlessly for the producer. Java.lang.reflect was the answer.

Once you can wrap your head around the idea its pretty simple to write. Essentially what we are attempting to do is shield the caller from having to know anything about the other method that is going to invoked behind the scenes.

public Class MyConcreteProxyClass implements InvocationHandler {

   Object obj;
   public MyDynamicProxyClass(Object obj)
     { this.obj = obj; }

  public Object invoke(Object proxy, Method m, Object[] args)
         throws Throwable {
     try {
        // do stuff
     } catch (Exception e) {
        throw e;
     }
        // return something
     }
}

Now we have to create the proxy interface which the caller will implement. Any of the methods of this class that he implements and fire will invoke method on the MyConcreteProxyClass object.

public interface ProxyInterface{
    public Object myMethod();
}

Now its time to create wire everything up. There are several ways to do this which help hide this proxying like a static factory method but this simply demonstrates how this works.

ProxyInterface proxy = (ProxyInterface)
Proxy.newProxyInstance(obj.getClass().getClassLoader(),
         Class[] { ProxyInterface.class },
          new MyConcreteProxyClass(obj));


This is now one of my favorite books on Java which I am probably going to read again just to be sure I have soaked up as much information as I can. This is a practical book written by a practical person who understands his audience, engineers. He is straight to the point with his code examples and doesnt bore you to death explaining every little detail or referencing lines of code and functions 8 pages back. I wish more technical writers would be this concise and straight to the point. There is something to be said for any technical writer who can get their message across with fewer words. No where else can I think of a better example of the old adage “less is more”. I would highly recommend this book to any engineer, especially associates.

Unfortunately for us we dont have a full time operations team at the moment. Its me really. The engineer, the frontend web developer, and admin. Im lucky that I saw it, put two-and-two together, and stopped it.

Anywho, this script kiddie managed to actually take down our appservers because our servlet container was only set to use 512 MB of RAM! This was not my doing although it has been operating fine for almost 2 years. Anyway, when we saw the increased spike in traffic we said great, maybe Googebot has come back around to save us from our perpetual Google dance. I noticed the increase in bot activity but thought nothing of it and went home. Little did I know it wasnt Googlebot.

The next day, i.e. today, I got into work to find our leads were still down but our page views were up. Thats odd? How and why could this be happening? I knew this wasnt people traffic because Google’s Urchin Tracker wasnt registering the requests because its Javascript (bots done fire it). It had to be a malicious user. I dug through yesterday’s logs and it turned out to be some script kiddie in Australia making about 25 requests per second from one IP address. What? No DDoS? Anyway, our machines performed well after more memory was allocated but they were thrashing a little.

Page load time

CPU load