TraceSource.TraceEvent() fails logging when Except

问题:

I have a call to TraceSource.TraceEvent() that is sometimes not writing to the Azure Diagnostics logs.

public class WorkerRole : RoleEntryPoint
{
    private TraceSource trace = new TraceSource(
        "ImportService", SourceLevels.Information);

    public override void Run()
    {
        ...
        try
        {
            ...
        }
        catch (Exception ex)
        {
            bool hasMsg = !string.IsNullOrEmpty(ex.Message);
            trace.TraceEvent(TraceEventType.Error, 0,
                "ex has message: " + hasMsg.ToString());   // this gets logged
            trace.TraceEvent(TraceEventType.Error, 0,
                "Inner exception message: " + ex.Message); // this does not
        }
    }
}

In certain cases, and I can't tell which since I can't read the Exception message, the second call is not found in the WADLogsTable. Are there certain characters that are not allowed, either by TraceSource or by DiagnosticMonitor?

To further narrow this down, the Exception in question is actually the InnerException of Exception: "There is an error in XML document (72, -499)". The XML that causes the Exception contains invalid character entities, eg . Could it be that the Exception message contains some of these character entities and the TraceSource fails to log them?

Edit: I was able to finally repro this in my dev environment and so I was able to examine the Exception in the debugger. The exception that won't log is an XmlException:

'', hexadecimal value 0x11, is an invalid character. Line 72, position -499.

In between the quotes is the non-printable character - it shows up as a black triangle in the debugger. So, this leads me to believe that my suspicion is correct - Some piece of the logging mechanism doesn't like the non-printable character. So, which piece? Or, more importantly, since it looks like I need to start sanitizing all of my strings when tracing, which characters should I look for to remove?

Is there some built in function that will sanitize a string, removing non-printable characters?

回答1:

Interesting. It looks like you'll need to HTML encode the exception string. This will turn quotes into e.g. " and your ASCII non-printing character into , or similar.

So:

    trace.TraceEvent(TraceEventType.Error, 0,
        "ex has message: " + HttpUtility.HtmlEncode(hasMsg.ToString()));   
    trace.TraceEvent(TraceEventType.Error, 0,
        "Inner exception message: " + HttpUtility.HtmlEncode(ex.Message)); 

should work just fine.

Frustratingly, HttpUtility is in System.Web, you'll need to add a reference to System.Web.dll to get this to go.

回答2:

The answer to another question helped me to figure out a solution. I added a couple of extension methods for convenience:

public static string RemoveControlChars(this string s)
{
    return Regex.Replace(s, @"(?![\r\n])\p{Cc}", "");
}
public static void TraceEvent(this TraceSource trace, 
    TraceEventType eventType, MyEvtEnum eventId, string message)
{
    trace.TraceEvent(eventType, (int)eventId, message.RemoveControlChars());
}

I like the added benefit of not having to cast MyEvtEnum to int every time I call TraceEvent and it adds a natural overload, so this feels like a double win.

It bothers me that I have to do this at all. One of the primary uses of a diagnostics system is to log Exceptions. Such a diagnostics system should be able to handle any string that an Exception message might contain. I also lose line breaks, which is frustrating. Edit: Losing line breaks was a side effect of RemoveControlChars(). I didn't realize that \r and \n are included as "control characters". I've updated my regular expression to not replace \r and \n characters.

I don't like accepting my own answer, so if you have an alternate solution or an improvement to mine, please post it and if it's better, I'll accept it.