How to architect a web service cache in c#

Question

I'm working on a web service to read data from a 3rd party feed, modify it a little and store it, then return it to my clients. It only needs to update from the 3rd party site periodically. It will be running as a WCF service in a web role in Azure.

At first I thought I would just always make a call to my parsefeed method, but make that call return if the last update was too soon...

   public void ParseFeed()
    {
        if (DateTime.Now > lastrun.AddMinutes(1))
        {
         //Fetch updated data into something shared here.
         //thedata is a public static in Global class 
         thedata = fetchdata();
         lastrun=DateTime.Now;            
        }
    }

But I guess that as the fetch can take 1-2s (its a web service) that multiple users will hit that code at once.

from http://support.microsoft.com/default.aspx?scid=kb;en-us;Q312607 Because static members of any class, including an application class, are not thread-safe, the user code must provide appropriate locking for access to static members. This applies to any static member that you add to the application class.

• I could use locking (not sure how) EDIT: lots of info here

• I could avoid the static var and use a cache, and put my data in it (but that would get removed at expiration and multiple users would try to fetch data)

• I could use a cache with a fake item of data in it (basically as a timer) and refresh when that expired - but that would refresh even if nobody was hitting the site. (May also not be thread safe)

• I can't really use an output cache because the clients query the data I return in a manner that probably makes each request unique: my service sorts and filters according to the request

BTW I am not worried about consistency of results over multiple instances on Azure. Each can fetch their own so I don't need to share state over multiple servers.

I get the feeling there is a simple solution, that I have totally missed. Ideas ?

Answer 1

From your question, it looks like:

• It is not acceptable to serve data that's more than a minute out of date
• It is acceptable if two service instances return different data due to refresh times being out of sync
• It is acceptable for callers to block for 1-2 seconds while data is refreshed

Assuming these are all true, then the simplest solution is just to use a static variable to store the data, with a lock construct around the whole check/refresh block. I wouldn't even bother trying to do anything clever like the double-check lock pattern; lock contention simply won't be an issue as the time spent in the critical region will pale into insignificance compared to the overhead of operating a web service, except when it is blocking and everybody has to block anyway.

Answer 2

Since you probably will be reading from the cache more than you will be writing, I would use a ReaderWriterLockSlim to ensure data can be read by multiple threads simultaneously but not written. I would also make sure that the cached data is immutable to make sure it's not altered by consumers.