implementing out-of-process cache using Redis in windows azure

Question

I've been working on a webpage that displays a table from a database I have in my azure cloud. In order to reduce calls to the DB directly for performance improvement I would like to build a cache for the page. Currently, I hold an in-memory cache (in-process) for the reads of the table. Now I would like to make an out-of-process cache, that should be updated from when writes are made, meaning inserts or updates (because after a value is updated or added, the in-memory cache will be no longer valid).

I was recommended on Redis, and specifically Book Sleeve, my question is where I can find some code samples to help me figure out how to start build the out-of-process cache with it and combine it in my current project.

Thanks in advance

Answer 1

If you want purely out-of-process, then it is pretty simple - something like the following, but noting that a BookSleeve is designed to be shared: it is fully thread-safe and works as a multiplexer - you shouldn't create / dispose them for every call. Note also that in this context I'm assuming you will handle serialization separately, so I'm simply exposing a byte[] API:

class MyCache : IDisposable
{
    public void Dispose()
    {
        var tmp = conn;
        conn = null;
        if (tmp != null)
        {
            tmp.Close(true);
            tmp.Dispose();
        }
    }
    private RedisConnection conn;
    private readonly int db;
    public MyCache(string configuration = "127.0.0.1:6379", int db = 0)
    {
        conn = ConnectionUtils.Connect(configuration);
        this.db = db;
        if (conn == null) throw new ArgumentException("It was not possible to connect to redis", "configuration");
    }
    public byte[] Get(string key)
    {
        return conn.Wait(conn.Strings.Get(db, key));
    }
    public void Set(string key, byte[] value, int timeoutSeconds = 60)
    {
        conn.Strings.Set(db, key, value, timeoutSeconds);
    }
}

What gets interesting is if you want a 2-tier cache - i.e. using local memory and the out-of-process cache, as now you need cache invalidation. Pub/sub makes that handy - the following shows this. It might not be obvious, but this would be doing a lot fewer calls to redis (you can use monitor to see this) - since most requests are handled out of the local cache.

using BookSleeve;
using System;
using System.Runtime.Caching;
using System.Text;
using System.Threading;

class MyCache : IDisposable
{
    public void Dispose()
    {
        var tmp0 = conn;
        conn = null;
        if (tmp0 != null)
        {
            tmp0.Close(true);
            tmp0.Dispose();
        }

        var tmp1 = localCache;
        localCache = null;
        if (tmp1 != null)
            tmp1.Dispose();

        var tmp2 = sub;
        sub = null;
        if (tmp2 != null)
        {
            tmp2.Close(true);
            tmp2.Dispose();
        }

    }
    private RedisSubscriberConnection sub;
    private RedisConnection conn;
    private readonly int db;
    private MemoryCache localCache;
    private readonly string cacheInvalidationChannel;
    public MyCache(string configuration = "127.0.0.1:6379", int db = 0)
    {
        conn = ConnectionUtils.Connect(configuration);
        this.db = db;
        localCache = new MemoryCache("local:" + db.ToString());
        if (conn == null) throw new ArgumentException("It was not possible to connect to redis", "configuration");
        sub = conn.GetOpenSubscriberChannel();
        cacheInvalidationChannel = db.ToString() + ":inval"; // note that pub/sub is server-wide; use
                                                             // a channel per DB here
        sub.Subscribe(cacheInvalidationChannel, Invalidate);   
    }

    private void Invalidate(string channel, byte[] payload)
    {
        string key = Encoding.UTF8.GetString(payload);
        var tmp = localCache;
        if (tmp != null) tmp.Remove(key);
    }
    private static readonly object nix = new object();
    public byte[] Get(string key)
    {
        // try local, noting the "nix" sentinel value
        object found = localCache[key];
        if (found != null)
        {
            return found == nix ? null : (byte[])found;
        }

        // fetch and store locally
        byte[] blob = conn.Wait(conn.Strings.Get(db, key));
        localCache[key] = blob ?? nix;
        return blob;
    }

    public void Set(string key, byte[] value, int timeoutSeconds = 60, bool broadcastInvalidation = true)
    {
        localCache[key] = value;
        conn.Strings.Set(db, key, value, timeoutSeconds);
        if (broadcastInvalidation)
            conn.Publish(cacheInvalidationChannel, key);
    }
}

static class Program
{
    static void ShowResult(MyCache cache0, MyCache cache1, string key, string caption)
    {
        Console.WriteLine(caption);
        byte[] blob0 = cache0.Get(key), blob1 = cache1.Get(key);
        Console.WriteLine("{0} vs {1}",
            blob0 == null ? "(null)" : Encoding.UTF8.GetString(blob0),
            blob1 == null ? "(null)" : Encoding.UTF8.GetString(blob1)
            );
    }
    public static void Main()
    {
        MyCache cache0 = new MyCache(), cache1 = new MyCache();
        string someRandomKey = "key" + new Random().Next().ToString();
        ShowResult(cache0, cache1, someRandomKey, "Initially");
        cache0.Set(someRandomKey, Encoding.UTF8.GetBytes("hello"));
        Thread.Sleep(10); // the pub/sub is fast, but not *instant*
        ShowResult(cache0, cache1, someRandomKey, "Write to 0");
        cache1.Set(someRandomKey, Encoding.UTF8.GetBytes("world"));
        Thread.Sleep(10); // the pub/sub is fast, but not *instant*
        ShowResult(cache0, cache1, someRandomKey, "Write to 1");
    }
}

Note that in a full implementation you probably want to handle occasional broken connections, with a slightly delayed reconnect, etc.