Does anyone have any experience using a second-level cache in a cluster?

In the docs section about the second-level cache, under read/write strategy, it says:



Can anyone shed more light on this? Specifically:

1. Why is locking required in a cluster but not in a single process?
2. Why don't the built-in cache providers support locking? Is it particularly difficult to implement? Are there significant performance penalties?

A possible means of implementing locking is suggested here (in reference to memcached): http://www.socialtext.net/memcached/index.cgi?faq#managing_a_memcached_cluster

(see section "Emulating locking with the add command")

The answer to question 1 is because only one mutable instance of the object exists on a single server. Where multiple mutable instances exist in a cluster. Thus you have to keep the objects synchronized.

I don't know the answer to question 2. I'd imagine locking objects across servers is more difficult and more resource intensive than using a solution like memcached. Memcached only stores one copy of the object across the farm so no locking is needed. Think of memcached is being a single big out of process hashtable spread across servers.

If you want to use nhibernate across a server farm or cluster you have these options.
1. Turn 2nd level caching off
2. Use a third party cache provider that supports farms/clusters
3. Use memcached with the built in NHibernate memcached provider.

Thanks for responding peterlemonjello. I see what you mean about the need to keep objects in sync if multiple instances exist in the cluster. However, are you sure that's the whole story? I spent some time sifting through the NH source in the Cache namespace, and although I can't make sense of all the details, it seems that the lock/unlock methods are used by the concurrency strategy to ensure proper transaction isolation. That's why it doesn't make sense to me that these methods are not implemented even for an in-process cache (eg SysCache). However, no one else seems bothered by this... I would just like to understand why...

Ultimately I would like to use memcached, but I need a better understanding of how its use will or will not affect transaction isolation. Do you know if there any issues to be aware of in this regard?

Thanks.

I'm not sure that's the whole story. I guess I kind of took a leap of faith after spending a lot of time reading about the 2nd level cache on the forums. I haven't spent much time looking at the code. I've just assumed it'll do what it's designed to do until a problem arises.

FWIW, we used to use a server farm with Syscache2 and sqlcache dependancies to deal with cache synchronization without any problems. We've since transitioned to memcached. So far we haven't had any transaction isolation problems. In fact it was a pretty simple migration to memcached.

Good to know - Thanks!