Any plans for a supported .Net Membase client?

Slegay,
I'm working to pull together an official support statement for the .Net platform for you. I didn't want to leave you hanging, but I will hopefully be able to get back to you tomorrow with more information.

Thanks!

Perry

While we're working on that, I think we can dig further into your issue. If you have the ability, can you do your "multi-get" test with only one node in the cluster? That will simplify things and then we can build up from there.

-Perry

Yep, I tested that already. With one node, works fine. Also, if I change the key transformer back to the default one (meaning no transformation), multi-get works fine. My guess is it has something to do with the splitting of the keys for consistent hashing. The strange thing is single key retrieval seems to work OK with TigerHashing enabled.

Regarding the dead pools, I commented out a line of code in the client. In PoolSocket.cs, the Read method was killing the socket (and therefore the pool itself) whenever an IOException was thrown (line 193). That seemed a bit overkill to me since that exception seems to be thrown very often by BasicNetworkStream.Read() on all the different configurations I've tested (usually because of an aborted connection). I trust the NorthScale cluster to identify and clean up dead pools, so that client behavior seemed unnecessary. It seemed to help. I'm not getting as many misses, and traffic to the DB server has been reduced dramatically.

Still, it'd be nice to get a complete supported offering from you guys (server + smart .Net client).

Thanks for your help.

Slegay,
- The first issue of multi-get hashing is being worked on as we speak.
- The second issue of the dead pools should be fixed in the latest code already. Can you download the latest source and compile it? [url]http://github.com/enyim/EnyimMemcached/archives/binary[/url]. In reality, the 'ioexception' is being thrown because the webclient's readwritetimeout was not set properly. The client aborts processing after a while when no data is received and the connection is still open which is exactly is what's happening when the configuration does not change.

Also, can you please run the unit tests and see if they fail with your config? [url]http://github.com/enyim/EnyimMemcached/issues#issue/9[/url]

Thanks!

Perry

Slegay, I'm happy to announce that we are now officially supporting the Enyim .NET client for memcached.

Did upgrading the client make a difference for you?

Thanks!

Perry

Hi there, sorry for the late response.

I can't give you an update on the TigerHashing problem. I pretty much gave up on that one since we could easily take care of hashing in our client wrapper and use the default transformer. That works well. I know the author's waiting for me to give exact steps to reproduce the problem, but it's a bit tricky at the moment since it depends on the actual keys and server config. I'll try to put something together, but since it's no longer an issue for us, it's pretty low on our priority list.

regarding the IOException, we've been running our tweaked version of the code for a month. I just changed the client to the patched version, we'll deploy it in our next release. I still need to run some tests first, I'll keep you posted.

Today though, a new problem came up. We're using NorthScale / Memcached as a centralized counter provider. We hadn't used the Increment / Decrement operations until now. We wrote some unit tests, and we're getting some weird issues when doing a Get on a Key whose value was incremented. Instead of returning the counter's value, we're getting null. The key is there though - doing an Add on that key returns false (as expected), and Increment / Decrement operations return expected results. But a simple Get seems to always be null.

What's the best way to send unit tests to you guys?