Feature Suggestion: Timeout between connection attempts about pymemcache HOT 5 CLOSED

@cgordon I had forgotten about this, and apparently missed your last comment anyway. We added https://gist.github.com/dcrosta/9473409 to our codebase to implement this for our needs -- documenting here in case it's helpful for anyone else in the future.

from pymemcache.

The Client class was designed to be as simple as possible, and to avoid any connection policy issues, like pooling, load balancing, consistent hashing and even retry logic. My goal was to make it as easy as possible to build those things around Client. The FallbackClient class is a good example of that. The underlying principle was to keep Client as clean and performant as possible, so users of Client didn't have to pay for any functionality they didn't want. That was largely motivated by python-memcaches decision to layer in load balancing and thread locals and other things that aren't completely necessary in the core client logic.

This request falls into sort of a gray zone for me. Can you give me more background on why you want this functionality? Are you worried about overloading a memcache server with repeated connections? Do you just not want to wait for the connect_timeout during periods when you are restarting your memcache hosts?

from pymemcache.

I see, that makes sense.

We actually connect to something that speaks the memcache protocol but which is not memcache itself. We do this from very high-volume services where each call matters relatively little, so python-memcache's policy of "I'll just lie to you for 30 seconds after a disconnect" is actually OK for us.

Maybe I'll experiment with a subclass of Client internally, and if it looks like something others might want I can submit here as another module, similar to the fallback client?

from pymemcache.

I agree with your earlier statement that this isn't a big enough feature to warrant a subclass, I just want to make sure I understand the need so that we add the right thing to Client.

Client already has ignore_exc, which will do what you want for "get" style requests (ignore exceptions and report no return result). So presumably what you're looking for is something similar for "set" style requests? What is happening in your architecture during the period in which memcache is disconnected? Are you restarting the proxy on the local hosts? Or restarting the remote memcache server connected to the proxy?

from pymemcache.

@drcosta I'm going to "time out" this issue :) Let me know if you'd like to revisit it.

from pymemcache.