Exponential backoff and dial retrying

[kjnsn]

When the reader/writer returned an io.EOF this was passed back to the client instead of marking the TCP connection as bad and attempting to redial. This implements an exponential backoff redial system.

[jonaz]

I just ran into this. I was running a Get and suddenly i got EOF instead of memcache.ErrCacheMiss

[googlebot]

Thanks for your pull request. It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

📝 Please visit https://cla.developers.google.com/ to sign.

Once you've signed, please reply here (e.g. I signed it!) and we'll verify. Thanks.

---

• If you've already signed a CLA, it's possible we don't have your GitHub username or you're using a different email address. Check your existing CLA data and verify that your email is set on your git commits.
• If you signed the CLA as a corporation, please let us know the company's name.

[jonaz]

@kjnsn @bradfitz any chance to get this in? Otherwise i need to fork. We have a haproxy between app and memcached that dont let connections live for ever :(

[tcolgate]

@bradfitz we're hitting this as well. I think I can work around it by disable idle connection timeout on the server, but EOF should probably be marking the connections dead. I'm happy to bring this PR, or something similar, up to date.

[tcolgate]

okay, so io.EOF isn't a resumableError, so the connection shouldn't get re-used, which is why my service isn't degenerating to a pile of cache errors over time. It should be safe to retry from the caller without any danger of clogging up the connection pool with closed connections.

[frikky]

Hey! Is there any way to get or something similar merged in?

I've had some request errors leading to eof, and would love at least some kind of retry system

[QuChen88]

Ping. Any plans on getting this PR merged?