Pass retry attempt count through middlewares

[underwoo16]

This would allow middlewares to instrument calls and retries with more granularity, as well as determine if raised errors are final or if they will be retried.

• Discussed in Broken Pipe and EOFErrors #119 where users saw an increase in errors reported to DataDog that actually resulted in successful calls due to the default of 1 reconnect_attempts in redis-rb / sidekiq

• This was specifically adapted from the idea suggested in Broken Pipe and EOFErrors #119 (comment)

There would still need to be changes coordinated in DataDog/dd-trace-rb#3820 to close the initial issue (#119) but this is a generic approach that would also benefit anyone writing their own middlewares

[byroot]

So I'm favorable to this change, the issue however is that it's not really backward compatible.

So I need to think about it.

[byroot]

Random thoughts:

• retry_attempt isn't that useful by itself, you got to know how many attempts total are allowed etc. It's not the most ergonomic.
• A way to provide the info that wouldn't break the interface would be to have the info of whether the error will be retried in the exception itself.

[underwoo16]

A way to provide the info that wouldn't break the interface would be to have the info of whether the error will be retried in the exception itself.

That would be a clean interface, I will take a look at this approach. It may be difficult since I think middleware sees exceptions before ensure_connected which handles the retries.

---

Another idea: How opposed are we to providing retry info via the config object which is already available in middleware calls? It handles a bit of retry logic https://github.com/redis-rb/redis-client/blob/master/lib/redis_client/config.rb#L143 - but I could also understand not wanting to add transient state to the config

[byroot]

I opened #255 as an alternative solution, I'd love to hear your feedback about it.

[underwoo16]

Tracking retry_attempt on the connection is clean and I like the error.final? interface 👍🏼 Thanks for taking a look, I'll close this in favor of #255, left a couple questions over tehre