async_run exits after a resolve, connect or TLS handshake failure

[anarthal]

After the first failure, async_exec returns, even if it's configured to retry indefinitely.

[mzimbres]

There are some important things to keep in mind when working on this issue. Boost.Redis implements only the simplest reconnection strategy where it waits for a fixed amount of time and then retries.

The next natural step would be to support exponential backoff as the nodejs client does.

More critical than exponential backoff is reconnecting over a sentinel, something two users already expressed interest. The complex thing about this however is that requests that set cancel_on_connection_lost will remain suspended during failover and then will have to be executed in the new instance.

I thought at first that I could offer built-in sentinel support, but I am not sure anymore because a proper use of sentinel might involve subscribing to health events, that might bloat the current connection (see the Pub/Sub messages section).

A better strategy might be to improve my original idea of having a separate function to resolve address, see here.

[anarthal]

I'd like to solve this without a functional change before the next Boost release cutoff.

Still, I've read the Sentinel docs to have a look at what it would imply. I'll comment on #237 so the conversation sticks to the relevant issue, in case users want to add something.