Enable keep alive on tcp connections via feature

[DoumanAsh]

This enables keep alive on tcp sockets via keep-alive feature

Do you want to have it configurable so that only ConnectionManager would enable it?
There is no harm in enabling it for short living connections as default intervals are quite long

Note that this doesn't really fix anything, but for long living connections it will speed up failure detection

Related to #885

package socket2 v0.5.3 cannot be built because it requires rustc 1.63 or newer, while the currently active rustc version is 1.60.0

Unfortunate sad state of rust networking

Just for reference redis(server) default value for keep-alive is 300 seconds (Ref)

[jaymell]

Thanks for this!

Do you want to have it configurable so that only ConnectionManager would enable it?

Is there any reason to limit it to ConnectionManager? Also, please correct me if I'm wrong, but currently the sync connection would need a similar setup in order to use keep-alives, correct? If so, I think it might be reasonable to feature-gate keep-alives/the socket2 dependency behind a new feature we could enable by default and let the whole lib use it. What do you think?

[DoumanAsh]

Is there any reason to limit it to ConnectionManager

It is unclear how user will use regular connection and whether he wants to use keep-alive.
but in general keep-alive itself has impact only on long living connections, so that's why I asked whether you'd prefer to enable it on all connections or only long living one.

Also, please correct me if I'm wrong, but currently the sync connection would need a similar setup in order to use keep-alives, correct

Yes, in similar manner it can be enabled for sync clients.

If you prefer to feature gate the whole keep-alive on all connections, then I'll do it for all connections

[DoumanAsh]

I refactored PR to hide socket2 behind keep-alive feature

[jaymell]

If you prefer to feature gate the whole keep-alive on all connections, then I'll do it for all connections

I think it sounds like a good approach!

[jaymell]

Sorry for the delay, but one more request: Can we enable the feature by default?

[nihohit]

not for this PR, in order not to block it, but we should probably also use this in sync connections, not only in async.

[jaymell]

@nihohit Sorry I'm not sure I follow your last comment.