Use SOCK_NONBLOCK to reduce system calls for outgoing connections

[panjf2000]

What this PR mainly does is:

1. Refactor the anetCreateSocket() to make it more generic for more socket arguments, and use SOCK_NONBLOCK if available, which will reduce two system calls (F_GETFL and F_SETFL) of enabling the non-blocking mode on each newly created socket.
2. Clean up the unused anetUnixGenericConnect() that calls anetCreateSocket().

SOCK_NONBLOCK for system call socket() is supported on most UNIX-like platforms (linux, dragonfly, freebsd, netbsd, openbsd, solaris, etc.). This improvement will significantly reduce the system calls considering how massively anetTcpGenericConnect() will be called when needed.

As for the cleanup, anetUnixGenericConnect was introduced in c61e692 and the only reference back then was from the createClient() in redis-benchmark.c which had been removed in ec8f066 and made it the dead code. Most of that dead code was also cleaned up in f657315, and it seems that the anetUnixGenericConnect got left out. Therefore, I also cleaned it up, but I'm not so certain about doing this cleanup in this PR. Maybe you would prefer to do it in a separate PR?

References

• socket(2) on Linux
• socket(2) on FreeBSD
• socket(2) on DragonFly
• socket(2) on NetBSD
• socket(2) on OpenBSD
• socket(3c) on Solaris
• socket(3socket) on illumos

[panjf2000]

@madolson @zuiderkwast Got a minute for this?

[zuiderkwast]

It looks good in general. Just a few small suggestions.

I'm not sure it has a very big impact though. It's only used for creating the listening socket and for creating outgoing connections, which is basically only for replication and cluster nodes (maybe something more?) but we don't use it for valkey-cli, valkey-benchmark and (IIRC) sentinel which use hiredis for this.

What's more important is accept that we do for each client, but we already use accept4() when it's available (setting options at the same time as accepting).

[codecov]

Codecov Report

Attention: Patch coverage is 70.83333% with 7 lines in your changes are missing coverage. Please review.

❗ No coverage uploaded for pull request base (unstable@f4f1bd6). Click here to learn what that means.

Additional details and impacted files

@@             Coverage Diff             @@
##             unstable     #293   +/-   ##
===========================================
  Coverage            ?   68.44%           
===========================================
  Files               ?      108           
  Lines               ?    61558           
  Branches            ?        0           
===========================================
  Hits                ?    42136           
  Misses              ?    19422           
  Partials            ?        0           

Files	Coverage Δ
src/unix.c	74.32% <ø> (ø)
src/anet.c	70.41% <70.83%> (ø)

[panjf2000]

It looks good in general. Just a few small suggestions.

I'm not sure it has a very big impact though. It's only used for creating the listening socket and for creating outgoing connections, which is basically only for replication and cluster nodes (maybe something more?) but we don't use it for valkey-cli, valkey-benchmark and (IIRC) sentinel which use hiredis for this.

What's more important is accept that we do for each client, but we already use accept4() when it's available (setting options at the same time as accepting).

Speaking of which, check out #294 as well.

[panjf2000]

Any updates?

[zuiderkwast]

Any updates?

I left it open for a week to let others have time to review. The to-be-merged label makes sure we don't forget it.