dgram: skip dns lookup for IPs

[evanlucas]

Previously, we would call dns.lookup even for IP addresses
in Socket#connect. This now skips that to improve performance.

Fixes: #39468

[schmod]

I'd avoided PRing this myself, because I couldn't figure out the best way to share isIP between net and dgram – I don't think we want dgram to depend on net, because it seems like it could introduce subtle changes to the order that modules load, which seems like it could break many APM implementations.

[evanlucas]

@schmod, good point, I got a rebase conflict somehow :/ so have to fix that regardless. With it loading internal/net for isIP, I'm not sure that would change anything as far as module load time?

[schmod]

With it loading internal/net for isIP, I'm not sure that would change anything as far as module load time?

True, I'm not really worried about that. It's more about whether or not anybody's hot-patching any modules used downstream of net.

That all being said, that sounds like an extreme edge-case, and might not be worth worrying about?

[jasnell]

While it may seem odd, this may be a breaking (semver-major) change. User code is permitted to provide a custom dns lookup function when the dgram/udp socket is created. Prior to this change, that function would be called for IP address values.

[jasnell]

This should include (a) a test case that verifies that the lookup function is not called and (b) a doc update indicating that a custom lookup function will not be invoked when the address is an IP address.

[schmod]

While it may seem odd, this may be a breaking (semver-major) change. User code is permitted to provide a custom dns lookup function when the dgram/udp socket is created. Prior to this change, that function would be called for IP address values.

We could preserve the current behavior if a custom lookup is provided, and postpone the breaking change to the next major release...

[JackFreelander]

@evanlucas - Would it be possible to revive this PR and complete the fix? There seems to be no work-around for dgram's socket implementation invoking DNS resolution on every single .send() call without this fix. It would be great to have this.