http: guard against failed sockets creation

[refack]

Fixes: #13045
Fixes: #13831
Ref: #13352
Ref: #13548 (comment)

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

Affected core subsystem(s)

http,async_hooks

[refack]

@nodejs/async_hooks silly question, what do I pass to nextTick if I don't have a new _handle? Where's the "ambient" asyncId?

[mscdex]

I don't think this check is necessary. It's only the socket creation that's an issue.

[refack]

Ack.
There's still the possibility that createSocket will be overridden to create sockets that do not have socket._handle.asyncReset (this is a method that only tcp_wrap and tls_wrap have). I should have guarded for that (not simply for the existence of _handle), but now I don't think that's an interesting use-case.

[refack]

Ping @nodejs/async_hooks

[trevnorris]

Two minor things. One is to change usage of nextTick() the other is a comment about unifying the way tests fail inside a Promise. I don't think I'd consider either a blocker.

[trevnorris]

You can pass null here and the internal nextTick() will grab the correct id.

[trevnorris]

Should we throw the fatalError() code in lib/async_hooks.js into test/common (https://github.com/nodejs/node/blob/v8.1.3/lib/async_hooks.js#L54-L68)? This way there's a standardized way of failing. e.g.

const { promiseFatalError } = require('../common'):

Promise.reject(42).catch(promiseFatalError);

[refack]

Ohhh, fancy 🎩!
I like!
New PR coming.

[refack]

CI: https://ci.nodejs.org/job/node-test-commit/10899/

[refack]

CI: https://ci.nodejs.org/job/node-test-commit/10919/