Reset connection after HEAD?

[ronag]

Continuing from nodejs/node#34350.

Some servers send a body with HEAD requests which is kind of un-parsable. There is however one way to be resilient against it. Always reset the connection after receiving the message headers for a HEAD request.

This would make things significantly slower though but much more resilient. How important are HEAD requests?

Should we just ignore this unfortunate case and hope that the target server behaves properly?

Opt-in/Opt-out option?

[ronag]

@szmarczak @delvedor

[szmarczak]

Can't we just throw on transfer-encoding: chunked?

[szmarczak]

If there's any data that we don't expect, we can also reset the connection at that point.

[ronag]

Can't we just throw on transfer-encoding: chunked?

That's a valid header for a HEAD response?

If there's any data that we don't expect, we can also reset the connection at that point.

Hard to determine. What if the extra data (i.e. HEAD body) contains data that looks like a proper response?

[szmarczak]

That's a valid header for a HEAD response?

Ah, you're right.

What if the extra data (i.e. HEAD body) contains data that looks like a proper response?

Then resetting the connection would make sense, but indeed, it would be a significant performance drop. But there are lots of edge cases like that so I wouldn't go for this.

IMO, if it looks like a proper response, it should be parsed, and if the path doesn't match with the pipelined request (or there isn't any) then throw reconnect.

[szmarczak]

But there are lots of edge cases like that so I wouldn't go for this.

E.g. GET receives content-length: 0 but there is actually a body...

[ronag]

Yea, I guess it's not really solvable.

[ronag]

E.g. GET receives content-length: 0 but there is actually a body...

We could reset connection after content-length: 0 as well 😄

[ronag]

So it turns out that Node actually does reset connection after HEAD. See, nodejs/node#12396.

I think it might make sense for us to do so as well.

Thoughts? @mcollina @szmarczak

[mcollina]

+1