Add fetch upload test on HTTP 421 response.

[yoichio]

Following whatwg/fetch#982,
fetch upload streaming on 421 (Misdirected Request) should be rejected.

[yutakahirano]

Don't we need a test for a non streaming request?

[yoichio]

I think so, but how might we test that requests should be sent twice?

[yutakahirano]

You can use https://web-platform-tests.org/tools/wptserve/docs/stash.html.

[yoichio]

Added a test for a non streaming request.

[yutakahirano]

LGTM. It's difficult to check the connection identity.

[annevk]

We should also test the service worker responding with with this status code.

[yoichio]

Added service worker test.

[annevk]

@yoichio in #24965 @MattMenke2 added connection tests and I suspect you could use similar infrastructure here to determine that the connection was not reused.

[yoichio]

Used network-partition-key.py to check connection reuse.
However, it seems Chrome reuses at least port id, which the script uses to identify the connection, so the python can not identify if connection is actually reused or not.
@MattMenke2 , any good idea to do that?

[MattMenke2]

I don't understand the question. I don't know what port id is.

[yoichio]

I want to test if fetch on 421 response retries fetching same request with new connection.
I thought network-partition-key.py could distinguish connections on which each request was sent.

[MattMenke2]

network-partition-key.py does create a stash containing one entry for each client address+port, and assumes they aren't reused across sockets. OSes typically assign TCP ports in a sequential order (which isn't great...), so client ports are typically not reused for the duration of the test.

There's never a guarantee that a connection will definitely be reused between two requests.

I'm not sure what this test is doing or why it cares about socket reuse, or perhaps I could be of more help.

[yoichio]

Thanks. I'm confused between connection, socket and port.
I don't have strong opinion to reuse network-partition-key.py to confirm the spec behavior.

I'd like to confirm fetch on 421 does retry on new connection in any way.

[MattMenke2]

I'm not sure exactly what you're trying to do. Your current modifications to network-partition-key.py look to be counting the total number of requests use a particular uuid, across all connections. Do you want to know the total number of requests use a particular connection instead?

[yoichio]

Do you want to know the total number of requests use a particular connection instead?

Yes.

[MattMenke2]

Do you want to know the total number of requests use a particular connection instead?

Yes.

Sorry for being a bit slow on the uptake.

So "address_key" identifies the connection. So something like ++server_state[address_key + b"connection_count"] should work, assuming you initialize it to 0. Or perhaps better, could make a dictionary for each address key with connection count and partition_id fields, and just increment the count on each use (At least I believe recursive dictionaries work with the API used to store state).

[yoichio]

Thanks.
Updated the test to count # of requests and connections.

[yoichio]

@annevk, could you take a look?

[annevk]

Looks good modulo nit.