[WIP] Sync HTTP2 stream id between source and destination.

[mike820324]

This PR change the following features.

Slightly changed the log format :

Put the direction(source/destination) in front of the message.

Upgrade Hyper-h2 to 2.5.1 :

Since hyper-h2 2.5.0, it will check whether the header field contains Host or not. As a result, update some unittests according to this.

Sync stream id between source and destination:

The original code will drop some frames(especially the priority update frame. I forgot the reason why we drop these) and hence causing the stream id between source and destination unsync. This PR remove some of the code in layer/http2.py and just send the same stream id to the other endpoint.

P.S @chhsiao90 can you kindly remind me that why we added this code. Orz

Pending some frames:

Due to the fact that we will pending the HEADER frame until we receive the StreamEnded event, we need to pending some of the frames that we receive until our HEADER frame is sent.

• PRIORITY frame
• RST frame
• WINDOW_UPDATE frame

Prevent duplicate priority frame:

When hyper-h2 receive HEADER frame with priority updated, it will actually generate two events which are

• RequestReceived
• PriorityUpdated.

Based on our current implementation,

• we will sending out the HEADER frame with the priority updated
• sending out PRIORITY frame with the same priority parameter of the HEADER frame.

As a result, I have added a simple logic on top of the receive_data method to check any duplicate PriorityUpdate event.

TODO:

• [x] unittest
• [ ] code cleanup

This is still WIP. @chhsiao90 please help me review this part since you are expert in this category. Any advice is thankful.

[chhsiao90]

Wow, finally, you remembered this repository!

[chhsiao90]

Simply ask, you drop the src <-> dest stream id mapping, and just straight forward with same stream id ? I think that would be ok!

[mike820324]

Yes, I have drop the steam id mapping. Do you remember why we add this mapping code, I believe there used to be a bug related to this, but I forgot. :(

[chhsiao90]

I remembered that legacy hyper-h2 didn't support for jumping the stream id.

[mike820324]

Oh, thanks for reminder. So it should be okay to remove the mappings part of code?

[chhsiao90]

If the test is ok, I think it's fine to remove!

[coveralls]

Coverage decreased (-0.7%) to 86.109% when pulling 81cf5d3253dd277a74db884db72a4738d8bdd0fe on http2-streamid-sync into eb8e14069189c6ff11a4953daacdb1bec8fa5db0 on master.

[coveralls]

Coverage decreased (-0.7%) to 86.109% when pulling 81cf5d3253dd277a74db884db72a4738d8bdd0fe on http2-streamid-sync into eb8e14069189c6ff11a4953daacdb1bec8fa5db0 on master.

[chhsiao90]

That me simply describe what I understand about the pending event and header receive event.

You introduce pending event is for the purpose to delay the event sending to the destination stream of WINDOW_UPDATE, PRIORITY, RESET that the stream still in idle state. And the header receive is just fulfill the pending event functionality.

But why we need this? Is the hyper-h2 restriction?