Re: [quicwg/base-drafts] HTTP/3 unidirectional streams arms race (#2559)

[Lucas Pardue <[e-mail]>]

> So having one extra round-trip when opening one greasing stream sounds not unreasonable.

I'm ok with that but it also risks affecting all unidirectional stream types. Server push somewhat skirts the issue by providing a signal (counting PUSH_PROMISEs) that enhances the clients knowledge. QPACK is part way there because non-zero SETTINGS_QPACK_MAX_TABLE_CAPACITY implies that some credit from both ends is required. 

Perhaps we could summarise some guidance for people about the RTT incurred by these streams and ways to avoid them.

A totally different alternative that might turn out in practice easier is to simply set max_initial_uni_streams to a very large number, and set the initial_max_stream_data_uni to a low number (8 bytes would be enough to peak at the stream type). This way the client has far more insight into what the other side is attempting to do and can exercise stream control as suited to its needs. All while avoiding this weird cycle.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/2559#issuecomment-476961392