Discussion: Exit criteria for bringing http2 out of experimental

[jasnell]

It's time to start figuring out what are the final steps necessary to bring http2 out of experimental?

Let's build a list.

• Expectation + evidence of API stability (e.g. list of user-space modules/applications building on top of http2)
  • http2.live
  • fastify
  • grpc
  • ...?
• Modules in citgm that exercise the http2 API
  • fastify
• Reports of http2 deployed in production
  • http2.live
• Framework adoption
  • fastify
  • restify
• full (or at least fuller) test coverage

/cc @nodejs/http2 @nodejs/tsc

EDITED BY TROTT: added test coverage. I took the "let's build a list" as an invitation to edit. Feel free to undo my edit if I'm misunderstood.
EDITED BY OFROBOTS: edited 'citgm' bullet to complete as fastify has been added.

[ofrobots]

Suggestion:

• Expectation + evidence of API stability (e.g. list of use-space modules building on top of http2)

[jasnell]

Added... although I made it "modules/applications" because it's more likely to have more deployed applications building on the API than it is to have individual modules.

[mcollina]

• Having some modules in CITGM that exercise the HTTP2 API.
• someone reporting having deployed http2 in production.
• frameworks adopting it (we have already two, restify and fastify) - should we wait for Express?

[jasnell]

Added!

[wesleytodd]

Express is in an interesting position right now on this. This is the furthest along PR expressjs/express#3390.

I did some benchmarks last weekend to start proving out a best way forward for that PR, and was planning on spending some time over the next few weeks moving it forward. But if you wait for 5.x, I am worried at how long it will take.

TBQH, I really want to push 5.x out because it has been just dragging on for too long. So if this is the nudge we need to push it over the finish line that sounds great.

EDIT: I also recently meet with @DonutEspresso about collaboration between express and restify, and he mentioned that they had not looked at http2 support yet. Maybe I am wrong? Clearly I am wrong: restify/node-restify#1489

[devinivy]

hapi supports http/2 push under the underdog module. But my understanding is that framework-level http/2 support is in the med/long-term roadmap. Basically, http/2 works well with hapi today under the compatibility API.

[leodutra]

Would be possible to enable HTTP/2 without having to use a cert and key?
There's a lot of microservices running using HTTP without SSL/TLS and being shielded by Nginx / other proxies with a single point SSL/TLS config.

I think the answer will be dependant on something deeper, maybe Node core modules, but I really wanted to check and ask for it.

[styfle]

Would be possible to enable HTTP/2 without having to use a cert and key?

Yes, however it will not work if you attempt to connect to the server via web browser.

See http2.createServer vs http2.createSecureServer and note the comments in the code snippet 😉

[mudrz]

Adding hapi issue link for reference hapijs/hapi#3584

[paambaati]

Should the list of userspace modules also include client-side request modules? None of the popular ones (request, axios, got, urllib & needle to name a few) support HTTP/2 right now.

[devsnek]

node-fetch is waiting on http2 to have agents, but otherwise there's already a pr open to it with initial support

[mcollina]

node-fetch is waiting on http2 to have agents, but otherwise there's already a pr open to it with initial support

Considering that there is no PR in the work for that, or that I didn't know about this requirement, it might be a very long wait. It would be really good to get that implementation going. My idea for that is described in #17746.

[mcollina]

Should the list of userspace modules also include client-side request modules? None of the popular ones (request, axios, got, urllib & needle to name a few) support HTTP/2 right now.

There is no Compatibility layer for the client. We might want to add that, but it was never on-scope for the initial release, and I would be somewhat concerned about making that a requirement for the exit criteria.

I wrote https://github.com/mcollina/h2url for that.

[jasnell]

@nodejs/tsc @nodejs/http2 ... at this point, the HTTP/2 API has stablized. There are ecosystem users (most significantly Google's grpc now builds on top of http/2 in core). There have been some major bug fixes. I think we are reasonably close to having http/2 ready to come out of experimental. I'd like to target the end of August for the official switch over.

Any objections? Any remaining tasks?