HTTP/1 Clear Text Requests to HTTP/2 servers wait forever, if called with DELETE method and Body

[ckosmowski]

Thanks for reporting an issue, please review the task list below before submitting the
issue. Your issue report will be closed if the issue is incomplete and the below tasks not completed.

NOTE: If you are unsure about something and the issue is more of a question a better place to ask questions is on Stack Overflow (https://stackoverflow.com/tags/micronaut) or Gitter (https://gitter.im/micronautfw/). DO NOT use the issue tracker to ask questions.

Task List

• Steps to reproduce provided
• [-] Stacktrace (if present) provided
• [-] Example that reproduces the problem uploaded to Github
• Full description of the issue provided (see below)

Steps to Reproduce

1. Create a Http Service with micronaut
2. Implement a simple GET route
3. Implement a second route with DELETE OR PUT with body parameters
4. Test the routes in any order (with simple HTTP/1.1 clients) -> it works as expected
5. Configure it to use Http/2
6. Test the routes again (with simple HTTP/1.1 clients)

Expected Behaviour

The routes will still respond correctly in any order

Actual Behaviour

• If we start with the DELETE or PUT route in our test, it will wait forever, no response is received, no exception is raised.
• If we start with the GET request first, and use the other route afterwards, all subsequent calls to DELETE and PUT will succeed.
• After waiting a while without calling, DELETE and PUT will fail again, until another GET request is issued.

Environment Information

• Operating System: Linux
• Micronaut Version: 2.2.0
• JDK Version: 8

Example Application

[volkerrichert]

I also had this issue with "PUT", as I described on gitter / questions on Nov. 25 18:16

[graemerocher]

Can one of you attach an example

[volkerrichert]

check out https://github.com/volkerrichert/mnIssus4643 and run tests. It will fail on put.

reorder test wit get first or switch to http 1.1

[volkerrichert]

current test will fail with:

Read Timeout
io.micronaut.http.client.exceptions.ReadTimeoutException: Read Timeout
	at io.micronaut.http.client.exceptions.ReadTimeoutException.<clinit>(ReadTimeoutException.java:26)
	at io.micronaut.http.client.netty.DefaultHttpClient$12.exceptionCaught(DefaultHttpClient.java:2258)
	at io.netty.channel.AbstractChannelHandlerContext.invokeExceptionCaught(AbstractChannelHandlerContext.java:302)
	at io.netty.channel.AbstractChannelHandlerContext.invokeExceptionCaught(AbstractChannelHandlerContext.java:281)

[graemerocher]

@volkerrichert the example works if you inject an HTTP/2 client

@Inject
@Client(id = "/", httpVersion = HttpVersion.HTTP_2_0)
RxHttpClient client;

So the actually problem is HTTP 1.1 clients can't talk to an HTTP/2 server, that clears things up. Investigating... not sure if a bug yet.

[volkerrichert]

But why does it work after doing the GET - request first? This request shouldn't work also. Thats's stange.

[ckosmowski]

If this is the case, i would suggest a downgrade feature or something. Because at least in my case you rarely are aware of all the clients using your server and you are not in control of changing the clients.

[graemerocher]

@volkerrichert there is a bug in the way the Netty channel is downgraded to HTTP/1.1 which I am looking into. But the root cause is attempting to send HTTP/1.1 requests over HTTP/2

@ckosmowski in this case the downgrade support is not working but only if the request has a body (like a PUT or a GET)

It seems to be a core Netty issue because the CleartextHttp2ServerUpgradeHandler which we are using (a Netty class) discards the body and prevents it from being read again when attempting to do the connection upgrade.

[ckosmowski]

@graemerocher Thank you for claryfiing.

When CleartextHttp2ServerUpgradeHandler is involved would you expect this to work with self signed certificates and SSL enabled?

[graemerocher]

Yes SSL uses a completely different codepath and doesn't involve CleartextHttp2ServerUpgradeHandler

[yawkat]

This is fixed by #6300 for the test case by @volkerrichert (thanks!). I have not tested it with DELETE, but the patch should apply to any first request with a body.