Is there any place proper to log the grpc or http error info?

[inevity]

Bug Report

Just use tower-http trace's new_for_grpc.
And do server timeout test : client pass grpc-timeout 1H, and server build with timeout 1ms, here use tonic::server::timeout, not tower timeout.
Then no any log in the server tonic , even enable the tower-http trace and trace level.

Can hack the tonic service to log the time expired error.
But how can get the error info log by the middleware or get current rpc status from my service code?

Version

0.6.2

Platform

Crates

Description

[davidpdrsn]

The middleware probably gets applied in the wrong order which is a bug.

The only workaround for now would be for you to write to your own timeout middleware (perhaps use tower's) and apply it after TraceLayer.

Adding this to the 0.7 milestone but I don't it'll require breaking changes to fix.

[inevity]

1. tonic::transport::Server::builder().layer(mylayer).timeout().add_service....
   client got time expired and code Cancelled.
   but tracelayer cannot capture any error.
2. mylayer: tower::ServiceBuilder::new().timeout().layer(TraceLayer::new())
   then add this mylayer to the tonic Server::builder()
   h2 log error, but tracelayer cannot capture any error
   client got transport error, code: Unknown
3. mylayer: tower::ServiceBuilder::new().layer(TraceLayer::new())..timeout()
   tracelayer on_failure callback capture this error:response failed classification=Error: request timed out latency=18 ms
   h2 log same error as 2's log2: response failed classification=Error: request timed out latency=18 ms
   client got transport error, code: Unknown

And use the tower timeout , however, we as client got transport error, though tracelayer on-failure can correctly classifiy as the request timed out error .

So we need fix the tower timeout middleware or add a tower-http timeout middleware to support return time expired errror message. And deprecate the tonic::Server timeout method.

[inevity]

Just move the tonic::transport::Service::timeout code to tower-http. Now run ok .
Some questions:

• If use the TimeoutExpired Error, which show as a h2 error in the tracelayer log , the grpc client parse it message as 'transport error', Unknow status code. Did not find where the server code send/return this error?
• How return correct error/status code with message? which method we prefer?

  1. Can we use use tonic's status struct in the tower-http's timeout middleware code which i just create to reply a grcp status?

  2. Should we create a tower-http middleware map_response_error to convert this h2/or the wrong grcp status to correct status code(Canceld, time expired)?

  3. use tonic client to convert this status to correct.

Retest on the master of tonic now, preview on 0.6.2:
Even have fix(transport): Correctly map hyper errors #629, fix(tonic): Expose h2 error instead of reason #883, found the error still cannot convert by the try_from_error.

Error is 'tonic::transport::Error(Transport, hyper::Error(Http2, Error { kind: Reset(StreamId(1), INTERNAL_ERROR, Remote) }))', try_from_error process result: cannot not parse as grpc error, cannot parse as h2 error, can not parse by the source chain, then return the Error above.