grpc/grpc-node

Since version 1.10.9 grpc-node doesnt comply with grpc spec in case of non 200 response

RemiBou opened this issue · 7 comments

RemiBou commented

Problem description

When server returns HTTP 404 with "404 not found" in body grpc-node throws a RESOURCE_EXHAUSTED

Reproduction steps

Init your grpc client with a non grpc server, it should throw "UNIMPLEMENTED" error, instead it might throw "RESOURCE_EXHAUSTED"

Environment

  • grpcnode >= 1.10.9

Additional context

I think it's same issue as #2809
I want to add that this should be fixed in grpc-node as it doesn't comply with the grpc spec anymore. Reference
https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md

Implementations should expect broken deployments to send non-200 HTTP status codes in responses as well as a variety of non-GRPC content-types and to omit Status & Status-Message. Implementations must synthesize a Status & Status-Message to propagate to the application layer when this occurs.

When "404" is return by server then grpc status 12 and error UNIMPLEMENTED should be send to the client application.

Java and GO grpc library work that way and grpc-node used to work that way until this commit 674f4e3

murgatroid99 commented

OK, I see the problem: the RESOURCE_EXHAUSTED error from attempting to parse the response body is superseding the UNIMPLEMENTED error from the HTTP status.

  • 👍2
RemiBou commented

@murgatroid99 do you think it should be fixed ?

murgatroid99 commented

Nothing has happened with this yet.

erlendnils1 commented

We are experiencing a similar issue. A HTTP 502 error from the load balancer used to throw an UNAVAILABLE error, but now throws RESOURCE_EXHAUSTED "Received message larger than max" based on misinterpreting the response as a gRPC response.

tysseng commented

For others experiencing the same, the reported message body sizes in the case mentioned by @erlendnils1 are:

1633951815 = 0x61 0x64 0x20 0x47 or 'ad g', meaning the client received 'bad gateway'
1013478509 = 0x3C 0x68 0x74 0x6D or '<htm', when the client receives an error page from the GKE (Kubernetes) load balancer, which returns a message saying 'The server encountered a temporary error and could not complete your request. Please try again in 30 seconds.'

  • 😕1