Version grpc-js@1.8.2 is broken

[Elrendio]

Problem description

When updating to version 1.8.2 our scripts randomly stops without errors. It works when version is pinned to 1.8.1. I have not investigated further.

Environment

• OS name, version and architecture: Arch Linux amd64
• Node version 19.3.0
• Node installation method: yay
• If applicable, compiler version [e.g. clang 3.8.0-2ubuntu4]: unknown
• Package name and version [e.g. gRPC@1.12.0] grpc-js@1.8.2

[mbwhite]

We've also seen issues with 1.18.2 today. Initially it looked like a programming error that triggered the steam to close.

On a full duplex stream, a message is sent from client to server, the client always to instantly close the stream before the server can respond

[murgatroid99]

Can you share a trace of a client doing that, which you would get by running the client with the environment variables GRPC_TRACE=all and GRPC_VERBOSITY=DEBUG?

[mbwhite]

@murgatroid99 thanks will try and get the information - I see there's a PR as well; if there's a build for that available might be able to try it as well

[murgatroid99]

I published version 1.8.3 with the change from that PR. Please try it out.

[nikolaymatrosov]

I also observe the same strange behavior: my app silently exits unexpectedly. So I had to downgrade the package version even to 1.7.3.
Unfortunately, I cannot trace the reason even with debug env variables turned on.

The message I get when I use the debugger in the WebStorm is Process finished with exit code 130 (interrupted by signal 2: SIGINT). I understand it is probably not directly related to the @grpc/grpc-js package, but maybe someone else gets the same symptoms and could go into debugging further.

The most strange thing is that using 'grpc.service_config_disable_resolution': 1 instead of downgrading also works for me. And I do not understand why. The reason may be related to the DNS TXT resolution. But resolution errors should be ignored and not interfere with the success path.

[murgatroid99]

What do you mean by "Unfortunately, I cannot trace the reason even with debug env variables turned on."? Does it not output anything at all?

[nikolaymatrosov]

It outputs a lot of info, but I can't find there the reason why the process stops silently.

[murgatroid99]

The fact that disabling service config resolution stops the problem is especially weird because as far as I can remember, the only thing that has changed about the TXT resolution code path is adding that option to skip that code path. One possibility is that the TXT resolution is actually succeeding, and that is triggering the use of a service config that causes some misbehavior in other new code.