Fix the following issue:
Problem Statement
I’m seeing our vsrpc endpoint fall over when one of my RPC methods hits a panic — the client just gets a dropped/hung connection instead of a normal JSON-RPC failure for that call. It’s also awkward to verify because the debug service doesn’t seem to have a simple RPC method I can call that deliberately panics with a message.
Expected outcomes
Panic handling for vsrpc calls
- If a vsrpc RPC method panics while handling a request, that panic should be reported to the caller as a JSON-RPC error for that request instead of causing the endpoint, process, or connection to fail.
- After one RPC call panics and returns an error, the same server/connection should remain usable for subsequent RPC calls.
- This behavior should apply consistently to RPC methods adapted through the existing vsrpc handler helper APIs, whether the method returns only an error or returns a value plus an error.
Error details
- The JSON-RPC error for a recovered panic should use
jsonrpc2.InternalError. - The error message should start with
panic:, include the panic value, and include stack trace information.
- The JSON-RPC error for a recovered panic should use
Debug service verification
- The debug service should expose a callable vsrpc method named
TestPanicAsync. - Calling
TestPanicAsyncwith a message should deliberately trigger the panic path and return the same JSON-RPC internal error behavior, including the supplied message in the error details.
- The debug service should expose a callable vsrpc method named
Implementation notes
The specific recovery mechanism, helper structure, and placement of validation or error conversion are left to the implementer. Preserve the existing request unmarshalling, cancellation, and normal success/error behavior for non-panicking calls.