feat(node): Fork isolation scope in tRPC middleware

[andreiborza]

Writing tRPC handlers today and using top-level methods like Sentry.setTag isn't very intuitive as the isolations scope is not forked per procedure in our middleware.

This PR changes the middleware to fork the isolation scope, while this is not 100% correct, as it breaks the one isolation scope per process/request model, it should be more intuitive and work better for most users.

Resolves: #16262

[AbhiPrasad]

Should we have a check for something like so?

getIsolationScope() !== getDefaultIsolationScope()

so that we only use an isolation scope if it hasn't been explicitly set? This way we don't break server-use cases, like using trpcMiddleware in your express app.

[lforst]

I think this should be fine even on express. Middleware only runs on the server. What are your concerns?

[AbhiPrasad]

I'm just worried we rebind a new isolation scope which interferes with a framework integration.

Like if you use the trpc middleware in a cloudflare worker: https://trpc.io/docs/server/adapters/fetch#create-cloudflare-worker

import * as Sentry from "@sentry/cloudflare";
import { initTRPC } from '@trpc/server';
import { fetchRequestHandler } from '@trpc/server/adapters/fetch';
import { createContext } from './context';

const t = initTRPC.context().create();
const sentryMiddleware = t.middleware(
  Sentry.trpcMiddleware({
    attachRpcInput: true,
  }),
);
const sentrifiedProcedure = t.procedure.use(sentryMiddleware);
export const appRouter = t.router({
  getUserById: sentrifiedProcedure.input(...),
});

export default Sentry.withSentry(() => {}, {
  async fetch(request: Request): Promise<Response> {
    return fetchRequestHandler({
      endpoint: '/trpc',
      req: request,
      router: appRouter,
      createContext,
    });
  },
});

withSentry creates an isolation scope around fetch, but so would the getUserById trpc procedure. In this use case, we can just rely on the isolation scope from the fetch handler instrumentation.

[andreiborza]

The problem with this is that the isolation scope is already different from the default isolation scope, but the scope remains the same between the two middleware invocations which is why we end up with issue like tags "leaking".

[AbhiPrasad]

let's go for it - we can always re-evaluate this later.