Interceptors no longer work with RxJS retry operator

[geersch]

Did you read the migration guide?

• I have read the whole migration guide

Is there an existing issue that is already proposing this?

• I have searched the existing issues

Potential Commit/PR that introduced the regression

PR #11142

NestJS version

9.4.2 -> 10.0.3

Describe the regression

I maintain 2 packages @geersch/retry and @geersch/nestjs-retry. The former one offers a retry function supporting exponential backoff (powered by RxJS) and the latter allows you to use it via a NestJS interceptor.

@Controller('cats')
class CatsController {
  @Get()
  @UseInterceptors(new RetryInterceptor(FixedBackoffStrategy))
  findAll(@Headers('x-attempt') header: number): string {
    return 'all the cats';
  }
}

This worked fine with NestJS 9.x.x, but with NestJS 10 it breaks and the tests of @geersch/nestjs-retry package fail. The retry operator which is eventually piped into the next.handle() observable does retry X times, but it appears that the actual controller method is not invoked more than once anymore, breaking the retry mechanism.

Minimum reproduction code

geersch/retry#5

Input code

I made PR where I update the dependencies to NestJS 10.0.3.

geersch/retry#5

The automated tests (Github actions) fail for the @geersch/nestjs-retry package. See retry-interceptor.spec.ts for the failing tests. It's just a simple controller with a few controller methods and 4 tests which all fail.

You can reproduce this by cloning this repository, switching to the fix/nestjs10 branch and running the tests.

git clone git@github.com:geersch/retry.git
git checkout fix/nestjs10
yarn

Be sure to the build script (yarn build) from the root first before running the tests.

yarn workspace @geersch/nestjs-retry test --watch

Expected behavior

I would expect the controller method to be retried when I pipe this into the RxJS retry operator.

Other

No response

[shavidze]

Yes, I think this PR caused this issue. I'm not 100 sure but the potential reason is that the handle method of the handler object is still calling nextFn(i + 1), but now it's not wrapped in a function.

[micalevisk]

hi @nordfjord do you mind on helping us to investigate this issue? thanks in advance.

[nordfjord]

I would suggest reverting that PR if it's causing issues with something as important as retries 😬

[nordfjord]

I believe I have a fix in #11926

[kamilmysliwiec]

Let's track this here #11926