Preventing unhandled promise rejections in async functions

I am developing a HTTP proxy service and I have observed presence of an odd error message in my logs:

unhandledRejection RequestError: HTTP request error.
    at /dev/rayroute/raygun/src/factories/createRequest.js:107:13
    at processTicksAndRejections (internal/process/task_queues.js:93:5) {
  code: 'RAYGUN_REQUEST_ERROR',
  originalError: Error: test
      at /dev/rayroute/raygun/src/factories/createRequest.js:73:29
      at processTicksAndRejections (internal/process/task_queues.js:93:5)

It is odd because there are plethora of tests to ensure that all errors are handled. It is also odd because I have never never seen unhandled rejection while developing the service (only ever saw it in production logs).

The relevant code looks like this:

const activeRequestHandler = createRequest(requestDefinition);

if (incomingMessage.socket) {
  incomingMessage.socket.on('close', () => {
    if (responseIsReceived) {
      log.trace('client disconnected after response');
    } else {
      log.debug('client disconnected');

      activeRequestHandler.abort(new Error('CLIENT_DISCONNECTED'));
    }
  });
}

try {
  await actions.afterRequestActions(
    context,
    requestDefinition,
    activeRequestHandler
  );
} catch (error) {
  log.error({
    error: serializeError(error),
  }, 'afterRequest threw an error');
}

try {
  responseDefinition = await activeRequestHandler.response;
} catch (error) {
  log.warn({
    error: serializeError(error),
  }, 'an error occurred while waiting for a HTTP response');

  // [..]
}

It is pretty straightforward:

• createRequest initiates a HTTP request and returns a request handler
• the request handler can be used to abort the ongoing request (afterRequestActions aborts request after a hard-timeout); and
• it is used to resolve the response or eventual rejection of the promise

I have written tests to ensure that errors are handled when:

• request rejected
• request aborted
• afterRequestActions throws an error

, but all tests are passing.

🤔

It turns out that the problem was that in all my test cases actions.afterRequestActions was resolving/ being rejected before activeRequestHandler.response is resolved. Meanwhile, in production afterRequestActions contains logic that can take substantially longer to execute. I have also learned that even if you declare a try..catch block for your async function, if it resolves before it is await-ted, then you will get an unhandled rejection, i.e.

This code will not warn about unhandled rejection:

const delay = require('delay');

const main = async () => {
  const promise = new Promise((resolve, reject) => {
    setTimeout(() => {
      reject(new Error('Expected rejection.'));
    }, 100);
  });

  await delay(90);

  try {
    await promise;
  } catch (error) {
    console.error(error)
  }
};

main();

But this code will always produce a warning about an unhandled rejection:

const delay = require('delay');

const main = async () => {
  const promise = new Promise((resolve, reject) => {
    setTimeout(() => {
      reject(new Error('Expected rejection.'));
    }, 100);
  });

  await delay(110);

  try {
    await promise;
  } catch (error) {
    console.error(error)
  }
};

main();

The best solution is to add an auxiliary catch block, e.g. This is how I refactored my original code:

const activeRequestHandler = createRequest(requestDefinition);

// Without this we were getting occasional unhandledRejection errors.
// @see https://dev.to/gajus/handling-unhandled-promise-rejections-in-async-functions-5b2b
activeRequestHandler.response.catch((error) => {
  log.warn({
    error: serializeError(error),
  }, 'an error occurred while waiting for a HTTP response (early warning)');
});

if (incomingMessage.socket) {
  incomingMessage.socket.on('close', () => {
    if (responseIsReceived) {
      log.trace('client disconnected after response');
    } else {
      log.debug('client disconnected');

      activeRequestHandler.abort(new Error('CLIENT_DISCONNECTED'));
    }
  });
}

try {
  await actions.afterRequestActions(
    context,
    requestDefinition,
    activeRequestHandler
  );
} catch (error) {
  log.error({
    error: serializeError(error),
  }, 'afterRequest threw an error');
}

try {
  responseDefinition = await activeRequestHandler.response;
} catch (error) {
  log.warn({
    error: serializeError(error),
  }, 'an error occurred while waiting for a HTTP response');

  // [..]
}

Comments

[Greg Reimer]

Wow, it took me a bit to spot the problem before reading the rest of the post! I think there's a deeper problem, which is that createRequest() exposes a dangerous API to users by creating a reference to a promise and storing it implicitly.

const activeRequestHandler = createRequest(requestDefinition);
// activeRequestHandler.response is a promise that's
// now just hanging around, unhandled.

Anyone else using the API will have to be cognizant and use the same workaround. As an alternative, I like how the fetch API does things.

// creates and immediately handles a promise
const resp = await fetch(...);

// ...other logic runs here...

// creates and immediately handles a promise
const content = await resp.text();

Maybe createRequest() should change its API?

const activeRequestHandler = createRequest(requestDefinition);

console.log(typeof activeRequestHandler.response); // function

// other sync/async logic here...

await activeRequestHandler.response();

[Gajus Kuizinas]

For what it is worth, createRequest is just a light abstraction around got. It is got that uses a cancellable promise to implement cancel (/abort) functionality.

I do like your suggestion, though and I think it would be an improvement to got API. Please raise an issue with got. I am sure Sindre will appreciate the suggestion.

[Gajus Kuizinas]

The requirement to process interceptors in series is very much intentional.