Spawning process for every parallel request results in high CPU usage

[richardscarrott]

So this issue was raised a while back but there wasn't any reproduction steps which I guess is why it was closed.

I was profiling our node server and noticed from cold one of our APIs was spending a lot of time in the stripe SDK.

Here's a contrived way to reproduce the issue:

const paymentIntentIds = new Array(50).fill('pi_123');
await Promise.all(paymentIntentIds.map(id => stripe.paymentIntents.retrieve(id));

If run from cold (i.e. without having made a prior call to the SDK), it will spawn a new process for all 50 of the above requests because the getClientUserAgent caching is a little naive.

You can see the difference between a cold call:

And a 2nd warm call:

If you're wondering why one might be making so many parallel requests at once, our use case is fetching a page of orders in our DB which all have a FK reference to a payment intent in stripe (we don't currently replicate the payment intents in our DB) and unlike older stripe resources, there's no batch endpoint for payment intents AFAIK.

I discovered the issue in 7.5.3 but can also reproduce in v8.165.0.

[richardscarrott]

In the meantime, I've patched our stripe client with:

Stripe.prototype.getClientUserAgent = cb =>
  cb('HACK: https://github.com/stripe/stripe-node/issues/1202');

Hopefully won't cause any problems with the telemetry on your end?

[richardm-stripe]

Hello @richardscarrott, thank you for the report and the great detective work. A fix for this in #1205 has been merged and will be released soon.