Fluid Compute caused huge increase in latency and page rendering times

Help
1

Hi all,

We have a high traffic website that incurs a fairly large monthly bill, so we decided to try out Fluid Compute to see what efficiency improvements we could get. With the recent communications from Vercel regarding moving to Fluid, we thought it could only be an improvement.

While we did see lower overall usage, it made the site almost unusable due to a massive spike in TTFB and general function run time. According to our Sentry logs, the longest page load times were around 4 minutes, with regular page loads between ~10s and ~1min. The higher the traffic, the worse it got.

So we disabled it and got in touch with support. They suggested that it could be a problem with our application implementation, but offered no help in pointing out what that might be. They couldn’t provide any documentation that outlines potential problems with FC.

So my question to the community is, has anyone else experienced this? Or can anyone offer any insight into implementation patterns that should be avoided for FC?

I don’t think we’re doing anything too out of the ordinary. Tech stack:

  • Next 15 with App Router
  • Next Auth
  • Storyblok CMS
  • BigCommerce BE
  • Vercel Runtime Cache for Storyblok
  • AWS AppSync for microservice integration
  • Sentry error tracking
  • ISR for static pages

The site uses Catalyst as a base: GitHub - bigcommerce/catalyst: Catalyst - for Composable Commerce

Graph go up when FC enabled:

068PZ00000X8TUuYAN
068PZ00000X8TUuYAN954×422 36.9 KB

4

This turned out to be caused by the Storyblok JS Client! It had an internal rate limit and throttling mechanism that defaulted to 5 requests per second. This was fine in a browser or serverless environment, but when using Fluid Compute that rate limit was shared across many different functions. Increasing this limit has resolved the issue.

In response to this, Storyblok have now updated the JS client to have a default limit of 1000, which mirrors their CDN limit.

5

Thanks for following up! I had tried to deploy a catalyst app but couldn’t find any way to reproduce the TTFB issues, it didn’t occur to me the CMS would be the cause here

1 Like