Chromium instances behind a load balancer bugs with setRemoteInstance()

[timmywiff]

I have recently set up a load balancer with two servers, both running Chromium v124. This is to replace the Puppeteer bundled version of v88 which has generation issues.

I am generating two different PDFs in a loop. I have setRemoteInstance() pointing at the load balancer. One PDF seems to generate fine and inspecting it shows it was generated by Chromium v124. The other PDF has styling issues and inspecting it shows it was generated by Chromium v88.

Is there some default behaviour that will ignore setRemoteInstance() if it fails for any reason? I can't figure out what is going wrong.

I can see in the docs that

If no instance is available at the given endpoint (instance crashed, restarting instance, etc), this will fallback to launching a chromium instance.

I have no indication of any issues with my Chrome instances. I would assume that it would work.

[timmywiff]

It does seem like the load balancer is the weak link here. Using sticky sessions does not seem to help. Bypassing it and using the server directly works as expected. Does anyone have any advice?