I’ve had to do this multiple times over the past two years. It’s always a shock to me how hard it is.
The Lambda@Edge functions especially. They have 4 phases they can be injected into: viewer-request, origin-request, origin-response, and viewer-response. If you put your code in the wrong place you get hard to debug issues. Additionally you cannot “annotate” a request, so a decision at say “viewer-request” cannot pass information to later stages.
Also, deployments take 15 minutes at least which just further frustrates the debugging process.
Every AWS service is shockingly crusty. Even after spending years on AWS calibrating my expectations downwards, it regularly finds ways to surprise me.
I know, I know. For developers it beats obtaining permission every time you want to spend $5/mo on some plumbing and for managers it beats getting fired because you chose Google cloud and they canceled it the next year. Still... ugh.
This was actually one of my most delightful experiences with Azure Functions and Microsoft's API Management solution - overhead does not make it the right fit for every serverless architecture, but it did make orchestrating functions in this way a lot easier
I totally agree. CloudFront is maddeningly anemic, with hacks required to do the most basic things. Nginx/Varnish configuration is bad but CloudFront is worse. I wonder how much better competitors like Cloudflare and Fastly are.
The Lambda@Edge functions especially. They have 4 phases they can be injected into: viewer-request, origin-request, origin-response, and viewer-response. If you put your code in the wrong place you get hard to debug issues. Additionally you cannot “annotate” a request, so a decision at say “viewer-request” cannot pass information to later stages.
Also, deployments take 15 minutes at least which just further frustrates the debugging process.