60 replies

1. Private reply Mastodon 15 min later
2. Private reply Mastodon 21 min later
   1. Private reply Mastodon 33 min later
      1. Private reply Mastodon 16 hr later
         1. Private reply Mastodon 17 hr later
3. James Doc Mastodon 22 min later

   @zachleat Thanks for the write up here. Appreciate it.

   1. Zach Leatherman Mastodon 34 min later

      @jamesdoc you’re welcome!

4. Paweł Grzybek Mastodon 27 min later

   @zachleat this was a very interesting read Zach.

   At the company that I work for, we received a surprisingly high bill from Vercel and after looking closely into the usage breakdown we noticed that the number of used cache units is so so so high. Too high to be even possible taking into consideration our revalidation interval and number of pages. Unfortunately Vercel CS is not able to give us a detailed breakdown of where we spent these cache units 🤷‍♂️

   PS. I love the term „convenience tax” !

   1. Zach Leatherman Mastodon 37 min later

      @pawelgrzybek yeah, more data is needed to justify some of the numbers I was seeing. They didn’t compare with my experience on Netlify or AWS

5. James Miller Mastodon 29 min later

   @zachleat There are so many ways to do things in AWS! If you eventually want a friendlier first-party package/deploy story that isn't the result of clicking around the console, check out SAM. Also not sure how many requests per month are hitting it, but you could eliminate API GW in favor of a Lambda function URL and save those costs. Come on in, the water's fine!
   https://docs.aws.amazon.com/serverless-application-model/latest/developerguide/serverless-getting-started-hello-world.html
   https://docs.aws.amazon.com/lambda/latest/dg/urls-configuration.html

   1. Zach Leatherman Mastodon 31 min later

      @james oh fascinating. I’ll definitely have a look at those—I’ve only just started!

      1. John Christopher Mastodon 54 min later

         @zachleat @james

         https://sst.dev is pretty nice as well.

         1. Zach Leatherman Mastodon 59 min later

            @jgchristopher @james very nice! I’ve also been pointed at https://www.serverless.com/ and https://arc.codes/ which seem like similar abstractions

            1. James Miller Mastodon 1 hr later

               @zachleat @jgchristopher Yep, both are solid. I choose SAM because it’s AWS-native CloudFormation without a third party abstraction or licensing/cost considerations. But totally understand that's not a priority for everyone - third party tools definitely do some heavy lifting for you.

               1. Brian LeRoux 💚 Mastodon 1 hr later

                  @james @zachleat @jgchristopher fwiw Architect is a wrapper for the SAM transform and open governance so no licensing concerns

                  1. James Miller Mastodon 1 hr later

                     @brianleroux @zachleat @jgchristopher yeah arc.codes is great! licensing shade was only being thrown on serverless framework 😬

                     1. Brian LeRoux 💚 Mastodon 1 hr later

                        @james @zachleat @jgchristopher rightfully so. ironically but over the years we got a bunch of shade for being in a foundation. At least ppl get why now.

   2. Naiyer Mastodon 21 hr later

      @james @zachleat I'd also recommend SAM CLI for deploying lambda functions. It is relatively easy to use and you can take the reigns in your hand with a Makefile. I'd still recommend API Gateway if you want to implement your own authorization mechanism which you can deploy in another lambda and configure it for your API Gateway resources. Makes it convenient to put OIDC introspection in front of multiple REST endpoints.

6. Raymond Camden Mastodon 34 min later

   @zachleat query - you shared an estimated cost on Vercel, what is the estimated cost of Lambda?

   1. Zach Leatherman Mastodon 37 min later

      @raymondcamden so far the projected usage is still under the free tier for the month!

      1. Raymond Camden Mastodon 40 min later

         @zachleat nice! i dont think you mentioned that, i would. (its also 99% possible you did and i missed it.)

         btw, _damn_ good article!

         1. Zach Leatherman Mastodon 54 min later

            @raymondcamden thank you! I just pushed a small note about the AWS cost

            1. Cory Mastodon 1 hr later

               @zachleat @raymondcamden I'm over here with Cloudflare Pages, Digital Ocean for my admin, Supabase, B2, bunny.net, Echofeed, Feedpress, Plausible analytics 🙈

               1. Raymond Camden Mastodon 1 hr later

                  @cory @zachleat I've liked Cloudflares serverless stuff a lot. Their hosting has been on my "to test and blog" list for a while

                  1. Cory Mastodon 1 hr later

                     @raymondcamden @zachleat yeah, I quite like their workers stuff and I'm well below the paid threshold for those. I've got one for analytics, contact form, now playing display, schedule rebuilds and music tracking — 100,000 runs a day is super generous for the *tiny* scale I'm dealing with. 😅

7. Stuart Langridge Mastodon 2 hr later

   @zachleat “ because usage is priced via GB-Hrs, 1 vCPU should use 1 GB of memory. Instead, 1 vCPU is 1.7 GB of memory and 1000 GB-Hrs is actually 588 Hours of usage.”

   You know how washing powder people tell you “use one cup” but the cup they provide you has an almost invisible mark halfway up saying “1 cup” and if you fill the cup it’s actually 1⅔ cups? Yeah.

   1. Zach Leatherman Mastodon 18 hr later

      @sil ha! incredible parallel 🏆

8. Joe Lanman Mastodon 14 hr later

   @zachleat very nice! Small thing - was playing around and this syntax in your documentation always gives me 404

   https://v1.screenshot.11ty.dev/:url/

   eg

   https://v1.screenshot.11ty.dev/https%3A%2F%2Fgov.uk/

   it works if I add the other url params (size)

   1. Joe Lanman Mastodon 14 hr later

      @zachleat also wondering if you considered Cloudflare, I agree AWS is painful

      https://developers.cloudflare.com/browser-rendering/get-started/screenshots/

      1. Zach Leatherman Mastodon 17 hr later

         @joelanman a little bit! I know they have their own fork of puppeteer that might have worked nicely here https://github.com/cloudflare/puppeteer

   2. Zach Leatherman Mastodon 17 hr later

      @joelanman fixed, thanks!

9. benschwarz Mastodon 22 hr later

   @zachleat if this were Twitter Vercel staff would be all up in replies explaining what happened… but they aren’t, they never left Twitter.

   To me the fact they Vercel and all thier staff were able to look past what X has become for the sake of their hype machine is a bad signal

   1. Zach Leatherman Mastodon 22 hr later

      @benschwarz …yeah. It isn’t just Vercel—a lot of the JS-dev ecosystem is that way. Definitely hurt my career when everything split.

      1. Pelle Wessman Mastodon 23 hr later

         @zachleat @benschwarz But we get quality over quantity over here ☺️

      2. benschwarz Mastodon 23 hr later

         @zachleat I can use a boatload more 🤯 emoji if that helps?

         1. Zach Leatherman Mastodon 23 hr later

            @benschwarz the only engagement that performs a meaningful function on mastodon is boost 😅 (favorites are for good feelings only)

   2. Jan Lehnardt Mastodon 23 hr later

      @benschwarz this.

10. Pelle Wessman Mastodon 23 hr later

    @zachleat Serverless functions are essentially the good old PHP CGI model adapted to the cloud and given a hype name.

    Just like PHP struggles with idle threads during async I/O, so does serverless functions.

    Depending on which serverless function provider one uses it ranges from being autoscaled worker with a strict one payload at a time processing (and no prefetch of next item) to being fresh executions for each and every request.

    1. Pelle Wessman Mastodon 23 hr later

       @zachleat Whenever possible I would opt to run workers myself and work on dialing in how many simultaneous payloads and prefetched ones they should have to maximize throughput, latency and cost

       1. Pelle Wessman Mastodon 23 hr later

          @zachleat I remember reading up on that theory when I worked with RabbitMQ some 5-10 years ago, something like this: https://www.rabbitmq.com/blog/2012/05/11/some-queuing-theory-throughput-latency-and-bandwidth

          Also get reminded of good old https://gearman.org/ which we eg used with our PHP setup on Flattr back in the days and could handle extreme spikes in activity without any spikes in cost – only spikes in queue lengths

          1. Francis 🏴‍☠️ Gulotta Mastodon 1 day later

             @voxpelli @zachleat yes and you get guaranteed resources per request, and price per request, and incredible scaling capabilities. It’s absolutely not appropriate for every workload, but can adapt quickly and be very cheap. And despite some companies best efforts to downplay it, it’s often pretty trivial to move to a server execution model when it makes sense.

             1. Francis 🏴‍☠️ Gulotta Mastodon 1 day later

                @voxpelli @zachleat I used to work in publishing and we would occasionally get viral successes. While the primary content is incredibly cachable. There was quite a lot of features that weren’t. Lambda helped us slice our server costs 90%. While the operational cost was higher during peaks. We only paid for it during peaks. And had very simple ops model.

                1. Pelle Wessman Mastodon 1 day later

                   @reconbot @zachleat Yeah, like with everything it’s about making informed choices and uses the tools for what they do best

                   We used Lambda@Edge when I worked at a newspaper to handle some of the scalability challenges

11. Munawwar Firoz Mastodon 1 day later

    @zachleat You can skip api gateway cost and save with lambda function URLs

    (and to prevent "denial of wallet" attack, possibly set concurrency limit)

    Other than that I've written few optimization tips:

    https://codepasta.com/2024/04/19/optimizing-puppeteer-pdf-generation

    BTW GCP cloud run makes this so much more easier with a simple Dockerfile

12. Francis 🏴‍☠️ Gulotta Mastodon 1 day later

    @zachleat

    > Mind-boggling that the same service would use ~1276 GB-Hrs per month on Vercel and ~101 GB-Hrs on AWS. Mind-boggling that the same service would use ~1276 GB-Hrs per month on Vercel and ~101 GB-Hrs on AWS.

    This is with all the caching in place, not the same workload hitting lambda, right?

    In either case yes mind boggling, and I used to run the team that owned that.

    1. Zach Leatherman Mastodon 1 day later

       @reconbot yeah, with caching—71% cache misses on the Vercel side. ~85% cache *hits* from CloudFront.

13. Zach Leatherman Mastodon 1 day later

    Found some stats:

    28.8% cache hits on Vercel 😬
    81.2% cache hits on CloudFront

    1. Jason Garber Mastodon 1 day later

       @zachleat money printer go brrr

    2. Ryan Townsend (old account) Mastodon 1 day later

       @zachleat all CDNs are not created equal when it comes to cache eviction.

       Yet another thing often selected because of some flashy feature or promise that has knock-on significant impact on origin load and financial implications 🙄

       Remember folks: the difference between 95% and 90% cache hit ratio is a doubling of origin traffic!

       1. Zach Leatherman Mastodon 1 day later

          @ryantownsend what‘s the difference between 90% and 28% 😬

          1. Ryan Townsend (old account) Mastodon 1 day later

             @zachleat bankruptcy.

14. Private reply Mastodon 2 days later
    1. Private reply Mastodon 2 days later
       1. Private reply Mastodon 2 days later
          1. Private reply Mastodon 2 days later
             1. Private reply Mastodon 2 days later
15. Compare Serverless Providers Mastodon 5 days later

    @zachleat Thanks! This is awesome! Keep it up!

16. Zach Leatherman Mastodon 27 days later

    Update after one billable month: the 11ty Screenshots service cost $1.95 on AWS 🏆

    1.16M screenshots served 👀
    86.2% cache hits

    1. Bob Monsour Mastodon 27 days later

       @zachleat With this edit, you answered my question.

    2. Andy Bell Mastodon 27 days later

       @zachleat cost them more to raise an invoice for you lol