Full Width 
Shortcuts 
Sign Up 
Log In 
Articles 
Reading List More from Jibran’s Perspective
I have failed, and that is exactly what I had hoped for a few months ago in this blog post. This is a good failure. It has taught me things, lessons I can use in the future to avoid failing this way again. But first a bit of context. What did I fail at? In February of 2024 I decide to try my hands on my first “Indie Hacker” hustle, something that would make me money on the internet without having to trade my time for it. A product instead of consultancy services that I usually provide. I had seen a number of people on Twitter (X) rave about how well their bootstrap templates were doing; and I had just gotten out of a consultancy project where I needed to connect a Next.js frontend to a Django backend. I thought it was the perfect project to start my indie hacking journey. I put up a launch post and started working, updating a build log as I went along. I gave myself until 28th March 2024 to finish it. That of course did not happen. Let’s talk about why I failed and what I learned. Episode 1: The one where I don’t understand the meaning of MVP My initial plan was to build a Django+Next.js boilerplate template the provided all of these: the base template that provided a Django backend & Next.js frontend working authentication b/w the backend & frontend Dockerfile that would create the backend & frontend containers for deployment Terraform scripts to setup an infrastructure on AWS Celery + Redis for background task processing TailwindCSS for the frontend (comes mostly for free with Next.js) social auth This looks like something achievable in a week or two of work - but only if you’re working full time on this. I failed to consider that I have a day job and a life. I was barely able to tick of the first two of these deliverables by the time my 6 week deadline came up. As a good friend told me later, I should have focused on the minimum amount of value I could deliver. Just having the first two things on my list be done would have been enough. I couldn’t charge the $20 I had planned for, but I could have charged $1-$5 for just that. And if no one was interested in spending the cost of a coffee on the MVP of the template, that would have been a good signal that this wasn’t going anywhere in it’s current shape. Instead, by focusing on building something much bigger, I robbed myself of the ability to validate the idea quickly. I spent all my available time coding the template instead of trying to talk to potential customers about it. Lesson 1: Scope down aggressively. Episode 2: Where I jumped on the hype-wagon I settled on building a boilerplate template because that’s what I had seen a lot of people on Twitter/X doing lately; I’m chalking this down to recency bias. I had no personal interest in a boilerplate template. It’s also not a product that I would personally use. I have so far made one project that uses this tech stack. Most of my other projects are Django, and Ruby on Rails. The most successful boilerplate templates I come across are from people who made a bunch of projects in 1 tech stack then realized they needed to do the same thing over-and-over again; which they then packaged into a template they could use. Selling to others was a bonus at first I guess. I was very enthusiastic about the project at the start, but as time went on I had to force myself to work on it. My lack of interest in this type of project was a big factor. Another factor was there being no way to see the fruits of my labor. I am currently working on an analytics dashboard for another client (a RoR project) and every time I build a feature, I love to play around with it in my free time. I test how it works, make sure the UX is a good one, and just play around and admire the app I’ve made. Without me using my template to build new projects, I lacked that feedback loop. Without the loop, I quickly lost interest. Lesson 2: Build something I can use myself. This isn’t a job I’m getting paid for, so the only motivation I have initially until it starts generating money is to build something interesting for myself. Episode 3: Where I had nothing for potential customers to play around with This is related to the 1st lesson. Because I didn’t have a path to quickly get something out there, there was no way for me to get my “product” into the hands of people who could test and provide feedback. I think the problem with a boilerplate template style of product is that you can’t give people a half-backed thing and ask them to test it. Unlike other SaaS apps, there’s no mid-way version of a template. Customers have to “buy-in” to use your template with any project they are starting. With SaaS, users can sign up and test, and then leave if they don’t like it. There’s no easy way of testing with a template. Lesson 3: Build something that can be tested by potential customers easily. For now, I’m going to stick with SaaS style web apps. Conclusion Moving forward: I’ll be working on web app products that users can sign up for and test very quickly. My next few experiments/products will be things that I can use myself as well. I’ll post what I’m going to work on next when I decide and have some time away from my job & freelance projects that are currently in progress.
If you’re just looking for implementation instructions, skip my ramblings and go straight to the code here. I’m currently working on my first project after deciding that I needed to fail more and practice finishing projects instead of abandoning them midway once they got “boring”. Anyways… This one is till in it’s interesting phase, so here’s a blog post with some things I learned yesterday while working on it. The project is a boilerplate template that should make it easy for devs. to start a new project with a Django backend and a Next.js frontend, something I had to struggle with recently. The problem The first thing I’m looking to solve is authentication. That was my biggest challenge when working on the contracting project that inspired this template. While there are a number of good posts around how to setup authentication b/w Django & Next.js, nothing “definitive” came up and I had to cobble together a weird mess of Django+DRF (Django Rest Framework) and Next.js+NextAuth, sharing a token from Django that was masquarading as a JWT token for Next.js. It wasn’t pretty and I knew I could do better. The options I considered 2 options for authenticating the Next.js frontend with the Django backend: Token based auth. On logging in, a user receives a token that is stored in local storage by the frontend and send with every request to the backend. Session/Cookie based auth. This is how authentication works in Django by default and is very easy to get started with - it basically comes for free out of the box when you start a new Django project. While token based auth. is what almost everyone suggests to use when using a Next.js frontend with any backend technology, I wanted to give session based auth. a try. I was curious what it would take to make it work - if it was even possible. tl;dr: It was possible to use cookie/session auth. b/w Django & Next.js - though with a few constraints which make it less appealing than the token based solution What follows are my notes on how to set it up, the problems I faced, and why for the template I’m going to go with token based auth. instead. Learning how CORS & Set-Cookie works It took me a few hours to get my head around how cross-origin requests and cookies work together, but the actual implementation was surprisingly straight forward. This “mini-quest” gave me a chance to learn a lot about how CORS and cookies work, and I’m happy with the time I spent on this. These are the resources which helped me the most (all are from MDN): Cross-Origin Resource Sharing Same-origin policy Using HTTP cookies Set-Cookie And finally, there was a surprise waiting for me! Browsers are almost universally making changes to restrict 3rd party or cross-domain cookies because of their privacy implications. Here’s a nice article from MDN about it: Saying goodbye to third-party cookies in 2024. This is the reason why; while this approach works, I won’t be using it in the template. More on that later. Implementation Implementing the session based auth. b/w Django & Next.js is pretty simple. Django configuration Install the django-cors-headers Python package. Add "corsheaders", to your INSTALLED_APPS. Add the "corsheaders.middleware.CorsMiddleware", middleware, right above the existing CommonMiddleware. Set CORS_ALLOWED_ORIGINS = ["http://localhost:3000"], replacing the URL with your frontend URL. Set CORS_ALLOW_CREDENTIALS = True Configure settings.py to allow cross-domain access for the session cookie. Set SESSION_COOKIE_SAMESITE = "None" Set SESSION_COOKIE_SECURE = True Next.js configuration No configuration is needed on the frontend. However, you do need to use the credentials: "include", option when using the fetch() API to access your backend. Here’s a minimal example. "use client"; import { BACKEND_URL } from "@/constants"; async function signIn() { const loginData = new FormData(); loginData.append("username", "admin"); loginData.append("password", "admin"); return await fetch(`${BACKEND_URL}/accounts/login/`, { method: "POST", body: loginData, credentials: "include", }); } async function whoAmI() { console.log( await fetch(`${BACKEND_URL}/accounts/me/`, { method: "GET", credentials: "include", }), ); } export default function Home() { return ( <main className="flex min-h-dvh w-full flex-col justify-around"> <h1 className="text-center">Home</h1> <button className="" onClick={signIn}> Sign In </button> <button onClick={whoAmI}>Who Am I</button> </main> ); } That’s it. That simple piece of code & configuration took me hours to find. Hopefully you can use this example to skip all that time spent trying to figure things out. Side quest log: Initially, I was not using the credentials: "include" option in the signIn() function above; thinking that I didn’t need to send any cookies with the login call, only the second API call to the /accounts/me endpoint. That mistake cost me about 2 hours of debugging time. If I had RTFM correctly the first time, I would have seen this: include: Tells browsers to include credentials in both same- and cross-origin requests, and always use any credentials sent back in responses. The credentials: "include" not only controls if cookies are sent, but also if they are saved when returned by the server. Why I won’t use this solution in the template Browsers are phasing out 3rd party cookies (Saying goodbye to third-party cookies in 2024) and adding features to work around that restriction where needed. The simplest way that doesn’t require much change is to use Cookies Having Independent Partitioned State (CHIPS). To enable CHIPS, you simply put a Partitioned flag on your Set-Cookie header, like so: Set-Cookie: session_id=1234; SameSite=None; Secure; Path=/; Partitioned; Unfortunately, there’s no straight forward way to do this in Django for now. There’s an open issue to resolve this, but looking at the comments, it won’t likely be solved anytime soon. Considering this, I opted to use the token based auth. method for my template. I’ll write a blog on that once I get it working over the next few days.
Links: Gumroad page Build Log My accidental new years resolution was to work on the 1 problem that has plagued me for my entire adult life; failure to commit and focus. I decided to work in 6 week “sprints” (inspired by Shape Up) and complete the projects I start - for some known definition of complete. This is the 1st project I have decided to work on. I’ll work on this from today (15th Feb 2024) to (28th Mar 2024). I’ll follow-up then with another post talking about how it went. The project The goal is to make & sell a Django + NextJS boilerplate template. What’s a boilerplate template? It’s the source code for a project that’s already setup with many things that are needed in a new project; for example: Stripe subscriptions functionality Background jobs CSS framework User/team management A great example is Saas Pegasus, which seems like an amazing boilerplate loved by many people. My boilerplate is going to be much simpler - and also much cheaper. SaaS Pegasus comes with so many features that it’s worth the $249 starting price. I’m aiming for $5-$10. Goals My goal is to sell this boilerplate to at least 10 people - and have them be happy using it. This means: talking to prospective customers and seeing if this can be useful to them. People will have the option of scheduling a 15 minute pre-purchase call with me for $5 to see if this would be useful to them. The payment is purely to make sure that I only spend time talking to people who are somewhat serious about purchasing. providing excellent after sales support. I’ll include a 60 minute setup call with me for any purchase. While a 60 minute call for a $10 sale isn’t scalable, it’s a great way for me to talk to customers at the start. having a no questions asked refund policy. My experiences with running an e-commerce store in the past tell me this is an amazing way to build trust. provide on-going support, updates, and fixes over email. build a mailing list of people interested in my work who I can email when I launch my future projects. The deliverable The boilerplate will allow developers to quickly start a project that uses Django for the backend and NextJS for the frontend. My recent experiences with another project in this tech stack required me to spend significant time on: figuring out how to setup authentication b/w Django & NextJS (this took the most time & effort) setting up Django Rest Framework so I could write APIs that would be used by the frontend writing Docker files that would build 2 containers - backend & frontend writing Terraform scripts to deploy those containers to AWS ECS writing config & scripts to run the project on Gitpod so it could be easily worked on by my team members My plan is to build a boilerplate that already has most those features built in, plus a few extras: Celery with Redis for background task processing Tailwind CSS for the frontend (in my project I used ChakraUI but Tailwind would be a better option for a boilerplate) If there’s demand for it, a stretch goal is to include social auth (sign-in with Google/Apple/etc) Once complete, I’ll put this on Gumroad and create a landing page there. From then on, it’s all about marketing it; that’s the part which I have no experience with and hope to learn the most from. The marketing plan This is the area where I lack any experience; so I’m not sure how I’m going to market this. Some ideas I have: build it in public on Twitter. I have a tiny Twitter following (312 followers) so not sure how useful this could be. But I have to try something. share it with people asking how to setup Django & NextJS on forums like Reddit, Stackoverflow, and others. maybe write a blog post on how to setup Django & NextJS and then link to the boilerplate from there. The blog post would provider all the steps necessary for the basic setup and the boilerplate would go beyond that with something that’s ready to use. The build log I’d also like to create a build log with this project. This will be a daily note of what I did for this project. I’ll keep it in my notes app Reflect and periodically put it here in this blog post. These daily notes might also serve as content for my build-in-public marketing strategy.
I was recently listening through the How to succeed at failing series on the Freakonomics podcast and started to think about how often I had failed in the past few years. The first answer was - not too much. I couldn’t think of too many instances of where I had “failed”. This was not a good thing. It was very much a sad thing that I couldn’t quickly think of many things I had failed at. Why is not failing bad? I realized I didn’t fail a lot because - I didn’t try a lot. I didn’t fail because I didn’t have good goals. Sure, I have a folder full of projects I thought of, researched, some that I even started working on; that I gave up on pretty soon after starting. But those don’t count as failures right? I was just playing around, not really aiming for anything. It was just a fun side project. That’s why I didn’t fail a lot, and that’s also why it’s not a good thing. I need to fail more Which means that I need to set goals and work towards them. Without goals to aim for, I keep spending my time in the same unproductive loop. Think of a shiny new project. Play around a bit, maybe learn a few things. Give up when things get hard or boring. Repeat While I do learn some things, I don’t really learn new things. Picking up another web backend framework when I’m very good at 1 already (Django if you’re looking to get something built 🙂) is very small incremental progress. I should be learning about running a business, marketing, sales, how to talk to people, cold calling, etc. Instead I’m playing around with NextJS/Remix because it’s easier to do and there’s very little chance of me failing at it. What I’m planning to do Any new project that I start, I will have an achievable goal for it. Something that can be called “done”. Achievable is different for everyone. Creating a small SaaS and have 5 people pay for it might be childs play for some, it’s my Everest for now. The plan is to aim for things just outside of my reach – to work towards those and grow in the process. Have a deadline. For now I’m starting with 6 week cycles of project work. In order meet deadlines, I need to have a clear picture of the goal. Goes back to the first thing on this list. Having a clear definition of “done” is important. But being able to cut scope to achieve some version of the goal is also important. - I’ve just finished reading Shape Up by Jason Fried and I like the Basecamp approach of deciding on an appetite (time commitment) to a project and cutting scope to meet that appetite. - That’s my plan moving forward. Each project I work on will be “shapped” into an achievable outcome and will have an allocated appetite - when I want to launch it or decide to abandon it. Continuing it for the next 6 week cycle will be another option but should be done sparingly. This approach of timeboxing projects and having an achievable goal is also discussed by Zack Freedman in this video Here’s what’s preventing you from finishing projects. See it when you get the chance. This wasn’t supposed to be a new years resolution I started thinking about this over 2 months ago and have been planning to write a post on it. Today seemed like a good day for it. So I guess this is my new year’s resolution. Fail more, fail better.
More in programming
A clip of me singing a funny song from Gilbert and Sullivan’s Ruddigore back in 2013
In this video, we look at why fork() needs copy-on-write, how it works inside the kernel, and a memory usage problem that Instagram encountered with Python.
Comments require commitment, but they’re worth it.
Basecamp 5 runs on Puma in cluster mode: one master process with preload_app! and 63 single-threaded workers per host, deployed as a Docker container with Kamal. We serve Basecamp from several sites. Each site has its own web hosts and a read replica of the database, and writes go to a single primary database in one of them. On our busiest hosts, each deploy left up to 2,000 requests waiting while the new workers warmed up. We reduced those queues by running signed-in requests through the app in the Puma master, before it forked the workers. Why 63 single-threaded workers? Basecamp has always served web requests from processes rather than threads. It ran on Unicorn, which only does processes, until we moved to Puma in January 2025, and we kept the same setup: workers (Concurrent.physical_processor_count * 1.3).ceil threads 1, 1 preload_app! On a 48-core host that’s 63 workers, each handling one request at a time. We chose 1.3 after benchmarking HEY in 2023, when we moved our apps out of the cloud and onto our own hardware. We tested several combinations of workers and threads with a mix of GET and POST requests on a 32-vCPU VM. Every multithreaded configuration we tested was slower and handled fewer requests than single-threaded workers. Adding workers beyond about 1.2 to 1.3 per vCPU brought little benefit. The threaded workers spent a lot of their time waiting for Ruby’s global VM lock. That made single-threaded workers a good fit for this workload, and we use the same setup for Basecamp. An app that spends more time waiting on its database or other services may benefit from more threads, so benchmark your own app. The other reason is the app itself. Basecamp has class-level state in places and has never needed to be thread-safe. With one request per process, it still doesn’t. Processes do use more memory than threads, and preload_app! reduces the difference. The master loads the app once and the workers share its memory through copy-on-write until they write to it. Shopify’s comparison of Ruby execution models explains the trade-off well. In the HEY benchmark the best setup came to about 260 MB of PSS per core, where PSS counts each shared page once, split between the processes using it, and the gap to a threaded setup was smaller than we’d expected. What Puma does on each host when a container starts: one master, then 63 forked workers that share its memory until they write to it. Two things about this setup matter for the rest of the post. A worker that’s compiling or loading something is fully blocked — there’s no other thread to pick up the next request. And whatever the master has in memory before it forks, all 63 workers share. Whatever they build after the fork, they build 63 times. What happens when we deploy Kamal starts the new container alongside the old one, and kamal-proxy moves the host’s traffic across as soon as the health check passes. At that moment, the new workers have handled health checks but no customer requests. preload_app! means the master loads the app once and the workers inherit it through fork. That covers the code. It doesn’t cover anything Ruby and Rails set up on first use: YJIT compiled code. YJIT compiles a method once it’s been called a certain number of times. The master calls very little during boot, so every worker compiles the same methods again on its own first requests. Compiled templates. Action View turns each ERB template into a Ruby method the first time it’s rendered. The schema cache. Active Record reads each model’s columns from the database the first time that model is used. Inline caches and memoized values throughout Ruby, Rails and the app. All 63 workers did all of this at once, while serving the traffic the old container had been handling a second earlier. In the test environment with YJIT on, the first request to a project page on a cold process took 652 ms, 151 ms of it YJIT compiling. The same request to a warm process took 28 ms. In production, CPU time per request peaked at around 200 ms while kamal-proxy moved traffic to the new container, against about 30 ms once the workers had warmed up. A host with spare CPU absorbs this. Every one of our web hosts has 48 cores and 63 workers, but each Amsterdam host serves around 250 requests per second, against 25 to 60 at our other sites. In Amsterdam the slow first requests turned into a queue. At a peak-hour deploy, the Puma backlog on an Amsterdam host reached anywhere from 250 to 2,238 requests, and kamal-proxy’s p99 response time hit about 10 seconds. Eron, our Director of Operations, had been tracking this since June. Another server in Amsterdam would help, but it would take weeks to arrive, so we also wanted to make deploys cheaper on the hardware we already had. What didn’t work We tried a few things first. In June, Donal tested the first two on a single Amsterdam host, comparing it with its neighbors, and they ruled out two likely causes. Warming each worker’s database connections. Puma’s before_fork hook clears the master’s connections, and each worker opened its own on its first request. Opening them in before_worker_boot instead made no difference. Queries on a freshly booted production host were already under a millisecond, so connections weren’t the problem. A synthetic request in each worker. Next, each worker made a few requests in before_worker_boot to an internal controller that touched every model. That ran the middleware, routing and Active Record paths, but it ran them in 63 workers at once — exactly the CPU spike we were trying to avoid. And a request with no real data renders no real views, so most of the app stayed cold. Spreading YJIT compilation out. Delaying YJIT in each worker by a random interval spread the compiling out over a few minutes, but every worker still ran interpreted until its delay ended. The queue didn’t change. Reforking from a warm worker. This is what Shopify’s Pitchfork does: let one worker serve traffic until it’s warm, then fork the others from it. Puma has an experimental version called fork_worker, and on beta it worked — the reforked workers were warm after three to five requests, where fresh ones took up to 30 seconds. But with fork_worker the template is worker 0, and it keeps serving requests. If it exits, the workers waiting to be forked never start (puma/puma#3596). If it gets no traffic, the refork never happens, which is what we saw on beta. Instacart have a mold_worker patch that promotes a warm worker to a template that stops serving, but it isn’t in a Puma release. We have a branch of it, and we may come back to it. That last experiment did show us where the fix was, though. Everything a warm worker has that a cold one lacks is in its memory, and fork copies memory. The master already has the app loaded. It just never runs it. Run the requests in the master So now, before the master binds its socket and forks, it makes the app’s own requests, in-process, the way a signed-in user would. Rack has a hook for exactly this. Rack::Builder#warmup takes a block that’s called once with the built app, before the server starts. rails server builds the app from config.ru, so the change to boot is one line: require_relative "config/environment" warmup { WarmUp.configured.run } if ENV["WARM_UP"] run Rails.application With preload_app! this runs in the master, and the workers inherit whatever it did. Puma binds its socket after the app is built, so until the warm-up finishes the health check’s connection is refused and kamal-proxy keeps retrying. No request reaches a worker that hasn’t been warmed. The warm-up has three steps. After precompiling the views, it gives the page requests and schema loading a shared 20-second budget, checked before each page or model. 1. Precompile the views actionview_precompiler reads every template for its render calls and compiles each one with the locals it’s passed. For us that’s 1,394 templates in about two seconds. A first request to a project page then compiles 2 templates instead of 44. 2. Request the pages, signed in A small browser class makes the requests through Rack::MockRequest, with the two cookies a real sign-in sets, then goes back for each page’s lazy Turbo frames: class WarmUp::Browser def initialize(signed_in_as:) @client = Rack::MockRequest.new(Rails.application) @headers = { "HTTP_USER_AGENT" => "Basecamp warm-up", "HTTP_COOKIE" => cookie_for(signed_in_as), "bc3.warm_up" => true } end def visit(path) page = get(path) frames_in(page).each { |id, src| get(src, "HTTP_TURBO_FRAME" => id) } end private def get(path, headers = {}) @client.get("https://#{host}#{path}", @headers.merge(headers)) end def frames_in(page) Nokogiri::HTML5(page.body).css("turbo-frame[src]").map { |frame| [ frame["id"], frame["src"] ] } end end The requests are signed in. The user is a monitoring account we already use for automated checks, and the pages are its own project, Campfire, to-dos, documents and messages. Public pages weren’t enough: after warming up with signed-out pages only, the first signed-in request to the projects page still took 131 ms, because authentication, the signed-in controllers and their views had never run. With signed-in pages it took 40 ms. cookie_for writes the same signed cookie the sign-in controller does, using the app’s own cookie jar, so there’s no API token and no secret to store. The frames are followed. The busiest HTML requests in production aren’t pages at all but Turbo frames — the sidebar badge, the inbox, the navigation menus. The browser parses each page and requests its <turbo-frame src> URLs with the Turbo-Frame header, so those controllers and views get warmed too. Our first four pages turned into 60 requests. The requests are excluded from rate limiting. They are internal, so they do not count against the rate limits that apply to real visitors. 3. Load the rest of the schema The page requests load the schema for the models they touch. The last step loads the rest, from the read replica: ApplicationRecord.reading do models.lazy.take_while { time_left? }.each { |model| model.load_schema if model.table_exists? } end The step checks 261 models and loads any schema information still missing. Those database round trips add up when the primary is far away: outside a request, Active Record uses the writing role, and from a host a long way from the primary each round trip is tens of milliseconds. Reading from the local replica brings the step down from about 20 seconds to 3.5. The pages go first because they load most of the schema anyway. If the time budget runs out, the step stops, logs how many models it got through, and the workers load the rest on first use like they always did. Rails can also load the schema from a dumped cache file at boot (bin/rails db:schema:cache:dump), which would make this step unnecessary. We don’t ship one in our image yet, because the dump needs a database to read from at build time, and we have several databases to cover. It’s on the list. What to close before the fork Running requests in the master opens things the master never opened before, and every worker inherits them. Two processes writing to the same socket will corrupt each other’s traffic, so you need to know what’s open before you fork. The way to find out is to list the master’s open file descriptors — ls -l /proc/<pid>/fd — before and after a warm-up, in an environment set up like production. Development wasn’t enough for us: it stores files on disk, so our S3 connections only showed up in production. Then, for each thing that’s open, check how its library handles a fork. We found three kinds: Already handled. Plenty of libraries detect a fork on their own, either by recording the PID they connected from and reconnecting in the child, by opening per-process files, or by resetting their thread pools. Redis clients, metrics libraries and concurrency libraries tend to be in this group. Check, but you probably don’t need to do anything. Already closed. Database connections are the classic one, and most Puma configs already clear them in before_fork. Anything else that’s opened per process — we have a SQLite cache the workers open on boot — needs closing when the warm-up finishes. Needs a new step. HTTP clients with keep-alive connections are the ones to look for: cloud SDKs with connection pools, tracing exporters, error reporters. They usually have no fork handling at all. We empty the aws-sdk connection pools in before_fork, and we run the warm-up untraced so the OpenTelemetry exporter never opens its connection to Tempo in the first place. Once that’s done, before_fork finishes with Process.warmup, which Ruby 3.3 added for this purpose: a major GC, a heap compaction, and every surviving object promoted to the old generation, so the memory pages the workers share change as little as possible afterwards. Choosing the pages The first list was the four pages that ran the busiest requests on beta. Once the warm-up was live, production showed us which endpoints were still cold. For one deploy, we compared each endpoint’s mean duration in the six minutes after kamal-proxy moved traffic to the new container with the same endpoint an hour later, then multiplied the difference by the number of requests in those six minutes. That gives the extra time each endpoint cost us because it was cold: Endpoint Cold Warm Requests in 6 min Extra seconds Campfire 246 ms 70 ms 6,490 1,140 Projects (JSON API) 84 ms 50 ms 22,077 771 Docs & Files 262 ms 177 ms 4,996 421 To-dos tool 205 ms 113 ms 4,018 371 To-dos (JSON API) 33 ms 16 ms 18,738 320 The pages already in the warm-up showed what to expect: the project page kept a 36 ms gap after a deploy, and the to-do page 10 ms. We’ve proposed adding these five requests, and expect them to add about five to seven seconds to the page step. The two JSON endpoints were a surprise. The warm-up’s page list had no API requests in it, so nothing on the API path had run before the first real request: not the API controllers, and not the Jbuilder templates rendering real records. Precompiling the views covers JSON templates too, but it isn’t a substitute for running the request. Results The warm-up is on for all 68 web hosts. With the first four pages it took 12 to 16 seconds per host: about 2 seconds to precompile the views, 7 to 9 for the 60 requests, and 3.5 for the schema. Deploys take that much longer per host, and we raised the deploy timeout from 30 to 60 seconds to cover it. In Amsterdam, at a peak-hour deploy: During deploy Before After Peak Puma backlog per host 250–2,238 requests 19–223 requests Peak kamal-proxy p99 about 10 s 2.4–4.8 s Peak CPU time per request 201–214 ms 88–132 ms Peak database time per request 56–69 ms 39–47 ms The same eight hosts at three deploys on 1 October, an hour apart, as the warm-up went from one host to four to all eight. The deploy in the middle, with four hosts warmed and four not, shows why every host needed the warm-up. Each warmed host recovered faster on its own: mean request duration peaked at 130 to 173 ms, against 203 to 311 ms on the hosts that weren’t warmed. But the backlog on all eight was about the same, because they were all waiting on the same database. Mean request duration on each host at the 07:21 UTC deploy. Blue hosts warmed up in the master before forking, orange hosts did not. Memory came down too. The workers now share compiled templates, YJIT code and the schema with the master instead of each building their own copy. On beta, the view precompiler alone took a busy worker’s private memory from 174–202 MB to 119–135 MB. Thirty minutes after the deploy, the web containers used about 39 GB less memory than the previous day’s containers at the same age and traffic. Amsterdam served most of our traffic at the times we tested. In Amsterdam, each new container used about 2 GB less just after traffic moved to it, which lowers the peak while the old and new containers overlap. Working with Claude Claude Code helped throughout. It combed through the per-worker backlogs and per-endpoint timings in Prometheus and Loki after each deploy, worked out the cold-versus-warm cost of each endpoint, and prepared the changes and the pull request descriptions with the benchmarks in them. We decided what to try, deployed it and read the results. If you do this Warm the master before it forks. Compile common code and templates and load their schema in the master, so workers inherit that work. With preload_app!, Rack::Builder#warmup runs before the workers start accepting traffic. Use the app’s real requests. Public pages, internal endpoints and synthetic queries warm the paths they run and nothing else. Signed-in requests to real records, frames included, run what production runs. Measure the cold penalty per endpoint. The difference between an endpoint’s cold and warm duration, times its request count after a deploy, ranks the pages worth adding. Ours weren’t the ones we’d have guessed, and two of them were JSON. Check what the warm-up leaves open. List the master’s file descriptors after a warm-up and account for every one before the fork. Two of ours needed changes. Set a time budget. A warm-up that runs long on one slow host fails the deploy on that host. Ours gives the page requests and schema loading a shared 20-second budget, checked before each page or model, puts the most valuable pages first, and logs what it skipped. Reforking from a warm worker, as Pitchfork does, solves the same problem continuously rather than once at boot, and it would warm paths no fixed list of pages covers. We may still get there: our branch brings Instacart’s mold_worker up to date with Puma’s main branch and fixes the bugs we found in it. But warming the master works with the Puma we already run, took a few days to implement, and substantially reduced the queues after deployment.
