DevOps for Foundlings

Yehuda Deutsch

DevOps for Foundlings - JJC - 2026-06-28

What is DevOps?

DevOps is whatever stops "it works on my machine" from becoming "it's broken for everyone else"

  • What is your Dev role?
  • What is your Ops role?
  • How big is your Dev/Ops gap?
  • Where are your current release pains?
DevOps for Foundlings - JJC - 2026-06-28

Who am I?

Yehuda Deutsch

Software Developer with a sweet tooth for infrastructure

Principal DevOps Engineer at Palo Alto Networks
SDLC / Internal Platform

Open Source / Data / Knowledge activist

Self Hoster

DevOps for Foundlings - JJC - 2026-06-28

Define your route

The highway is not always the fastest route, and a fast car might not always get you where you need

  • How many users?
  • How many developers?
  • What's the cost of downtime?
  • What's your budget?
DevOps for Foundlings - JJC - 2026-06-28

Define your MVP

Your MVP is the smallest thing that works for someone who isn't you

  • What's the one job this product does?
  • What can break without anyone caring?
  • What does "shippable" mean for you?
  • What's deliberately out of scope?
  • Would you use it as-is?
DevOps for Foundlings - JJC - 2026-06-28

Version your source code

Version control isn't just about the code - it's your safety net, your history, and your rollback button

  • Everything in git
  • Tag your releases
  • Keep branches simple
  • Treat your deploy script like code
  • Source of Truth
DevOps for Foundlings - JJC - 2026-06-28

CI/CD is about guardrails

CI/CD isn't about fancy pipelines - it's the thing that catches you before you hit the ditch

  • CI catches your mistakes
  • CD prevents the scary deploy
  • It protects you from yourself
  • Start with one file
  • A green pipeline is trust
DevOps for Foundlings - JJC - 2026-06-28

Environments

You really need two: one to break things, one to serve users.
The rest are different degrees of acceptable breakage

  • Production
  • One non-production
  • Same config, different values
  • Skip the environment zoo
DevOps for Foundlings - JJC - 2026-06-28

Configuration and Secrets

One .env file committed to git is all it takes to ruin your week

  • Config is sourced from environment variables
  • Same build, every environment
  • Secrets never in git
  • If you leak a key, rotate it immediately
DevOps for Foundlings - JJC - 2026-06-28

Observability

You don't need Datadog, You need to know your app is down before your client emails you

  • The minimum viable monitoring - Is it up? Is it erroring? Is it slow?
  • Get an alert when something breaks
  • Read your logs
  • Alert on symptoms, not noise
DevOps for Foundlings - JJC - 2026-06-28

Rollbacks and Charge Ahead

If you can't roll back, every deploy is a gamble - Small projects can't afford gambles

  • Keep the previous version
  • One-command rollback
  • Health checks matter
  • Deploy small, deploy often
  • Hotfix should be a normal deployment
DevOps for Foundlings - JJC - 2026-06-28

The Lo Ta'asses (Don'ts)

  • Don't jump to buzz tools because of the buzz
  • Orchestration is an obstacle when you don't need it
  • Microservices will split your mind if used too early
  • Don't over compensate with tools, use the few systems that do more than use many tools that cover all
  • Don't be a "Poshet Yad" (Beggar)
  • Don't let an "Eved Ki Yimloch" situation (Reverse Position)
  • Security Fatigue is real and mean - don't overprotect
DevOps for Foundlings - JJC - 2026-06-28

ATAQ?

DevOps for Foundlings - JJC - 2026-06-28

This is the way

x59.us

y@x59.us

gitlab.com/uda/talks

DevOps for Foundlings - JJC - 2026-06-28

* Productization - how do you make your code a product * Productionization - how do you make your code shippable * Where are the cracks things fall through, what gears squeak every time * Are you afraid of "Thursday / Friday Deploys"? do you have large Merge Request you can't review?

* 10 friends, 500 customers, or a dream of "going viral someday"? * Just you, or a friend helping on weekends? * If your blog is down for a day, do you scream? If your client's invoicing tool is down, you're getting calls * $0, $5/month, or $20/month? This single number eliminates 90% of your tool choices

If it can't be used by a stranger, it's not shippable - it's a demo * Strip it to a single sentence. If you can't, you're building two products and shipping neither * No password reset? Fine. No analytics? Fine. No login at all? Maybe fine. Know what's actually essential * A URL someone can visit? A working API? A CLI tool someone can install? Define the finish line before you start running * Write down what you're not building yet. Auth, payments, scaling, polish - the "later" list protects your "now" * If you wouldn't, why should anyone else? Be your own harshest first user

* Code, configs, Dockerfiles, deployment scripts, even your notes - If it matters, it's tracked * You should always know "version 1.3 is what's running right now" instead of guessing * Trunk-based or GitHub-flow - no 14-branch GitFlow ceremony for a project only you touch * It gets reviewed, versioned, and tested, not left as an untracked deploy.sh that only works on your machine * Your code should be the source of truth to everything but secrets

* Tests, linting, type checks on every push. It's not judging you, it's saving you from shipping a typo to prod * If every merge auto-deploys, there's no "big release night" where everything goes wrong. Small, frequent, forgettable * Working late? CI runs the tests so you don't have to think about it. Tired deploy is a safe deploy * 30 lines of CI YAML (GitLab, Github, Woodpecker), Don't build a Jenkins empire for a project only you touch * When it's green, you ship with confidence. When it's red, you fix it before anyone else notices

* Where your users are, Treat it with respect * local dev, or a staging environment if you need to test on real infrastructure * Don't build a staging environment that's secretly nothing like prod, Same Dockerfile, same base, only secrets and scale differ * dev, QA, staging, pre-prod, UAT is enterprise sprawl, You don't have those problems

* You can template the config into code, but there is no benefit over loading from environment variables * One Docker image works everywhere, Only the env vars change * .env in .gitignore, always, Use your platform's secret management * Hope in this case will definitely harm you, act ASAP

* Uptime checks, Error tracing, APMs etc. * Email, Slack, Discord webhook - noisy channels will be disregarded * That means - make them readable * "the site is down" yes, "CPU hit 80% for 30 seconds" no

* Your PaaS does this automatically. If you deploy manually, keep the last working artifact. * Whether it's a PaaS button, a git revert + redeploy, or swapping a Docker tag. Practice it once before you need it * If your deployment goes out unhealthy, auto-rollback or at least get alerted * The smaller the change, the smaller the blast radius, the easier the rollback * Depending on the scale, hotfix should simply be like any other version release

* Resume-driven development - choosing tools because they're trendy, not because you need them * One container on a $5 VPS would've been fine * Start with a modular monolith; split when something actually needs to scale independently * 3 monitoring tools, 2 CI systems, and you can't remember how they connect * Spreading across 8 free services with fragile limits instead of paying $5 for one reliable thing * Don't spend 3 weekends setting up Terraform for a static site with 20 visitors * Protection is important, but when it breaks usability people will get used to circumventions