Notes from the production trenches.

Practical, opinionated posts on the things that bite teams running real workloads: AWS, Kubernetes, Terraform, CI/CD, cloud cost and technical leadership. Written in Athens.

Cloud cost

  1. It's not the $33–38/month idle fee. It's the $0.045–0.052 per GB of traffic that never needed to leave the VPC. Most AWS accounts pay this tax silently, every month, on TBs of S3 reads.

  2. Most startup AWS bills are not too high because of one big mistake. They are high because of a dozen small ones nobody owns. This is the order we work through them in, with the Cost Explorer view that exposes each one.

  3. Hetzner is far cheaper per server and per terabyte of traffic. AWS sells you time back through managed services. Which one is right depends less on price lists than on what your team is willing to run itself.

Security and compliance

  1. security · compliance

    SOC 2 on AWS: a technical checklist for startups

    Your first enterprise customer sent a security questionnaire and asked for a SOC 2 report. Here is what SOC 2 actually requires of an AWS estate, which services cover which controls, and what evidence to keep from day one.

  2. A prospect sent a security questionnaire and asked for "SOC 2 or ISO 27001". You can afford one this year. Here is how to choose based on who your customers are, and how to avoid doing the work twice.

  3. DORA has applied since January 2025, and supervisors have moved from asking whether you are ready to asking for evidence. This is what the regulation expects from the engineering side of a small fintech on AWS, pillar by pillar.

  4. Most NIS2 guides for Greece are written by law firms. This one is for the people who have to build it: what Law 5160/2024 and its ministerial decisions actually require, and the order we would implement it in.

  5. The legal guides tell you what NIS2 requires. They rarely tell the engineer what to switch on. This is the mapping we use from the directive's security measures to concrete controls on AWS.

  6. Your application runs in eu-central-1. Your error tracker is in Oregon. The IP address that triggered the error is now personal data sitting on a server in Oregon. You may have just imported a sub-processor problem you didn't know you had.

  7. Most startups we meet still log in to AWS with IAM users and keep access keys on laptops. IAM Identity Center replaces all of that with one login per person, short-lived credentials and access you can revoke in one place. This is the setup we use, from enabling it to deleting the last IAM user.

  8. Long-lived AWS keys in GitHub secrets are the most common credential leak we find in audits. OIDC replaces them with short-lived credentials that only one repository, branch or environment can get. Here is the full setup, including the trust-policy details most tutorials get wrong.

  9. aws · architecture

    AWS Organizations from day one

    The cheapest blast-radius control AWS sells is free, and most teams don't turn it on until after the first time someone deletes a production database from the wrong shell window.

Migration and architecture

  1. Most migrations don't fail on cutover day. They fail in the weeks before, when nobody wrote down the cron job on the old server. This is the checklist we work through with Greek teams moving to AWS, GCP or Azure.

  2. Heroku is not shutting down, but it is no longer moving forward. Here is what actually changed, how the main alternatives compare for a team in Europe, and when a full move to AWS is worth the extra work.

  3. Every Heroku concept has an AWS equivalent, but the mapping is not one-to-one and the database is the hard part. This is the order we migrate in, the commands we use, and the cases where we tell teams to stay put.

  4. postgres · migration

    Zero-downtime PostgreSQL migration to Amazon RDS

    Moving a live Postgres database to RDS is a replication problem followed by a five-minute cutover. This is the method we use, the limitations that bite, and the queries we run before we let traffic in.

  5. aws · containers

    ECS vs EKS: which should a startup choose?

    Both run your containers well. The difference is how much platform you want to own. Here is an honest comparison, the point where Kubernetes starts paying for itself, and how to move from ECS to EKS later without a rewrite.

  6. The right answer for most Greek teams isn't the one trending on Hacker News. It's the one that bills you in EUR, runs near Athens, and matches the cloud your hires already know.

  7. Three years after the licence change, the fork has its own features, Terraform has kept shipping, and Pulumi now runs HCL. Here is how they actually differ and what we pick for startups.

  8. Every Terraform codebase reaches the same fork: keep one giant state file, or split it. The most popular split — by team — feels organised and is almost always wrong.

Reliability and operations

  1. Small teams don't need an incident management programme. They need one page that tells whoever got paged at 3am what to do next. Here is ours, plus how to run on-call with 3 to 10 people without burning anyone out.

  2. Most startups do not need a second live region. They need backups an attacker cannot delete, a restore they have actually timed, and a written plan. This is how we set that up on AWS.

  3. All three can keep your telemetry in the EU. The real differences are how the bill grows, how hard it is to leave, and how many engineer-hours the stack eats every month.

  4. Every new engagement starts the same way: one week, five dimensions, one page of findings. We've published the rubric so any team can self-run it before deciding whether they need outside help.

  5. Each of these patterns looks reasonable in isolation — sometimes even responsible. They're the ones that get added when something goes wrong. Then they cause the next thing that goes wrong.

  6. Mutable image tags break rollbacks, defeat caching and turn an outage post-mortem into archaeology. Pin image digests instead — your future self at 03:00 will thank you.

Leadership and hiring

  1. Published benchmarks for fractional CTO pricing in Europe now exist, but they come from vendors and they skip Southern Europe. Here is what they say, where they disagree, and what each price level should actually buy you.

  2. Most pages ranking for this question quote US or offshore rates. These are the European numbers, with sources, plus what actually moves the price and how to compare two quotes that look nothing alike.

  3. Most early-stage Greek startups don't need a full-time CTO yet. They need someone senior making the right technical calls a few hours a week. Here is how to tell which one you are.

  4. These three options are not really competitors. A freelancer and an agency are ways to get code written. A fractional CTO is a way to decide what gets written and check that it was written well. Most founders need to choose who does which job.

  5. Technical due diligence rarely kills a deal on its own. What it does is slow the round, cut the valuation or add conditions. Most of what reviewers flag can be fixed in a few weeks if you know what they look for.

  6. The Greek tech market is healthier than it has been in fifteen years and still produces the same hiring complaint from every CTO: "we can't find anyone." The complaint is half right and half a job description problem.

Tell us what you need