From Hetzner Cloud to a Six-Account AWS Landing Zone: Isolated Environments and Continuous Compliance
Customer: TolaData GmbH · Industry: Software and Internet · Country: Germany · Segment: SMB
From Hetzner Cloud to a Six-Account AWS Landing Zone: Isolated Environments, Continuous Compliance
The challenge
TolaData's platform is used by international development organisations to plan, track, and evaluate the impact of their programmes. The software was doing its job. The foundation underneath it was holding the business back.
Everything ran on Hetzner Cloud, provisioned and maintained by hand. There was one environment, so testing and live traffic shared the same space and a mistake in one could reach the other. Deployments were manual, which meant releases depended on the availability of specific engineers. And because there was no organisation-wide record of configuration or activity, questions from partners and funders about who changed what and when were answered from memory rather than from evidence.
As the platform grew across more countries and more organisations, none of that was going to hold.
What we built
A landing zone with real account boundaries
Six AWS accounts in Frankfurt, built on AWS Control Tower and AWS Organizations. Three of them carry workloads. The other three exist to keep the platform honest: a management account that owns the organisation, an audit account that aggregates compliance findings, and a log archive account that receives the activity trail and is written to but not worked in.
The point of the separation is blast radius. A misconfiguration, a compromised credential, or a bad change is contained inside the account it happened in. And because the audit and log accounts are separate from the accounts they observe, the record cannot be quietly edited by the people it covers.
Three environments that stay apart
Production, staging, and sandbox each get their own account, their own network, and their own boundaries. Engineers can experiment in sandbox without any path to customer data. Changes prove themselves in staging before they reach production.
All three are defined in Terraform from the same shared modules for networking, compute, and container registry, so staging is a genuine rehearsal for production rather than an approximation of it. Nothing in any environment exists because someone once clicked a button, which means any environment can be reviewed, repeated, or rebuilt.
Delivery that runs itself
Every merge triggers a deployment through GitHub Actions. The pipeline assumes a temporary AWS role instead of holding any stored keys or passwords, so there are no long-lived credentials to leak or rotate. Container images are locked once published and scanned continuously for vulnerabilities, and every deployed version is tagged so rolling back a bad release is a single, known operation rather than an incident with an audience.
Because infrastructure and application code live in different repositories, the pipeline dispatches Terraform changes across repository boundaries, so a platform change follows the same reviewed, automated path as an application change.
Compliance that is continuous, not periodic
Governance is enforced at the organisation level rather than configured per project, which means it applies to accounts that do not exist yet.
Four preventive policies sit above every account. Region lock makes it impossible to create a workload outside Frankfurt. Another prevents the activity trail from being switched off, whoever is asking. These are not alerts that fire after the fact; the action simply fails.
Underneath that, AWS Config records every configuration change across all six accounts and aggregates the results into the audit account. Organisation-level CloudTrail delivers a complete API history to the isolated log archive account. GuardDuty watches for threats, Inspector scans container images continuously, and VPC Flow Logs capture network activity. AWS Security Hub brings the findings from all of it into one prioritised view, so the security posture of six accounts is a single dashboard rather than six investigations.
Nine containerised services now run in production on Amazon ECS with AWS Fargate, five of them AI-enabled, with scheduled and compute-heavy data processing handled by AWS Batch so it never competes with the platform customers are using.
What changed
TolaData ships on every change instead of on a schedule set by human availability, and release ownership moved from individual engineers to cross-functional delivery teams, so no single person is a bottleneck or a risk.
Environments are genuinely isolated, so the cost of a mistake in testing is a mistake in testing. Compliance evidence is produced continuously as a by-product of running the platform: complete configuration history, a full API trail in an account nobody can edit, and one place to look for security findings across the organisation.
Data stayed in Germany before and stays in Germany now, but it is no longer a matter of which provider the servers happen to sit with — creating a workload anywhere else is blocked outright.
An independent penetration test validated the production security posture after migration, and the AWS Well-Architected Framework Review closed with zero high-risk issues across all six pillars.
AWS Control Tower · AWS Organizations · AWS Security Hub · AWS Config · AWS CloudTrail · Amazon GuardDuty · Amazon Inspector · Amazon ECS · AWS Fargate · AWS Batch · Amazon ECR · Amazon CloudWatch · AWS Systems Manager · Amazon VPC · Elastic Load Balancing · Amazon S3 · AWS IAM Identity Center



