Cloud and DevOps

Infrastructure defined in code, and the pipelines around it. Terraform across major providers, Kubernetes and Docker on top, and the operational layer that decides whether any of it still works at four in the morning. I have been doing this since the tools were called config management and ran over SSH.

01

What I cover

Infrastructure as code

Networks, compute, and storage in version control, so the environment can be rebuilt from nothing and the changes reviewed before they touch anything. Terraform across every major provider.

Containers and orchestration

Images that build the same way twice, and clusters configured from a manifest rather than from memory. Includes the parts that only matter once something has already gone wrong.

Configuration management

Servers brought to a defined state by running something, not by following a checklist. Existing bash and Ansible setups converted so they can be re-run without drift.

CI and CD

Builds, tests, image scanning, and deployment. Pipelines built to tell you what broke rather than that something broke, which is a small design decision with a large effect at three in the morning.

Backups

Backups that have been restored. A backup nobody has tested is a hypothesis, so a restore test is part of the job rather than a separate exercise nobody has time for.

Legacy migration

Moving running systems to code without downtime that the business notices. Bash scripts turned into Ansible roles, or a server rebuilt from a documented definition rather than copied from itself.

02

Work done in this area

Hetzner private cloud
Virtual private cloud set up with Terraform, including public access to the clusters through a load balancer, defined entirely in code.
Stack
Terraform, Hetzner, load balancing, networking
Docker Compose for Azure
Docker Compose files produced for an Azure deployment, with the client told plainly that they would need their own DevOps staff to implement the work.
Stack
Docker, Azure, Compose
Hardened images and dependencies
New Dockerfiles and package manifests produced to harden an Azure environment, with the open source dependencies declared explicitly.
Stack
Docker, Azure, Kubernetes, Terraform
Pipeline build and scan ordering
A pipeline that pushed and then scanned its image, fixed so the scan result gates what gets deployed.
Stack
CI, Docker, image scanning
Bash to Ansible
An existing backup script migrated to Ansible so it runs repeatedly without drift. Client reported the change as helpful and delivered on time.
Stack
Ansible, Bash, AWS, Redis
Jenkins for GitLab
Pipeline built and delivered on time. Asked for again by a client who came back for follow up work.
Stack
Jenkins, GitLab CI, shell
Google Compute Engine
Compute instance set up and supported for an ongoing engagement.
Stack
GCP, Python, cloud computing
Proxmox cluster
Highly redundant multi node cluster configured for failover.
Stack
Proxmox VE, Linux, shared storage
Deployment and hosting
Dynamic website deployed with DNS and hosting configured, delivered on time with quality the client did not expect for the budget.
Stack
DNS, web hosting, cloud computing, PHP
Google App Engine deployment
Reporting application deployed on Google App Engine and handed over with documentation.
Stack
Google App Engine, PHP, Python
03

Questions

Do you work in AWS, or should I be on Hetzner?

Both. Hetzner costs a fraction of AWS for equivalent work, and it comes up often with clients who did not know it was an option. What decides it is where your team already is and what is already built there, not which provider is fashionable this year.

Will you recommend Kubernetes?

No, not by default. Most of the systems I work on do not need it, and a single container on a small host is simpler to run, cheaper, and has fewer ways to fail at three in the morning. Kubernetes earns its cost once you have several services, several people deploying, and a real reason to schedule across more than one machine.

Can you take over infrastructure somebody else built?

Yes, and the first piece of work is usually smaller than people expect. Runbooks are frequently wrong rather than missing, which is easier to deal with, and infrastructure nobody has touched in two years is usually still working for a specific and documentable reason.

Is your backup a backup?

Ask whether anyone has restored from it. If the answer is no, then what you have is a process that writes files, and you do not know whether those files are complete or readable. Restoring one file takes minutes and settles the question. See Linux and systems for the host side.