Kubernetes upgrade without downtime

We upgrade your cluster (AKS or EKS) to a new version without an outage of your services. We start from the assessment, so we know the blocking issues and fix them in advance.

from CZK 145,000

2 to 3 weeks, price per 1 cluster and 1 version, each extra cluster +CZK 75,000. Excl. VAT.

Book a consultation

What we offer

We upgrade a non-production cluster first. Production is then upgraded in waves, in a window you choose within working hours. After every wave we check the applications, and we keep the original nodes ready until you confirm the upgrade.

The Kubernetes control plane cannot be rolled back to the previous version. That is why we put the weight on preparation and the non-production test. Nodes can go back to the original node pool, and if a problem appears after the control plane upgrade, we fix it forward.

We do not write applications. We work only on infrastructure, cloud and DevOps.

When it makes sense

Your cluster version is close to end of support, or you already pay for extended support.
You keep postponing the upgrade because of the risk of an outage.
Your internal team has no upgrade experience or no time for it.

What you get

A cluster on a supported version with no service outage.
A proven procedure you can reuse for the next upgrade.
An updated runbook and recommendations for the next upgrade.

Why a Kubernetes upgrade cannot wait →

What exactly we do

Preparation: Fixing the blocking issues from the assessment in the infrastructure (add-ons, deprecated APIs in manifests and Helm charts), adding PodDisruptionBudgets and probes together with your developers, verifying backups with a restore test.
Non-production test: Upgrading the non-production cluster with the same procedure and verifying it with your tests.
Production upgrade: Control plane upgrade, a new node pool and applications moved over in waves, with a check after every wave.
Validation and close: Monitoring after the upgrade, removing the original node pool once you confirm, updating the runbook.
One version at a time: The price is for 1 cluster and an upgrade by 1 minor version. Kubernetes is always upgraded one minor version at a time. If you are several versions behind, we price the further steps from the assessment.

Who works on it

Three people work on the upgrade. It is led by a senior who decides every step in production.

Senior, project lead
4.25 days

Responsible for
Upgrade plan, go/no-go criteria, the decision on every wave, communication with you during the window, problem solving.

What you get from them
Main contact, go/no-go decisions
Middle
9 days

Responsible for
Fixing blocking issues, upgrading non-production and production, add-ons and node pools.

What you get from them
The upgrade carried out
Junior
7.5 days

Responsible for
Preparation, tests after every wave, monitoring during the window, report and runbook.

What you get from them
Tests, report, runbook

We assign work by complexity: juniors do routine work from templates, the middle takes the harder parts, and the senior owns the design, quality control and risk decisions. You do not need to decide who does what. The effort in days is indicative and only explains the price. The price is fixed and does not depend on the days actually worked.

How the project runs

A typical upgrade takes 2 to 3 weeks, of which 8 to 14 days are work, depending on the number of applications and the assessment findings. We go to production only after a successful non-production test and your approval.

Week 1: preparation
Goal: Remove everything that could put the upgrade at risk.
- Fixing the blocking issues from the assessment (add-ons, deprecated APIs in the infrastructure).
- Adding PodDisruptionBudgets and probes together with your developers.
- A restore test from backup.
- Checking quotas for the new nodes.
- Window plan, communication and go/no-go criteria.
Output: A cluster ready for the upgrade, an approved plan. Who: middle on fixes, junior on backups and quotas, senior on the plan
Weeks 1 and 2: non-production
Goal: Prove the procedure with no risk to production.
- Upgrading the non-production cluster with the same procedure as production.
- Verifying the applications with your tests.
- Adjusting the procedure to what we find.
Output: A proven procedure. Who: middle on the upgrade, junior on tests
Week 2: production
Goal: Upgrade production with no outage.
- The go/no-go decision with you.
- Control plane upgrade.
- A new node pool and applications moved over in waves.
- After every wave a check of pods, errors, latency and alerts.
- The original node pool stays ready for a way back.
Output: Production on the new version. Who: middle carries it out, senior decides, junior monitors
Week 3: validation and close
Goal: Confirm stability and hand over.
- Monitoring operations for 2 to 3 working days.
- Removing the original node pool once you confirm.
- Upgrade report, updated runbook, recommendations for the next upgrade.
Output: Upgrade closed and documentation handed over. Who: junior on the report, senior on recommendations

How we work together

Main contact: The project lead (a senior). Accountable for the result, deadlines, risks and decisions. Day to day you can talk to anyone on the team directly.
Communication: A shared Slack or Teams channel and a ticketing system of your choice. Every question and change leaves a trail.
Kick-off: At the start we confirm the scope, access, contacts and the dates of the key milestones.
Weekly status: A short 15 to 30 minute sync: what is done, what is next, what we need from you and which risks we see.
Final readout: We present the results to your team and management and go through the next steps.
Escalation: Junior → middle → project lead. If a deadline or a major risk is at stake, the project lead contacts you the same day.
Working hours: Working days, 8:00 to 17:00 Central European Time. Production changes happen in a window you choose, always through your change management.

What we need from you

The timeline holds if we have the following. If something is delayed, the schedule moves by the same time and we tell you straight away.

A completed assessment (ours or equivalent).
Cluster-admin rights and cloud access for the duration of the project.
A non-production cluster. If you do not have one, we agree on a temporary test cluster whose cloud costs are yours.
Tests or smoke tests to verify the applications.
A window within working hours and approval in your change management.
Application owners available during the window.

Access and security

Every team member has their own named account. We do not use shared accounts.
We ask only for the permissions a phase needs. Where read access is enough, we work read-only.
We do not take data out of your environment. Our documents contain only the summaries the result needs.
When the project ends, you remove our access, or we remove it on request and confirm it.
If you wish, we sign an NDA before the kick-off.

What is included and what is not

Included in the price

Preparation and fixing of blocking issues in the infrastructure
Upgrade of non-production and production by 1 minor version
Upgrade in waves with a check after every wave
Report, updated runbook, recommendations

Not included: priced separately or on your side

Code fixes for applications incompatible with the new version (we describe them, your developers make them)
Further clusters (+CZK 75,000 each)
An upgrade by several versions (priced from the assessment)
Night and weekend windows, 24/7 on-call

If we find during the project that reality differs from what we assumed, we tell you before we work outside the scope. We send a short estimate of the impact on price and timeline and continue only once you agree.

Deliverables and acceptance

Deliverable	Contents	Form
Upgraded cluster	Control plane and all node pools on the target version	In your environment
Upgrade report	Progress by wave, check results, deviations	PDF
Runbook	Updated upgrade procedure	Markdown in the repository
Recommendations	What to prepare for the next upgrade	Part of the report

The project is done when:

The control plane and all node pools are on the target version.
All applications are running and Ready.
Operations run for 2 working days after the upgrade without an incident caused by the upgrade.
We have handed over the report and the runbook.

We hand over the deliverables and go through them with you. We incorporate comments in one round within 5 working days of receiving them.

Risks and how we handle them

Risk	How we handle it
An application has no replicas or PDB	We add them before the upgrade, or agree a short window with the application owner.
Deprecated APIs in application charts	We flag them in the assessment. Your developers fix them before the upgrade.
Not enough quota for the new nodes	We request a quota increase during preparation.
A problem after the control plane upgrade	The control plane cannot be rolled back, so we fix forward. That is why we always test on non-production.
An application problem on the new nodes	We move the applications back to the original node pool, which we keep until you confirm.

Price and terms

from CZK 145,000 excl. VAT for 1 cluster and an upgrade by 1 version, each extra cluster +CZK 75,000.

Duration: 2 to 3 weeks.
The work happens within working hours. You choose the production window.
Prices exclude VAT, invoiced in CZK or EUR.
Work outside the scope: a separate quote, always in advance.

What can follow

Next steps

1A short 30 minute call: we look at your environment and check that the service fits what you need.
2We confirm the scope, the start date and the contact people.
3We send the contract and the list of access we need to start.

Frequently asked questions

Do we need an assessment?

Yes, the upgrade builds on it. If you do not have one, we start with it.

Why only within working hours?

During working hours everyone is available, application owners included, and any problem is solved fastest.

What if we are several versions behind?

We upgrade step by step, one version at a time. We price the further steps from the assessment.

Tell us what you are working on.

Thirty minutes, free. We reply within one working day.