Home
›
Articles
›

Moving your app onto AWS without a weekend of downtime

Moving your app onto AWS without a weekend of downtime

By Ibi Hasanli

·

·

4 min read

Moving your app onto AWS without a weekend of downtime is a planned cutover, not a hope that Saturday night will be enough. You build a working copy on AWS first, switch DNS in a short window once the data is in sync, and keep a way back. Lower the DNS time to live ahead of the switch, then wait, because resolvers keep the old answer until the TTL they already have expires.

Build a staging copy on AWS first

The new environment has to take real traffic in a test, not only look right in a screenshot. Build it in the customer's AWS account, in a UK or EU region unless there is a reason not to, and prove the application, certificates and backups there before anyone's public DNS changes.

Eight Mile's migration work follows that order: mapping what you have, rebuilding the setup as code, and proving the replacement before switching off the original. The new platform runs beside the old one, so a rollback stays possible. If you do not hold the AWS login, start with leaving a freelancer with your AWS root login. If the application was inherited, see what to check before you inherit a custom app. An unpatched box is a separate risk, covered in the EC2 server nobody has patched since launch.

Move in this order 1. Staging Build a copy on AWS and prove it first 2. Lower TTL Shorten the TTL then wait for the old value to expire 3. Cut over Sync the data, check the new side, switch DNS 4. Way back Keep the old host live so you can switch back Do not raise the TTL again until the new host has been stable.
Build the AWS copy first, lower the DNS TTL and wait, then switch. Keep the old host live so you can switch back.

A planned DNS switch-over

DNS is how visitors find the new host. A record's TTL, in RFC 1035, is how long that answer may be cached before the source should be consulted again. Amazon's Route 53 best practices recommend temporarily shortening TTLs before you change a critical record, so you can switch, watch and roll back quickly, then raise the TTL once the new settings are stable. For a domain already in use, the Route 53 record guidance suggests a shorter value first, such as 300 seconds.

Do that ahead of the cutover, then wait. Route 53's troubleshooting page is plain: changing the TTL now will not make the change take effect faster, because resolvers already have the previous value cached, and they will not ask again until that old TTL has passed. Route 53 says updates to its own name servers typically show up in less than a minute, but that does not clear caches on the internet.

While you wait, get the data across and keep it current. AWS Database Migration Service can do a full load and then ongoing replication so the target stays in step with the source. AWS is also clear, in its notes on change data capture, that this is not real-time replication, that there is no latency SLA, and that lag can run to several minutes or longer. When the copy is close, those docs describe stopping the application, letting remaining changes apply, then pointing the app at the new database. A backup restore test belongs in the same plan. Before you change the record, check the site, logins, payments and mail on the new side. If you do not control the domain, see who actually owns your business domain name.

The DNS switch, step by step 1. Lower the TTL On the record you will change, ahead of time 2. Wait Until the old TTL has expired in caches 3. Switch Point DNS at the new host; keep the old one live 4. Raise the TTL Only after the new host has been stable Changing the TTL at the last moment does not speed up the switch.
Shorten the TTL before the cutover and wait for the old value to expire. Changing it at the last moment does not speed things up.

A way back

Keep the old host running through the window, and after it, until you are sure. If the new side fails, switch the DNS record back. A short TTL, already in place, is what makes that rollback useful. Do not raise the TTL again until the new host has been stable.

How Eight Mile runs a move

Our infrastructure and cloud work includes designing, building and running infrastructure on AWS, and moving workloads onto AWS. Customers keep the accounts. We work with access to them. A review establishes the plan and the price. Billing for scoped work starts only once a written scope and price are agreed. Delivered work is covered for twelve months. After that, care is a separately agreed retainer, as in what managed AWS hosting includes and why your AWS bill and your management fee are two different bills.

Get the cutover terms in writing before you start.

  • The window: when the switch happens, and who approves it.
  • Data sync: how the copy is made and kept current, and when writes stop on the old side.
  • The old host: how long it stays live after the switch.
  • Rollback: what triggers a switch back, and who can do it.
  • Hours: any out-of-hours cover for the window, written into the arrangement. Ordinary enquiries are handled during London business hours, Monday to Friday.

The useful test is whether you can name the staging URL, the TTL you lowered and when it expires, and the step that puts DNS back. If moving your app onto AWS without a weekend of downtime still sounds like a long outage, the plan is not finished.

If you want that plan agreed on an AWS account you own, talk to an engineer. Tell us what is running today and when you need it moved.

AWS
Infrastructure

Leaving a freelancer with your AWS root login

Leaving a freelancer with your AWS root login is a risk even after a good build. How to take back control with named IAM, MFA and a handoff.

·

4 min read

AWS
Infrastructure

The EC2 server nobody has patched since launch

The EC2 server nobody has patched since launch is a risk once the OS is out of support. What to check, and what managed patching should include.

·

4 min read

Infrastructure
Cloud

Cloud bill went up and nobody can say why

Cloud bill went up and nobody can say why: usually orphaned resources, missing tags, or one shared account. Fix ownership first.

·

3 min read

AWS
Infrastructure

AWS vs managed hosting for a small SaaS

AWS vs managed hosting for a small SaaS: when simple platforms fit, and when owned cloud, IAM and infra in git win.

·

3 min read

CRM
SaaS

CRM for a small firm off the shelf or built

CRM for a small firm off the shelf or built: when HubSpot-class tools fit, and when a system on your own job data serves trades and clinics better.

·

4 min read

Websites
Hosting

Uptime monitoring for a small business website

Uptime monitoring for a small business website only works with a named engineer and response path. What a care plan or managed hosting should cover.

·

3 min read

TradeOps
SME

When WhatsApp is running your jobs

When WhatsApp is running your jobs, groups and voice notes become the job board. What a proper ops system replaces, without taking phones away.

·

4 min read

Software
Customer service

The info@ inbox nobody owns

The info@ inbox nobody owns is where a UK SME's enquiries sit with no named owner and no reply time.

·

4 min read