Compute gets the attention, but a large share of user-visible incidents come from the three layers in front of it: name resolution, certificates and caching. None is complicated; each has a characteristic way of going wrong.
DNS: the TTL decides your options
Time-to-live controls how long resolvers cache a record, and therefore how quickly you can move traffic. A long TTL is efficient and slow to change; a short one is flexible and slightly chattier. The practical approach is to lower the TTL a day before a planned migration and raise it again afterwards.
- Keep TTLs moderate normally; lower them ahead of planned changes.
- Remember that some resolvers ignore very low TTLs.
- Watch registrar expiry dates: an expired domain is a total outage.
- Keep DNS configuration in code alongside the rest of your infrastructure.
- Use health-checked records if you need automatic failover.
TLS: automate renewal or it will lapse
Certificate expiry is among the most predictable outages there is, which makes it the most embarrassing. Use automated issuance and renewal, and monitor the expiry date independently of the renewal system so that a silent failure is still noticed.
- Automate issuance and renewal.
- Alert at thirty days remaining, independently of the renewal automation.
- Terminate TLS in one place – the load balancer or proxy – not on every instance.
- Redirect plain HTTP to HTTPS and enable strict transport security once stable.
- Include every hostname you serve, including the one nobody remembers.
CDNs help some things and not others
| Content | Benefit |
|---|---|
| Static assets | Large: served from an edge near the user. |
| Cacheable API responses | Real, if you set cache headers thoughtfully. |
| Personalised pages | Little, unless you cache fragments carefully. |
| Write requests | None; they travel to the origin regardless. |
| Origin availability | Partial: stale-while-revalidate can hide brief outages. |
Summary
Manage DNS as code with TTLs you have thought about, automate certificates and monitor expiry separately, and use a CDN where content is genuinely cacheable. These three layers are where a surprising number of outages begin.
