Lesson 3 of 8 in Networking for DevOps

DNS: how names become addresses

Core9 min

DNS is the distributed system that turns names such as example.com into addresses and other records. A resolver asks a chain of servers for the answer on your behalf and caches it for a time set by the record's TTL.

Draft. This lesson is written and waiting for a person to run every command and review it. It is safe to learn from, but check the official documentation for anything you depend on.

Read these first: IP addresses, CIDR and subnets

The problem this solves

People and configuration files use names, while the network only moves packets between addresses. Hard-coding addresses breaks the moment a server moves, so every service needs a lookup layer.

DNS is also the first thing that fails and the last thing people check. A wrong record, a stale cache or a missing entry produces errors that look like the service is down.

How it works

How a recursive resolver finds an answer
How a recursive resolver finds an answerYour app asks a resolver, and the resolver asks the root, the top-level domain and the authoritative servers in turn before returning the answer.1 ask, 5answer back2 who runs .com?3 who runsexample?4 what is the address?Your appwants example.comResolvercaches answersRoot serversknow who runs .com.com serversknow who runs exampleAuthoritativeholds the real record
  • People and clients
  • Network
  • Control
  • Data

Your app asks a resolver, and the resolver asks the root, the top-level domain and the authoritative servers in turn before returning the answer.

  1. Your app to Resolver: 1 ask, 5 answer back
  2. Resolver to Root servers: 2 who runs .com?
  3. Resolver to .com servers: 3 who runs example?
  4. Resolver to Authoritative: 4 what is the address?

Your program does not talk to the root servers. It asks a recursive resolver, usually one run by your network, your cloud or a public provider. If the resolver already has an unexpired answer, it replies from its cache. Otherwise it does the walk in the diagram.

Each step narrows the question. The root servers say which servers handle .com. The .com servers say which servers are authoritative for example.com. The authoritative servers hold the actual records and give the final answer, which the resolver caches and passes back.

Common record types are A (name to IPv4 address), AAAA (IPv6), CNAME (this name is an alias for another name), MX (mail), TXT (free text, used for verification) and NS (which servers are authoritative). Every record has a TTL, the number of seconds a cache may keep it.

Before DNS is asked, the system normally checks local sources such as /etc/hosts. So the address your application sees can differ from the one dig shows, because dig queries DNS directly and skips those local sources.

Try it yourself

These commands only read DNS. dig ships in the dnsutils or bind-utils package on Linux and is present on macOS.

  1. 1
    See which resolver your machine uses
    cat /etc/resolv.conf

    A nameserver line names the resolver. On many Ubuntu systems it shows a local address such as 127.0.0.53, which is a local forwarder. On macOS use scutil --dns.

  2. 2
    Look up a record
    dig example.com A +noall +answer

    The answer line has the name, the TTL in seconds, IN, the record type and the address. Run it twice and watch the TTL count down when your resolver serves a cached answer.

  3. 3
    Watch the whole walk from the root
    dig +trace example.com

    This performs the lookup itself, step by step. You will see the root servers, then the .com servers, then the servers for the domain.

  4. 4
    Trigger a name that does not exist
    dig this-name-does-not-exist.invalid

    Find the header line with status: NXDOMAIN, which means the name does not exist. The .invalid ending is reserved and never resolves.

  5. 5
    Compare two resolvers
    dig @1.1.1.1 example.com +short
    dig example.com +short

    @1.1.1.1 sends the query to Cloudflare's public resolver instead of your default. Differing answers point at caching or split-horizon DNS.

  6. 6
    See what applications see
    getent hosts example.com

    getent uses the same lookup path as most programs, including /etc/hosts. On macOS use dscacheutil -q host -a name example.com.

What breaks, and where to look

  • curl: (6) Could not resolve host: api.example.internal, or dig shows status: NXDOMAIN.

    WHY IT HAPPENS
    The name has no record in the zone the resolver consults. Common reasons are a typo, a record never created, or a private name asked of a public resolver that cannot see it.
    CHECK FIRST
    Run dig <name> and read the status line. Then run dig @<authoritative-or-internal-resolver> <name> to see whether the record exists at the source.
  • You changed a record and some machines still go to the old address.

    WHY IT HAPPENS
    Resolvers cache the old answer until its TTL runs out. A long TTL set before the change means the old value can linger for that long.
    CHECK FIRST
    Run dig <name> and read the TTL in the answer. Lower the TTL well before a planned move, then change the record after the old TTL has passed.
  • dig returns the right address and your application connects somewhere else, or fails.

    WHY IT HAPPENS
    The application uses the system lookup path, which can include /etc/hosts, search domains or a different resolver, and dig bypasses them.
    CHECK FIRST
    Compare dig <name> +short with getent hosts <name>, then read /etc/hosts and /etc/resolv.conf.

When not to use this

  • DNS is a poor tool for instant failover. Caches obey the TTL in their own time, and some clients ignore it, so use a load balancer or health-checked routing when a switch must happen at once.
  • /etc/hosts is fine for a quick local test and a hazard in production, because the override is invisible to everyone else and easy to forget.

Check yourself

Why does your app ask a resolver instead of the root servers?

The resolver does the multi-step lookup for you and caches results, so most lookups are answered without repeating the walk.

What does the TTL on a record control?

How many seconds a resolver may reuse the answer before asking again.

What does status: NXDOMAIN mean?

The queried name does not exist in DNS.

Lesson 18 of 92 in the DevOps path

Search

Search cheat sheets and projects