How DNS works: a beginner's guide

We sometimes get emails from people who are trying to point their custom domain at PythonAnywhere so that they can host their website, but are struggling to set up their DNS settings. Normally DNS setup is pretty simple, but sometimes people can get bogged down due to confusing interfaces on their registrar's site, or complexities in the terminology people use.

The parts of DNS that you need to know about in order to host a website are actually not all that complicated, but some domain registrars have complicated, hard-to-understand interfaces. Either they assume that you understand all of the technical details about how the whole thing works -- which makes it hard for first-timers -- or they try to put a simple user-friendly interface on top of it, but simplify it so much that it's actually harder to use because they're hiding important stuff from you.

Given that basic DNS stuff really isn't all that hard, we felt that it would be a good idea to post an explanation, going from the basics up to some slightly deeper stuff. This post is written so that if you only want the basics, you can just read the first part, while if you want a deeper understanding -- either out of interest, or because your domain registrar has got such a low-level interface that you need to -- then you can keep reading.

It's worth noting that for most people, you don't need to know any of this stuff to set up a website on PythonAnywhere, even with a custom domain; it's meant more as an explanation so that people who do run into problems with their registrar have the background knowledge they need to solve the problem -- or, indeed, to explain to the registrar's tech support team what the problem is. And, of course, it's a bit of light reading for people who are just interested in this stuff :-)

The basics: domain names and IP addresses

When a browser wants to connect to, it needs to know which computer on the Internet is hosting that site. The string is a hostname (technically it's a "fully qualified domain name"), but at the underlying network layer, all computers are identified by IP addresses, which are numerical. So somehow the browser needs to find out which numerical IP address it should use when it wants to talk to

It does that by asking a DNS server -- in full, a Domain Name System server. Normally this will be a server that's provided by your ISP or your local network administator; in general, when your computer joins a network, the details of the DNS server are part of the information it gets from the router. So to convert a hostname to an IP address, the browser makes a system call to the operating system saying "please get me the IP address for", the operating system sends a message to the DNS server with the same request, and the DNS server responds with the IP address: something like "" (in IP version 4, the version that's currently most widely used, IP addresses are normally written as four numbers between 0 and 255, separated by dots -- I'll use similar addresses as examples later on). Now the browser can connect to the server at that IP address, send it an HTTP message saying "please send me the contents of the front page for", and the server will respond with the appropriate stuff.

What all this means is that when you want your custom domain to be hosted on PythonAnywhere, you're essentially setting things up so that all of the DNS servers across the Internet know what IP address to provide when someone wants to access your site. (This, by the way, is why we say that the changes you make "may take some time to propagate across the Internet" -- it's not just one central database that needs to be updated; instead, different ISPs will pick up the change at different times. A little more about that later.)

DNS records

The explanation above is, of course, a bit simplified. The DNS database isn't just a mapping from hostnames like to IP addresses. Instead, for each "domain name", it can keep a bunch of different types of records. I put "domain name" in quotes back there because the technical meaning of the words here is slightly different to what people normally use it to mean. When we normally use the phrase "domain name", we mean something like "", "", or "". That is, we mean the portion of a network address that identifies a particular organisation. We'd expect (for example) Google to own the domain name, to have a website at, and for people who work there to have email addresses like

But in the technical parlance of DNS, a "domain name" is something a little more subtle. The Wikipedia article is (of course) a good explanation, but a summary is:

  • There is a "root" domain called .
  • There is a subdomain of the root domain called com.
  • If you buy then you own, which is is a subdomain of com.
  • If you create a website at then technically you've created a subdomain of with the name

(You'll notice that all of the domain names above end with the . to represent the root domain. Technically domain names always should, but no-one ever bothers, so I won't use the extra . in the rest of this article.)

Basically, any level of the hierarchy is a domain name. A number of different kinds of "records" can be associated with a domain name. The domain might have various records, and so might Any domain can have a number of different records, including multiple records of the same type.

Some examples of record types are:

  • A ("Address") records, which are how you specify an IPv4 address like These tell DNS that if someone wants to treat the domain name as the name of a computer, then this is the IP address they should use.
  • MX ("Mail eXchange") records, which specify what email server should be used for a domain. If someone wants to send email to, the email systems involved will use the MX record for to work out which computer to deliver it to.
  • TXT ("TeXT") records, which are pretty much free-form text and are used for various purposes including verifying senders of email messages.

There are a bunch of others -- NS, SOA, and so on. But apart from A records (which you can see are obviously important in the example above where we got the IP address for a hostname), there's one other interesting kind of record: CNAME.


When a browser (via its operating system) asks its DNS server for the IP address corresponding to a hostname, the server can respond in a number of different ways:

  • If it doesn't know what the IP address is, it will just return a response saying so, so that the browser can display an appropriate error page. (For example Chrome will say something like "’s server IP address could not be found.")
  • If it has an A record for the hostname, then it will just return the record and the browser will use the IP address contained in it.
  • If it doesn't have an A record, but does have a CNAME (abbreviated from "canonical name") record then it will return that. The CNAME record doesn't contain an IP address; instead, it contains another hostname. The browser can then send the DNS server a second message, saying "OK, then -- so what's the IP address of that hostname" -- and hopefully the response this time will have an IP address. (Of course, it could just respond with another CNAME record, which would require a third lookup, and so on.)

CNAME records, on the face of it, look like a somewhat roundabout and inefficient way to do things. Instead of making one request to get from a hostname to an IP address, a browser will need to make two or more. So why is it that when you create a website with a custom domain on PythonAnywhere, we ask you to set up a CNAME record to point to a hostname like, rather than just providing you with an IP address so that you can create an A record?

The answer is that CNAMEs are actually really useful -- and also not all that inefficient in practice.

The usefulness first: let's imagine you had a website hosted at, and you set up an A record with your registrar to point it at an IP address that belonged to PythonAnywhere. Your site would be up and running, and everything would work fine so long as that IP address was always the right one for your site. But if it changed, you'd need to log in to your registrar again and update it.

But IP addresses sometimes have to change. Sometimes they get blocked in certain countries. Sometimes a specific IP address might be subjected to a denial-of-service attack, and be unusable. Or sometimes we at PythonAnywhere might want to move your site from one IP address to another simply to balance out load across our cluster of servers.

If you're using a CNAME record, then IP address changes like that are something you don't need to worry about. We can update the A records for our own hostnames, like, because we control the DNS settings for and all of its subdomains. Because your website is pointed at us using a CNAME, browsers that want to connect to your site will start using the new IP address without you needing to do anything. They'll ask for the IP address of, they'll get a CNAME response saying that it's the same as the one for, so they'll look up the IP address for that and will get the correct new address from the A record that comes back.

But we can't update DNS records for your domain -- only you can do that -- so if you use a A record, when the IP address changes you'll have to update it yourself every time. If you happened to be away, or if we don't have up-to-date contact details for you, it might be some time before you knew about the problem and were able to fix it.

So what about the inefficiency? We always make sure that the addresses point to an A record rather than another CNAME, but that's still, in theory, two lookups to go from to an IP address -- one to get the CNAME, and one to get the IP address from the hostname stored there. According to Cloudflare, the average ISP has a 70 millisecond round-trip time for queries, so that might mean that your site would load up 70 milliseconds slower if it needs to do two queries rather than one. Not a huge amount of time, but given that research shows that people are less engaged with slower websites, every millisecond counts.

The answer here is twofold -- firstly, your computer will generally cache the results of DNS lookups for some time. So this extra time will only impact some hits to your page. Secondly, many DNS servers are pretty smart -- if someone asks for a hostname, and the result is going to be a CNAME record, they know that the next request is likely to be for the CNAME's value -- so they'll attach the results for that to their response as well. So, for example, a browser might say "what's the address for" and the DNS server would reply " has a CNAME pointing to Oh, and by the way, has an A record pointing to the IP address". The browser can then just use the IP address directly; there's only one lookup, but the CNAME is fully resolved.

So using CNAME records to point your website at PythonAnywhere means less maintenance for you, and in general no real performance overhead.

There is one case where CNAMEs can be problematic, though:

Naked domains

One problem with CNAMEs is that in general, you can't use them for "naked" domains. A naked domain is the domain name that you buy from your registrar, something like -- without anything like www. in front of it. For relatively arcane technical reasons a naked domain can't use a CNAME, only subdomains like -- and so if you want without the www. to point to PythonAnywhere, you have to use an A record.

That's why we suggest that if you want your site to be accessible without the www., you should host it at, and then use a redirection service so that people who visit the address without the www. will be transparently redirected to the address with it. Most registrars have this kind of thing built in. You tell them to do the redirection, and they'll set up an A record pointing to one of their servers, and that server will do the redirect when it receives a request -- and because, unlike us, they can update A records on your behalf, they can fix things if it needs to change. If your registrar doesn't support it, there are third-party services you can point to with an A record you set up yourself.

See our help pages for details of how to set up a redirect -- we have links to the appropriate documentation for a bunch of popular domain registrars, and also some links to third-party services if you need to use them.

Some DNS providers support what they call "ALIAS" records. These are not part of the DNS standard, but the way they work is that you create a record that looks a bit like a CNAME -- it maps from a domain name of yours to a hostname. The way they work internally is that when a client requests the domain, the DNS provider just does the lookup for the hostname that the domain name points to internally, and returns what looks like an A record -- kind of like the cached CNAME example above. So, for example, a browser might say "what's the address for" and the DNS server would see that it had an ALIAS record for CNAME pointing to, so it would look up the IP address for and would reply (bending the truth a little that " has an A record pointing to the IP address".

Their advantage over CNAME records is that they can be created for naked domains, so you can use one to host a site at on PythonAnywhere without needing to use an A record.

However, because they are not a part of the DNS standard, we don't officially support them. Still, they will in general work, and they're better than using an A record.

How to use all of that information

Hopefully by now it should be clear what you need to do when pointing a custom domain to PythonAnywhere; you log into your domain registrar's website, and find the place where you set up your DNS configuration. Once you're there, you set up a CNAME record to point to the value shown when you look at your website's configuration on the "Web" page inside PythonAnywhere. If there are any other A or CNAME records for then you should delete them (because having two records for the same hostname will potentially confuse the DNS). But don't delete any other records -- there may be things like MX records, which you'll remember are used for email, or "NS" or "SOA" records, which are low-level DNS stuff that you shouldn't touch without knowing pretty clearly what you're doing.

Different registrars have different interfaces, but with most of them it's reasonably easy to find the bit you need once you know what you're looking for. We also have some links to the appropriate documentation on the sites of popular registrars on our help pages.

One thing that does sometimes confuse people is that many registrars only require you to type in the bit that goes before your domain name when specifying a record -- that is, for you would set up a CNAME with a "name" of www and a "value" of If you created one with a "name" of, then the CNAME would actually point the hostname at PythonAnywhere, which would be unhelpful :-)

Sometimes people are told by their registrar that instead of setting up a CNAME, they need to provide a name server (or even two name servers). Oddly, this sometimes happens even when the registrar in question really does support CNAMEs. Without wanting to call any registrar out in particular, if you are told that, but your registrar appears on the list on our help pages then the customer service person you're talking to is a little confused, and you should check out the documentation that we link to.

But some registrars really do not allow you to set up CNAMEs; they require you to specify name servers. This can happen, for example, with some country-level domains. Explaining what that means is what we'll move on to next.

Domain registrars versus DNS and name servers

Providing the ability to register domains, and providing the DNS stuff for those domains are, technically speaking, two different services. Given that in general it's pretty pointless to own a domain without being able to associate it with an IP address so that you can host a website there, most companies that do domain registration provide both registration and DNS services as a bundle, so you don't need to know about the separation -- but it is there, and if you find yourself having to use a registration-only service, you need to know a bit more about how things work.

The first thing to explain is what a name server is. Remember that earlier we said that when a browser wants to connect to, it asks its local DNS server -- the one provided by the ISP -- what the IP address is. Obviously, the DNS server needs to find the answer somehow. It might have cached the DNS records for the hostname you're looking for due to some previous request, but if this is the first time it's heard about this hostname, it will have to pass the query on to a different server. This is the "authoritative nameserver" for the domain; different domains will have different nameservers set up for them. In a normal setup, when you buy, there will have to be one or more computers somewhere on the Internet who, when asked for the details of, can give the official answer -- a CNAME, an A record, or whatever. So if the DNS server belonging to your ISP doesn't know the IP address for, it needs to ask the nameserver that is authoritative for for the answer.

You might wonder how your ISP's DNS server knows how to find out what those nameservers are -- how do they know which nameserver is authoritative for The answer is that they ask the nameservers that are authoritative for "one level up" -- that is, if they want to know which nameservers are authoritative for, they'll ask the nameservers that are responsible for .com to tell them. Likewise, if asked "which nameservers are authoritative for", they would ask the nameservers that are authoritative for, which might first require them to ask who is authoritative for .uk in order to make that query. (You might in turn wonder how the whole thing gets bootstrapped -- how the DNS server works out the nameservers that are authoritative for these top-level domains like .com and .uk. The answer is that there are some special servers that handle that -- that is, servers that are right at the top of the tree, and are authoritative for the . "root" domain we mentioned a while back.)

So -- when you register a domain name with a normal "bundled" registrar like GoDaddy, they will do three things:

  • Register you as the owner of the domain with the organisation who is responsible for keeping track of such things (which will be a different one for each top-level domain -- Verisign for .com, Nominet UK for .uk, and so on)
  • Set up one or more of their own nameservers so that they can give authoritative answers to questions about your domain.
  • Tell the authoritative nameservers one level up from your domain (that is, the .com ones for that their -- that is, the registrar's -- nameservers are authoritative for your domain.

Once that's done, you're all set -- you can use their web interface to update your DNS settings, and when you do that they'll pass the changes on to the appropriate nameservers.

However, if you are using a registrar which doesn't do DNS for you, they won't do the last two steps. They'll just register the domain. They will, however, have a way to tell the one-level-up nameservers which nameservers are authoritative for your domain -- but you'll need to provide them with the names of some servers in order for them to do that.

Setting up your own nameserver is really tricky, but luckily you don't need to do that. There are free services out there that can handle all of it for you, so you just need to sign up with one of them. They'll provide you with the addresses of some nameservers that you can then pass on to your registrar, and a web-based interface so that you can set up all of your DNS records like the CNAME and so on. One that consistently get good reports from our users is the FreeDNS service from Namecheap.


Hopefully that was all reasonably clear. You now know the basics of how DNS works, and how it interacts with domain registration. If you've got any questions, please drop us a line at -- and additionally, if you think there are other things that you'd like us to add to this article, please do let us know!