This guide is an effort to summarize established best practices and strong recommendations for authors of DigitalOcean tutorials. It is intended to provide a foundation for consistency, technical correctness, and ease of use in DigitalOcean’s instructional material.
This guide is by nature both a work in progress and an opinionated document, based on the growing experience of in-house technical writers and editors, community managers, and engineers. Its recommendations are subject to change, and are written specifically for educational content with a broad range of readers and end-users in mind.
Many tutorials will rely on existing tutorials as prerequisites. Put all prerequisites for the article (including any nested prerequisites for the prerequisites) into the article, rather than having deeply nested prereq lists.
In rough, descending order of preference, use the following installation mechanisms:
curl | bashpatterns) require a judgement call on whether or not to use them.
curlinstall scripts piped to the shell, with an appropriate warning about inspecting scripts.
In general, avoid unnecessary complication. For unpackaged software installed from source or binaries, you should generally accept the default installation prefix unless it’s very unusual or introduces conflicts.
An init script, conforming to official recommendations for the distribution, should be given for service-oriented software, if not provided by the package or other installation method.
On Linux systems, put self-contained binaries or directories in
/opt and standalone scripts in
Ubuntu and Debian systems should have
unattended-upgrades with at least security updates installed and configured. We recommend no auto-reboot or auto-update all, given context.
We generally recommend using the
apt command for Ubuntu 18.04 and newer. Change
apt-cache to use
apt. When updating apt commands, do not use the
-y flag; readers should be guided through any necessary inputs and prompts.
For CentOS and Rocky Linux tutorials, we now recommend using
dnf, which has superseded
yum and provides better performance.
If a tutorial relies on the latest updates, call
upgrade during Step 1 setup or as needed. Call
update first so that your server pulls the latest versions of packages. When including
upgrade, which downloads and installs new versions for every package, please be aware that some users may choose to keep certain packages at a lower version.
Make sure to use native init system commands, even when legacy compatibility commands are available. For instance, use
sudo systemctl start [service_name] even though
sudo service [service_name] start will work.
Provide information about how to enable or disable the service from starting at boot. Indicate how to inspect outcome of service-related commands when not clearly indicated by the interface (
systemctl status, etc).
Prefer restarts over reloads for services as a rule of thumb. In most cases, it’s more important to ensure a known state than avoid a split-second service interruption, and restarts are also more useful in the case of a complete service failure.
Unless it’s part of a config management workflow, prefer user-data scripts, and prefer cloudinit scripts to bash scripts in user-data in most cases.
Explain where and how to access logs for installed services. Where relevant, explain
journalctl commands for checking service status and log output. Where possible, offer concise suggestions for diagnosing common failure cases.
Make sure to handle log rotation for any cases where it’s not handled by packages or other installation mechanisms.
For following plaintext log files, use
tail -F, not
tail -f, as the latter will not track a file across renames and might cause confusion if logs are rotated while a user is watching them.
sudo users instead of using root directly. Reference the appropriate initial server setup guides which explain this task as a prerequisite.
On Debian-based distributions, add and remove users with
adduser sammy and
deluser --remove-home sammy respectively; on RHEL-based distributions, use
adduser sammy (set a password with
passwd sammy if necessary) and
userdel -r sammy.
sudo privileges with
usermod -aG sudo sammy on Ubuntu. CentOS is a little more complicated. Modern versions use
usermod -aG wheel sammy, but some versions require
visudo to uncomment
wheel group permissions first. Specifically, on CentOS 5,
sudo needs to be installed and the wheel group needs to be uncommented with
visudo; on CentOS 6,
sudo is already installed, but wheel needs to be uncommented; CentOS 7 has
sudo and the wheel group is already set up.
When using privilege escalated commands, make sure to test them as written. To pass environment variables through
sudo -E command_to_run (if trusted with entire environment) or
sudo FOO=BAR command_to_run. For instances that require a root shell, use
sudo -i. For instances that require redirection, use
tee -a to append to rather than replace the destination file:
[sudo] command_to_run | sudo tee [-a] file_to_change.
For interactive shells, assume Bash on GNU/Linux systems, mentioned explicitly when relevant. On FreeBSD, use tcsh, as it’s available out of the box and has useful features.
For text editors, we include the copy “use [preferred] or your favorite text editor”, and include the following beginner-friendly editors in commands for those copy and pasting. On Linux, we default to
nano; on FreeBSD, we default to
ee. vi(m) is permissible, but avoid it in introductory topics where it might present a stumbling block for beginners.
For file transfer, we generally recommend
sftp in most cases for its interactive and scp-alike uses, though it lacks push functionality, so
scp is acceptable as well.
rsync is useful for backups and large transfers (or many small files). Do not use FTP under any circumstances. We also make an effort to standardize on
wget because of its robustness.
wget’s advantage is mostly recursive download (i.e. a special use case which is not common for our kind of content).
On machines that ship with
iproute2 utilities, we prefer them to the
net-tools suite, which is considered obsolete. In general,
iproute2 utilities like
ss will have better support for multiple interfaces, IPv6, new kernel functionality, etc. So likewise, we should use
ip route over
ip addr show over
ifconfig, etc. Sometimes the older utilities output is a bit cleaner by default, but the output itself is a bit less trustworthy since they don’t handle edge cases as well. When possible, we will control the more verbose output using available flags.
Within the context of systems administration tutorials, generally avoid lengthy custom scripts and long shell scripts.
Author-written scripts (and possibly other resources) should live in a per-article repository in the do-community GitHub account with a link back to the published tutorial. Follow good scripting practices in general. For example, put any variables the user will have to fill in at the top of the script, preferably in a well marked section. Provide in-line comments where needed to provide human-readable scripts. Ensure that your descriptions of the code are not exclusively provided in comments, but that you also provide prose descriptions with further explanation than appears in comments.
bash and avoid Bash-specific features when portability or cross-platform reuse are a concern. Use the shell and coreutils/standard Unix tools for small tasks; avoid introducing new dependencies purely for glue-language tasks unless the benefits are substantial. Prefer
#!/usr/bin/env interpreter to
In general, use
cron for scheduling recurring tasks, but systemd timers are also acceptable.
your_domain when possible instead of documentation IP ranges or
When downloading scripts or data, ensure that the user is in a writeable directory or paths are explicitly specified. For files which should be available for reference or reuse, use the user’s home directory, unless they belong in some standard well-defined path elsewhere on the filesystem (such as
/etc). For throwaway files, use
When a tutorial relies on a specific directory, be sure to provide the
cd commands that route the reader to that directory before they run commands.
We recommend the Debian-style configuration directories for distributions that don’t structure it that way by default. Always test configuration changes (Apache uses
sudo apachectl configtest, and Nginx uses
sudo nginx -t).
/var/www/html should be used as the document root for all web servers. Nginx’s
/usr/share/nginx/html default should be changed because that directory is owned by and can potentially be modified by package updates. This is no longer a problem in Ubuntu 16.04, but will remain relevant for previous releases.
For Apache or Nginx tutorials, use dedicated virtual host blocks instead of editing the default config files. This route can avoid common mistakes and maintain the default files as the fallback configuration as intended.
Encrypt and authenticate all connections between systems. Do not encourage (explicitly or implicitly) users to send credentials or transmit non-public data in the clear.
Specifically, passwords and key material must not be transmitted over unsecured connections. Database connections, logging, cluster management, and other services should ideally be encrypted at all times.
Web-based control panels must be served over HTTPS connections, and TLS/SSL should be used for services where it’s supported. All web servers should be HTTPS-enabled (or capable, at least). Use a certbot prerequisite to provide SSL certification. Public-facing services like plain HTTP are permissible, as users may still want or need to offer them, but should be strongly discouraged in the general case, especially for dynamic content. For articles that provide a plain HTTP connection, add a note or warning label to discourage plain HTTP and encourage HTTPS.
Avoid practices which constitute low- benefit security through obscurity or theatrics, like changing the default SSH port. Do configure a firewall. Our distro-specific recommendations are
ufw for Ubuntu,
iptables for Debian, and
firewalld for CentOS. However,
iptables is most consistent across platforms, and has many tools that hook into it.
Maintain the default SSH port as a norm. Changing the port should only be done in specific situations where that is a primary concern.
Disable password authentication and use key-only authentication for root or, alternatively, disable root login completely. Use strong SSH keys: at least 2048-bit RSA but recommended 4096; ECDSA is no longer recommended for technical reasons; and Ed25519 and elliptic curve algorithms are not widely supported enough.
Use passphrases for any interactive keys, but not for non-interactive processes. Set up or copy and change ownership on SSH keys from the root account to the user home directory. Install fail2ban where it’s practical.
Note that while SSH Agent Forwarding is necessary for normal workflows on platforms like CoreOS, it comes with some security concerns. Essentially, anyone with permissions on your host will be able to use the forwarding socket to connect to your local ssh-agent.
We strongly encourage the use of Let’s Encrypt for ease of use, and recommend TLS. Do use strong SSL security; look at https://cipherli.st/ (both modern and legacy recommendations).
For hosts without a domain name, we suggest a self-signed certificate of adequate strength. Again, look at https://cipherli.st/ plus the self-signed cert creation used in guides like this Self-Signed Certification on Nginx guide. Set up a strong Diffie-Hellman key when enabling encryption, as in Self-Signed SSL Certifications on Apache and Nginx.
We recommend VPNs as a solution for general encrypted communication between servers. VPNs become increasingly valuable when multiple services need to be protected between servers; instead of encrypting each service individually, all internal communication can be piped to the VPN. This approach is particularly useful if the services in question don’t support native encryption.
We generally recommend Tinc over OpenVPN for server-to-server communication. You can read this article on Ansible and Tinc for more details and implementation.
This is inherently an opinionated, living document, and will be updated often. Tech changes quickly and we at DigitalOcean do our best to keep up, but if you notice any errors or have feedback, we’ll be monitoring the comments below.
Thanks for learning with the DigitalOcean Community. Check out our offerings for compute, storage, networking, and managed databases.