zulip/docs/production/requirements.md

252 lines
11 KiB
Markdown
Raw Normal View History

# Requirements and scalability
To run a Zulip server, you will need:
- A dedicated machine or VM
- A supported OS:
- Ubuntu 22.04
- Ubuntu 24.04
- Debian 12
- A supported CPU architecture:
- x86-64
- aarch64
- At least 2 GB RAM, and 10 GB disk space
- If you expect 100+ users: 4 GB RAM, and 2 CPUs
- If you intend to [upgrade from Git][upgrade-from-git]: 3 GB RAM, or
2G and at least 1G of swap configured.
- A hostname in DNS
- Credentials for sending email
For details on each of these requirements, see below.
[upgrade-from-git]: upgrade.md#upgrading-from-a-git-repository
## Server
#### General
The installer expects Zulip to be the **only thing** running on the
system; it will install system packages with `apt` (like nginx,
PostgreSQL, and Redis) and configure them for its own use. We strongly
recommend using either a fresh machine instance in a cloud provider, a
fresh VM, [our Docker image][docker-zulip-homepage], or a dedicated
machine. If you decide to disregard our advice and use a server that
hosts other services, we can't support you, but [we do have some notes
on issues you'll encounter](install-existing-server.md).
#### Operating system
Ubuntu 22.04, Ubuntu 24.04, and Debian 12
are supported for running Zulip in production. You can also
run Zulip on other platforms that support Docker using
[docker-zulip][docker-zulip-homepage].
We recommend installing on the newest supported OS release you're
comfortable with, to save a bit of future work [upgrading the
operating system][upgrade-os].
If you're using Ubuntu, the
[Ubuntu universe repository][ubuntu-repositories] must be
[enabled][enable-universe], which is usually just:
```bash
sudo add-apt-repository universe
sudo apt update
```
[docker-zulip-homepage]: https://github.com/zulip/docker-zulip#readme
[upgrade-os]: upgrade.md#upgrading-the-operating-system
[ubuntu-repositories]: https://help.ubuntu.com/community/Repositories/Ubuntu
[enable-universe]: https://help.ubuntu.com/community/Repositories/CommandLine#Adding_the_Universe_and_Multiverse_Repositories
#### Hardware specifications
- CPU and memory: For installations with 100+ users you'll need a
minimum of **2 CPUs** and **4 GB RAM**. For installations with fewer
users, 1 CPU and 2 GB RAM is sufficient. We strongly recommend against
installing with less than 2 GB of RAM, as you will likely experience
out of memory issues installing dependencies. We recommend against
using highly CPU-limited servers like the AWS `t2` style instances
2019-12-13 01:30:22 +01:00
for organizations with hundreds of users (active or no).
- Disk space: You'll need at least 10 GB of free disk space for a
server with dozens of users. We recommend using an SSD and avoiding
2019-12-13 01:30:22 +01:00
cloud storage backends that limit the IOPS per second, since the
disk is primarily used for the Zulip database.
See our [documentation on scalability](#scalability) below for advice
on hardware requirements for larger organizations.
#### Network and security specifications
- Incoming HTTPS access (usually port 443, though this is
[configurable](deployment.md#using-an-alternate-port))
from the networks where your users are (usually, the public
2019-12-13 01:30:22 +01:00
Internet).
- Incoming port 80 access (optional). Zulip only serves content over
2019-12-13 01:30:22 +01:00
HTTPS, and will redirect HTTP requests to HTTPS.
- Incoming port 25 if you plan to enable Zulip's [incoming email
integration](email-gateway.md).
puppet: Admit we leave epmd port 4369 open on all interfaces. The Erlang `epmd` daemon listens on port 4369, and provides information (without authentication) about which Erlang processes are listening on what ports. This information is not itself a vulnerability, but may provide information for remote attackers about what local Erlang services (such as `rabbitmq-server`) are running, and where. `epmd` supports an `ERL_EPMD_ADDRESS` environment variable to limit which interfaces it binds on. While this environment variable is set in `/etc/default/rabbitmq-server`, Zulip unfortunately attempts to start `epmd` using an explicit `exec` block, which ignores those settings. Regardless, this lack of `ERL_EPMD_ADDRESS` variable only controls `epmd`'s startup upon first installation. Upon reboot, there are two ways in which `epmd` might be started, neither of which respect `ERL_EPMD_ADDRESS`: - On Focal, an `epmd` service exists and is activated, which uses systemd's configuration to choose which interfaces to bind on, and thus `ERL_EPMD_ADDRESS` is irrelevant. - On Bionic (and Focal, due to a broken dependency from `rabbitmq-server` to `epmd@` instead of `epmd`, which may lead to the explicit `epmd` service losing a race), `epmd` is started by `rabbitmq-server` when it does not detect a running instance. Unfortunately, only `/etc/init.d/rabbitmq-server` would respects `/etc/default/rabbitmq-server` -- and it defers the actual startup to using systemd, which does not pass the environment variable down. Thus, `ERL_EPMD_ADDRESS` is also irrelevant here. We unfortunately cannot limit `epmd` to only listening on localhost, due to a number of overlapping bugs and limitations: - Manually starting `epmd` with `-address 127.0.0.1` silently fails to start on hosts with IPv6 disabled, due to an Erlang bug ([1], [2]). - The dependencies of the systemd `rabbitmq-server` service can be fixed to include the `epmd` service, and systemd can be made to bind to `127.0.0.1:4369` and pass that socket to `epmd`, bypassing the above bug. However, the startup of this service is not guaranteed, because it races with other sources of `epmd` (see below). - Any process that runs `rabbitmqctl` results in `epmd` being started if one is not currently running; these instances do not respect any environment variables as to which addresses to bind on. This is also triggered by `service rabbitmq-server status`, as well as various Zulip cron jobs which inspect the rabbitmq queues. As such, it is difficult-to-impossible to ensure that some other `epmd` process will not win the race and open the port on all interfaces. Since the only known exposure from leaving port 4369 open is information that rabbitmq is running on the host, and the complexity of adjusting this to only bind on localhost is high, we remove the setting which does not address the problem, and document that the port is left open, and should be protected via system-level or network-level firewalls. [1]: https://bugs.launchpad.net/ubuntu/+source/erlang/+bug/1374109 [2]: https://github.com/erlang/otp/issues/4820
2021-12-07 21:13:47 +01:00
- Incoming port 4369 should be protected by a firewall to prevent
exposing `epmd`, an Erlang service which does not support binding
only to localhost. Leaving this exposed will allow unauthenticated
remote users to determine that the server is running RabbitMQ, and
on which port, though no further information is leaked.
- Outgoing HTTP(S) access (ports 80 and 443) to the public Internet so
that Zulip can properly manage image and website previews and mobile
push notifications. Outgoing Internet access is not required if you
[disable those
features](https://zulip.com/help/image-video-and-website-previews).
- Outgoing SMTP access (usually port 587) to your [SMTP
server](email.md) so that Zulip can send emails.
- A domain name (e.g., `zulip.example.com`) that your users will use to
access the Zulip server. In order to generate valid SSL
2019-12-13 01:30:22 +01:00
certificates [with Certbot][doc-certbot], and to enable other
services such as Google authentication, public DNS name is simpler,
but Zulip can be configured to use a non-public domain or even an IP
address as its external hostname (though we don't recommend that
configuration).
- Zulip supports [running behind a reverse proxy][reverse-proxy].
- Zulip configures [Smokescreen, an outgoing HTTP
proxy][smokescreen-proxy], to protect against [SSRF attacks][ssrf],
which prevents user from making the Zulip server make requests to
private resources. If your network has its own outgoing HTTP proxy,
Zulip supports using that instead.
2019-12-13 01:30:22 +01:00
Zulip does not, itself, require SSH, but most installations will also require
access to incoming port 22 for SSH access for remote access.
[ssrf]: https://owasp.org/www-community/attacks/Server_Side_Request_Forgery
[smokescreen-proxy]: deployment.md#customizing-the-outgoing-http-proxy
[reverse-proxy]: reverse-proxies.md
## Credentials needed
#### SSL certificate
Your Zulip server will need an SSL certificate for the domain name it
uses. For most Zulip servers, the recommended (and simplest) way to
get this is to just [use the `--certbot` option][doc-certbot] in the
Zulip installer, which will automatically get a certificate for you
and keep it renewed.
For test installations, an even simpler alternative is always
available: [the `--self-signed-cert` option][doc-self-signed] in the
installer.
If you'd rather acquire an SSL certificate another way, see our [SSL
certificate documentation](ssl-certificates.md).
[doc-certbot]: ssl-certificates.md#certbot-recommended
[doc-self-signed]: ssl-certificates.md#self-signed-certificate
#### Outgoing email
- Outgoing email (SMTP) credentials that Zulip can use to send
outgoing emails to users (e.g., email address confirmation emails
during the signup process, message notification emails, password
reset, etc.). If you don't have an existing outgoing SMTP solution,
read about
[free outgoing SMTP options and options for prototyping](email.md#free-outgoing-email-services).
Once you have met these requirements, see [full instructions for installing
Zulip in production](install.md).
## Scalability
2019-12-13 01:30:22 +01:00
This section details some basic guidelines for running a Zulip server
for larger organizations (especially >1000 users or 500+ daily active
users). These guidelines are conservative, since they are intended to
be sufficient for a wide range of possible usage patterns that may not
be applicable to your installation.
Zulip's resource needs depend mainly on 3 parameters:
- daily active users (e.g., number of employees if everyone's an
employee)
- total user accounts (can be much larger)
- message volume.
In the following, we discuss a configuration with at most two types of
servers: application servers (running Django, Tornado, RabbitMQ,
Redis, Memcached, etc.) and database servers. Of the application
server services, Django dominates the resource requirements. One can
run every service on its own system (as
[docker-zulip](https://github.com/zulip/docker-zulip) does) but for
most use cases, there's little scalability benefit to doing so. See
[deployment options](deployment.md) for details on
installing Zulip with a dedicated database server.
- **Dedicated database**. For installations with hundreds of daily
active users, we recommend using a [remote PostgreSQL
database](postgresql.md), but it's not required.
- **RAM:** We recommend more RAM for larger installations:
- With 25+ daily active users, 4 GB of RAM.
- With 100+ daily active users, 8 GB of RAM.
- With 400+ daily active users, 16 GB of RAM for the Zulip
application server, plus 16 GB for the database.
- With 2000+ daily active users 32 GB of RAM, plus 32 GB for the
database.
- Roughly linear scaling beyond that.
- **CPU:** The Zulip application server's CPU usage is heavily
optimized due to extensive work on optimizing the performance of
requests for latency reasons. Because most servers with sufficient
RAM have sufficient CPU resources, CPU requirements are rarely an
issue. For larger installations with a dedicated database, we
recommend high-CPU instances for the application server and a
database-optimized (usually low CPU, high memory) instance for the
database.
- **Disk for application server:** We recommend using [the S3 file
uploads backend][s3-uploads] to store uploaded files at scale. With
the S3 backend configuration, we recommend 50 GB of disk for the OS,
Zulip software, logs and scratch/free space. Because uploaded files
are cached locally, you may need more disk space if you make heavy
use of uploaded files.
- **Disk for database:** SSD disk is highly recommended. For
installations where most messages have <100 recipients, 10 GB per 1M
messages of history is sufficient plus 1 GB per 1000 users is
sufficient. If most messages are to public channels with 10K+ users
subscribed (like on chat.zulip.org), add 20 GB per (1000 user
accounts) per (1M messages to public channels).
- **Example:** When
[the Zulip development community](https://zulip.com/development-community/) server
2019-12-13 01:30:22 +01:00
had 12K user accounts (~300 daily actives) and 800K messages of
history (400K to public channels), it was a default configuration
single-server installation with 16 GB of RAM, 4 cores (essentially
always idle), and its database was using about 100 GB of disk.
- **Disaster recovery:** One can easily run a warm spare application
server and a warm spare database (using [PostgreSQL warm standby
replicas][streaming-replication]). Make sure the warm spare
application server has copies of `/etc/zulip` and you're either
syncing `LOCAL_UPLOADS_DIR` or using the [S3 file uploads
backend][s3-uploads].
- **Sharding:** For servers with several thousand daily active users,
Zulip supports sharding its real-time-push Tornado service, both by
realm/organization (for hosting many organizations) and by user ID
(for hosting single very large organizations)
Care must be taken when dividing traffic for a single Zulip realm
between multiple Zulip application servers, which is why we
recommend a hot spare over load-balancing for most installations
desiring extra redundancy.
If you have scalability questions or are unsure whether Zulip is a fit
for your use case, contact [Zulip sales or support][contact-support]
for assistance.
For readers interested in technical details around what features
impact Zulip's scalability, this [performance and scalability design
document](../subsystems/performance.md) may also be of interest.
[s3-uploads]: upload-backends.md#s3-backend-configuration
[streaming-replication]: postgresql.md#postgresql-warm-standby
[contact-support]: https://zulip.com/help/contact-support