Slavi Pantaleev ca786cc343 Revert "Upgrade Synapse (1.31 -> 1.32)"

This reverts commit f825c7c263.

Related to https://github.com/spantaleev/matrix-docker-ansible-deploy/issues/1010

2021-04-20 23:40:55 +03:00

31 KiB

Raw Blame History

Frequently Asked Questions

This documentation page tries to answer various Frequently Asked Questions about all things Matrix, with a focus on this Ansible playbook (What is Ansible? How does it work?).

This FAQ page does not intend to replace the matrix.org FAQ (please see that one too).

We've only started this FAQ recently, so it's still somewhat empty.

Also, we encourage you to not dig yourself into a hole by reading way too much. When you've heard enough, proceed to Prerequisites to get guided into installing Matrix.

Introductory

Where do I find more questions and answers about Matrix?

This is a Frequently Asked Questions page focused on this Ansible playbook (What is Ansible? How does it work?) for deploying a Matrix server.

For a lot more generic questions and answers, see the matrix.org FAQ.

What is Matrix? What is Element? What is Synapse? Why are you confusing me with so many terms?

Matrix is a new type of realtime communication (chat) network, the closest analogy to which is probably "email".

You don't just use the "email" protocols (SMTP, POP3, IMAP) directly though. There's a server somewhere which stores your data (@gmail.com, @yahoo.com, @hotmail.com, @your-company.com) and you access it by using these "email" protocols via some client program (Outlook, Thunderbird, some website, etc).

In the world of the Matrix chat protocol, there are various client programs. The first and currently most full-featured one is called Element (used to be called Riot.im and Vector.im in the past). There are many other clients. You can switch clients as much as you want until you find the one that is right for you on a given platform (you may use Element on your desktop, but Fluffychat on your phone, etc).

Matrix is also like email due to the fact that there are many servers around the world which can all talk to each other (you can send email from @gmail.com addresses to @yahoo.com and @hotmail.com addresses). It's the same with Matrix (@bob:his-domain.com can talk to @alice:her-domain.org).

If someone else is hosting your Matrix server (you being @user:matrix.org or some other public server like this), all you need is a Matrix client program, like Element.

If you'd like to host your own server (you being @user:your-own-domain.com), you'd need to set up a Matrix server program, like Synapse.

In short:

Matrix is the protocol - a set of rules about how the chat network operates
Element is a client program you can use to participate on the Matrix chat network via some server (yours or someone else's). There are also many other client programs.
Synapse is a server program you can use to host your very own Matrix server.

This FAQ here mostly focuses on installing various Matrix services using the Ansible automation tool. You can learn much more about Matrix in the matrix.org FAQ.

People I wish to talk to are not on Matrix. Can I talk to them?

You most likely can. Besides Matrix-native chats, Matrix also supports the concept of "bridging", which allows you to plug other networks into it.

This Ansible playbook can help you install tens of bridges for various networks.

Besides setting up your own bridges (preferable), you can also use some public bridges hosted by others.

How do I get started with Matrix?

One of Matrix's distinguishing strengths (compared to other chat networks) is its decentralized nature. There's not just one entity (company, organization) controlling the servers. Rather there's thousands of servers operated by different people - one server being insecure, slow or disrespective toward its users does not affect the rest of the network. To participate in that decentralization in its fullest, consider hosting your own server or using some public server other than the largest/default one (matrix.org).

There are 3 ways to get into Martix, depending on your technical ability and needs:

using the existing default server - the easiest way is to use an existing server. The largest public Matrix server is matrix.org and it's configured as a default server in clients such as Element and many others. Just use Element on the browser via that link (or download the Element app on a smartphone), create an account and start chatting.
using some other server - instead of using the largest public server (matrix.org), you can use another public one. Here's a list of public Matrix servers to choose from. Again, you download Element or some other client of your choosing and adjust the homeserver URL during login.
using your own server - running your own server puts you in ultimate control of your data. It also lets you have your own user identifiers (e.g. @bob:your-domain.com). See How do I set up my own Matrix server.

How do I set up my own Matrix server?

Normally, you'd first choose the Matrix server software you'd like to run. At the time of this writing (January/2021), there's only one fully-featured server program, so there's only one reasonable choice. That's Synapse.

There are many guides about installing Synapse. Using this Ansible playbook is just one way of doing it.

Naturally, we're biased, so our usual recommendation is to go with this Ansible playbook, instead of installing Synapse (and many many other things around it) manually. To get started with the playbook, start at the Prerequisites page.

What is Ansible? How does it work?

Ansible is an automation program. This "playbook" is a collection of tasks/scripts that will set up a Matrix server for you, so you don't have to perform these tasks manually.

We have written these automated tasks for you and all you need to do is execute them using the Ansible program.

You can install Ansible and this playbook code repository on your own computer and tell it to install Matrix services at the server living at matrix.DOMAIN. We recommend installing Ansible on your own computer.

Alternatively, you can download Ansible and the playbook itself directly on the matrix.DOMAIN server.

To learn more, see our dedicated Ansible documentation page.

Why use this playbook and not install Synapse and other things manually?

There are various guides telling you how easy it is to install Synapse.

Reading the documentation of this Ansible playbook, you may also be thinking:

I don't know what Ansible is. I don't know what Docker is. This looks more complicated.

.. so you may be leaning toward installing Synapse manually.

The problem with a manual installation is:

Synapse is written in Python. If not packaged for your distribution, you'd need to install various Python modules, etc., and keep them updated.
Synapse requires a Postgres database (it can run on SQLite, but that's very much discouraged). So you'd need to install Postgres as well.
you may also need a reverse-proxy server in front of it (nginx, Apache), so you'd need to be familiar with that
SSL is required, so you'd need to obtain Let's Encrypt (or other free or non-free) certificates for one or more domain names. You'd need to be familiar with certbot (when using Let's Encrypt) or similar software.
for each additional component you'd like to add (client like Element, bridge to some other chat network, Integration Manager (sitckers, other services), Identity Manager, etc.), you'll need to spend extra time installing and wiring it with the rest of the system in a way that works.
you'll likely get slower updates for all of these components, depending on your distro packaging or your own time and ability

The playbook, on the other hand, installs a bunch of components for you by default, obtains SSL certificates for you, etc. If you'd like, you can enable various bridges and other services with very little effort. All the components are wired to work together.

All services run in Docker containers (most being officially provided by each component's developers), so we're not at the mercy of distro packaging.

Why use this playbook and not just use the Docker image directly?

Reasons are similar to the reasons for not installing manually.

Besides Synapse, you'd need other things - a Postgres database, likely the Element client, etc., etc.

Using the playbook, you get all these components in a way that works well together out of the box.

What's different about this Ansible playbook compared to EMnify/matrix-synapse-auto-deploy?

This is similar to the EMnify/matrix-synapse-auto-deploy Ansible deployment, but:

this one is a complete Ansible playbook (instead of just a role), so it's easier to run - especially for folks not familiar with Ansible
this one installs and hooks together a lot more Matrix-related services for you (see above)
this one can be executed more than once without causing trouble
works on various distros: CentOS (7.0+), Debian-based distributions (Debian 9/Stretch+, Ubuntu 16.04+), Archlinux
this one installs everything in a single directory (/matrix by default) and doesn't "contaminate" your server with files all over the place
this one doesn't necessarily take over ports 80 and 443. By default, it sets up nginx for you there, but you can also use your own webserver
this one runs everything in Docker containers, so it's likely more predictable and less fragile (see Docker images used by this playbook)
this one retrieves and automatically renews free Let's Encrypt SSL certificates for you
this one optionally can store the media_store content repository files on Amazon S3 (but defaults to storing files on the server's filesystem)
this one optionally allows you to use an external PostgreSQL server for Synapse's database (but defaults to running one in a container)
helps you import data from a previous installation (so you can migrate your manual virtualenv/Docker setup to a more managed one)
this one is actually maintained

What kind of server do I need to install Matrix using this Ansible playbook?

We list our server requirements in Prerequisites.

Why not run Matrix on Kubernetes?

There's no reason not to run Matrix on Kubernetes.

However, that's overly complicated for thousands of us who just want to run a single small (and sometimes not so small) Matrix server, either using "cloud" servers or even a Raspberry Pi at home.

For us, a Kubernetes-based setup which requires a cluster of multiple computers and is more technically-involved is a no-go.

There are others working on automating a Matrix-on-Kubernetes setup, such as this Helm chart: https://github.com/dacruz21/matrix-chart.

Why don't you use Podman instead of Docker?

We like the philosophy of a daemonless container runtime, but Podman is just not ready for our use case yet.

Learn more about our past experiences/attempts to give Podman a chance, by reading this issue.

In short, alias podman=docker is a lie (for us).

Why use Docker?

Docker is one of our 2 hard dependencies (the other one being systemd).

It lets us run services in an isolated manner and independently of the (usually old) packages available for distributions.

It also lets us have a unified setup which runs the same across various supported distros (see them on Prerequisites).

Is Docker a hard requirement?

Yes. See Why don't you use Podman instead of Docker? for why we're not using another container runtime.

All of our services run in containers. It's how we achieve predictability and also how we support tens of different services across lots of distros.

The only thing we need on the distro is systemd and Python (we install Docker ourselves, unless you ask us not to).

Why don't you use docker-compose?

Instead of using docker-compose, we prefer installing systemd services and scheduling those independently.

There are people who have worked on turning this setup into a docker-compose-based one. See these experiments here.

Can I run this on a distro without systemd?

No. systemd is one of our 2 hard dependencies (the other one being Docker).

Can I install this on a Raspberry Pi?

Yes, you can. See our Alternative Architectures documentation page.

Whether a Raspberry Pi has enough power to give you a good experience is another question. It depends on your use case.

Also see: What kind of server specs do I need?.

What kind of server specs do I need?

This largely depends on your use case. It's not so much the number of users that you plan to host, but rather the number of large rooms they will join.

Federated rooms with lots of history and containing hundreds of other servers are very heavy CPU-wise and memory-wise.

You can probably use a 1 CPU + 1GB memory server to host hundreds of local users just fine, but as soon as one of them joins a federated room like #matrix:matrix.org (Matrix HQ) or some IRC-bridged room (say ##linux), your server will get the need for a lot more power (at least 2GB RAM, etc).

Running Matrix on a server with 1GB of memory is possible (especially if you disable some not-so-important services). See How do I optimize this setup for a low-power server?.

We recommend starting with a server having at least 2GB of memory and even then using it sparingly. If you know for sure you'll be joining various large rooms, etc., then going for 4GB of memory or more is a good idea.

Besides the regular Matrix stuff, we also support things like video-conferencing using Jitsi and other additional services which (when installed) may use up a lot of memory. Things do add up. Besides the Synapse Matrix server, Jitsi is especially notorious for consuming a lot of resources. If you plan on running Jitsi, we recommend a server with at least 2GB of memory (preferrably more). See our Jitsi documentation page to learn how to optimize its memory/CPU usage.

Can I run this in an LXC container?

If your distro runs within an LXC container, you may hit this issue. It can be worked around, if absolutely necessary, but we suggest that you avoid running from within an LXC container.

Configuration

Why install my server at matrix.DOMAIN and not at the base DOMAIN?

It's the same with email servers. Your email address is likely name@company.com, not name@mail.company.com, even though it's mail.company.com that is really handling your data for @company.com email to work.

Using a separate domain name is easier to manage (although it's a little hard to get right at first) and keeps your Matrix server isolated from your website (if you have one), from your email server (if you have one), etc.

We allow matrix.DOMAIN to be the Matrix server handling Matrix stuff for DOMAIN by Server Delegation. During the installation procedure, we recommend that you set up server delegation using the .well-known method.

If you'd really like to install Matrix services directly on the base domain, see How do I install on matrix.DOMAIN without involving the base DOMAIN?.

I don't control anything on the base domain and can't set up delegation to matrix.DOMAIN. What do I do?

If you're not in control of your base domain (or the server handling it) at all, you can take a look at How do I install on matrix.DOMAIN without involving the base DOMAIN?

I can't set up HTTPS on the base domain. How will I get Matrix federating?

If you really can't obtain an HTTPS certificate for your base domain, you can take a look at How do I install on matrix.DOMAIN without involving the base DOMAIN?

How do I install on matrix.DOMAIN without involving the base DOMAIN?

This Ansible playbook guides you into installing a server for DOMAIN (user identifiers are like this: @user:DOMAIN), while the server is at matrix.DOMAIN.

If you're fine with uglier identifiers (@user:matrix.DOMAIN, which is the equivalent of having an email address like bob@mail.company.com, instead of just bob@company.com), you can do that as well using the following configuration in your vars.yml file:

# This is what your identifiers are like (e.g. `@bob:matrix.YOUR_BASE_DOMAIN`).
matrix_domain: "matrix.YOUR_BASE_DOMAIN"

# This is where Matrix services
matrix_server_fqn_matrix: "matrix.YOUR_BASE_DOMAIN"

# This is where you access the Element web UI from (if enabled via `matrix_client_element_enabled: true`; enabled by default).
# This and the Matrix FQN (see above) are expected to be on the same server.
#
# Feel free to use `element.matrix.YOUR_BASE_DOMAIN`, if you'd prefer that.
matrix_server_fqn_element: "element.YOUR_BASE_DOMAIN"

# This is where you access Dimension (if enabled via `matrix_dimension_enabled: true`; NOT enabled by default).
#
# Feel free to use `dimension.matrix.YOUR_BASE_DOMAIN`, if you'd prefer that.
matrix_server_fqn_dimension: "dimension.YOUR_BASE_DOMAIN"

# This is where you access Jitsi (if enabled via `matrix_jitsi_enabled: true`; NOT enabled by default).
#
# Feel free to use `jitsi.matrix.YOUR_BASE_DOMAIN`, if you'd prefer that.
matrix_server_fqn_jitsi: "jitsi.YOUR_BASE_DOMAIN"

I don't use the base domain for anything. How am I supposed to set up Server Delegation for Matrix services?

If you don't use your base domain for anything, then it's hard for you to "serve files over HTTPS" on it -- something we ask you to do for the .well-known setup (needed for Server Delegation).

Luckily, the playbook can set up your Matrix server (at matrix.DOMAIN) to also handle traffic for the base domain (DOMAIN).

See Serving the base domain.

How do I optimize this setup for a low-power server?

You can disable some not-so-important services to save on memory.

# An identity server is not a must.
matrix_ma1sd_enabled: false

# Disabling this will prevent email-notifications and other such things from working.
matrix_mailer_enabled: false

# You can also disable this to save more RAM,
# at the expense of audio/video calls being unreliable.
matrix_coturn_enabled: false

# This makes Synapse not keep track of who is online/offline.
#
# Keeping track of this and announcing such online-status in federated rooms with
# hundreds of servers inside is insanely heavy (https://github.com/matrix-org/synapse/issues/3971).
#
# If your server does not federate with hundreds of others, enabling this doesn't hurt much.
matrix_synapse_use_presence: false

You can also consider implementing a restriction on room complexity, in order to prevent users from joining very heavy rooms:

matrix_synapse_configuration_extension_yaml: |
  limit_remote_rooms:
    enabled: true
    complexity: 1.0 # this limits joining complex (~large) rooms, can be  
					# increased, but larger values can require more RAM

If you've installed Jitsi (not installed by default), there are additional optimizations listed on its documentation page that you can perform.

I already have Docker on my server. Can you stop installing Docker via the playbook?

Yes, we can stop installing Docker ourselves. Just use this in your vars.yml file:

matrix_docker_installation_enabled: true

I run another webserver on the same server where I wish to install Matrix. What now?

By default, we install a webserver for you (nginx), but you can also use your own webserver.

How is the effective configuration determined?

Configuration variables are defined in multiple places in this playbook and are considered in this order:

there are defaults coming from each role's defaults file (role/matrix*/defaults/main.yml). These variable values aim to be good defaults for when the role is used standalone (outside of this collection of roles, also called playbook).
then, there are overrides in group_vars/matrix_servers, which aim to adjust these "standalone role defaults" to something which better fits the playbook in its entirety.
finally, there's your inventory/host_vars/matrix.DOMAIN/vars.yml file, which is the ultimate override

What configuration variables are available?

You can discover the variables you can override in each role (role/matrix*/defaults/main.yml).

As described in How is the effective configuration determined?, these role-defaults may be overriden by values defined in group_vars/matrix_servers.

Refer to both of these for inspiration. Still, as mentioned in Configuring the playbook, you're only ever supposed to edit your own inventory/host_vars/matrix.DOMAIN/vars.yml file and nothing else inside the playbook (unless you're meaning to contribute new features).

I'd like to adjust some configuration which doesn't have a corresponding variable. How do I do it?

The playbook doesn't aim to expose all configuration settings for all services using variables. Doing so would amount to hundreds of variables that we have to create and maintain.

Instead, we only try to make some important basics configurable using dedicated variables you can see in each role. See What configuration variables are available?.

Besides that, each role (component) aims to provide a matrix_SOME_COMPONENT_configuration_extension_yaml (or matrix_SOME_COMPONENT_configuration_extension_json) variable, which can be used to override the configuration.

Check each role's role/matrix*/defaults/main.yml for the corresponding variable and an example for how use it.

Installation

How do I run the installation?

See Installing to learn how to use Ansible to install Matrix services.

Of course, don't just jump straight to Installing. Rather, start at Prerequisites and get guided from there (into setting up DNS, configuring the playbook, etc).

I installed Synapse some other way. Can I migrate such a setup to the playbook?

Yes, you can.

You generally need to do a playbook installation (start at the Prerequisites page), followed by importing your existing data into it.

This Ansible playbook guides you into installing a server for DOMAIN (user identifiers are like this: @user:DOMAIN), while the server is at matrix.DOMAIN. If your existing setup has a server name (server_name configuration setting in Synapse's homeserver.yaml file) other than the base DOMAIN, you may need to tweak some additional variables. This FAQ entry may be of use if you're dealing with a more complicated setup - How do I install on matrix.DOMAIN without involving the base DOMAIN?

After configuring the playbook and installing and before starting services (done with ansible-playbook ... --tags=start) you'd import your SQLite (or Postgres) database and also import your media store.

I've downloaded Ansible and the playbook on the server. It can't connect using SSH.

If you're using the playbook directly on the server, then Ansible doesn't need to connect using SSH.

It can perform a local connection instead. Just set ansible_connection=local at the end of the server line in inventory/hosts and re-run the playbook.

If you're running Ansible from within a container (one of the possibilities we list on our dedicated Ansible documentation page), then using ansible_connection=local is not possible.

Troubleshooting

I get "Error response from daemon: configured logging driver does not support reading" when I do `docker logs matrix-synapse`.

See How can I see the logs?.

How can I see the logs?

We utilize systemd/journald for logging.

To see logs for Synapse, run journalctl -fu matrix-synapse.service. You may wish to see the manual page for journalctl.

Available service names can be seen by doing ls /etc/systemd/system/matrix*.service on the server.

Some services also log to files in /matrix/*/data/.., but we're slowly moving away from that.

We also disable Docker logging, so you can't use docker logs matrix-* either. We do this to prevent useless double (or even triple) logging and to avoid having to rotate log files.

We just simply delegate logging to journald and it takes care of persistence and expiring old data.

Also see: How long do systemd/journald logs persist for?

How long do systemd/journald logs persist for?

On some distros, the journald logs are just in-memory and not persisted to disk.

Consult (and feel free to adjust) your distro's journald logging configuration in /etc/systemd/journald.conf.

To enable persistence and put some limits on how large the journal log files can become, adjust your configuration like this:

[Journal]
RuntimeMaxUse=200M
SystemMaxUse=1G
RateLimitInterval=0
RateLimitBurst=0
Storage=persistent

Maintenance

Do I need to do anything to keep my Matrix server updated?

Yes. We don't update anything for you automatically.

See our documentation page about upgrading services.

How do I move my existing installation to another (VM) server?

If you have an existing installation done using this Ansible playbook, you can easily migrate that to another server using our dedicated server migration guide.

If your previous installation is done in some other way (not using this Ansible playbook), see I installed Synapse some other way. Can I migrate such a setup to the playbook?.

How do I back up the data on my server?

We haven't documented this properly yet, but the general advice is to:

back up Postgres by making a database dump. See Backing up PostgreSQL
back up all /matrix files, except for /matrix/postgres/data (you already have a dump) and /matrix/postgres/data-auto-upgrade-backup (this directory may exist and contain your old data if you've performed a major Postgres upgrade).

You can later restore these roughly like this:

restore the /matrix directory and files on the new server manually
run the playbook again (see Installing), but don't start services yet (don't run ... --tags=start). This step will fix any file permission mismatches and will also set up additional software (Docker, etc.) and files on the server (systemd service, etc.).
perform a Postgres database import (see Importing Postgres) to restore your database backup
start services (see Starting the services)

If your server's IP address has changed, you may need to set up DNS again.

What is this `/matrix/postgres/data-auto-upgrade-backup` directory that is taking up so much space?

When you perform a major Postgres upgrade, we save the the old data files in /matrix/postgres/data-auto-upgrade-backup, just so you could easily restore them should something have gone wrong.

After verifying that everything still works after the Postgres upgrade, you can safely delete /matrix/postgres/data-auto-upgrade-backup

How do I debug or force SSL certificate renewal?

SSL certificate renewal normally happens automatically via systemd timers.

If you're having trouble with SSL certificate renewal, you can inspect the renewal logs using:

journalctl -fu matrix-ssl-lets-encrypt-certificates-renew.service
or by looking at the log files in /matrix/ssl/log/

To trigger renewal, run: systemctl start matrix-ssl-lets-encrypt-certificates-renew.service. You can then take a look at the logs again.

If you're using the integrated webserver (matrix-nginx-proxy), you can reload it manually like this: systemctl reload matrix-nginx-proxy. Reloading also happens periodically via a systemd timer.

If you're using your own webserver instead of the integrated one (matrix-nginx-proxy) you may also need to reload/restart it, to make it pick up the renewed SSL certificate files.

31 KiB Raw Blame History