Without these:
- `--tags=install-synapse` and `--tags=install-all` would be incomplete
and will not contain Synapse worker configuration
- `--tags=install-synapse-reverse-proxy-companion` and
`--tags=setup-synapse-reverse-proxy-companion` would not contain
Synapse worker configuration
Running with a user (like `matrix:matrix`) fails if Etherpad is enabled,
because `/matrix/etherpad` is owned by `matrix_etherpad_user_uid`/`matrix_etherpad_user_gid` (`5001:5001`).
The `matrix` user can't acccess the Etherpad directory for this reason
and Borgmatic fails when trying to make a backup.
There may be other things under `/matrix` which similarly use
non-`matrix:matrix` permissions.
Another workaround might have been to add `/matrix/etherpad` (and
potentially other things) to `matrix_backup_borg_location_exclude_patterns`, but:
- that means Etherpad won't be backed up - not great
- only excluding Etherpad may not be enough. There may be other files we
need to exclude as well
---
Running with `root` is still not enough though.
We need at least the `CAP_DAC_OVERRIDE` capability, or we won't be able to read the
`/etc/borgmatic.d/config.yaml` configuration file (owned by
`matrix:matrix` with `0640` permissions).
---
Additionally, it seems like the backup process tries to write to at least a few directories:
- `/root/.borgmatic`
- `/root/.ssh`
- `/root/.config`
> [Errno 30] Read-only file system: '/root/.borgmatic'
> Error while creating a backup.
> /etc/borgmatic.d/config.yaml: Error running configuration file
We either need to stop mounting the container filesystem as readonly
(remove `--read-only`) or to allow writing via a `tmpfs`.
I've gone the `tmpfs` route which seems to work.
In any case, the mounted source directories (`matrix_backup_borg_location_source_directories`)
are read-only regardless, so our actual source files are protected from unintentional changes.
Without this, it's a string and borg says:
> At 'hooks.postgresql_databases[INDEX_HERE].port': '5432' is not of type 'integer'
> /etc/borgmatic/config.yaml /etc/borgmatic.d /tmp/.config/borgmatic/config.yaml /tmp/.config/borgmatic.d: No valid configuration files found
.. and fails to do anything.
This role is usable on its own and it's not tied to Matrix, so
extracting it out into an independent role that we install via
ansible-galaxy makes sense.
This also fixes the confusion from the other day, where
`matrix_postgres_*` had to be renamed to `devture_postgres_*`
(unless it was about `matrix_postgres_backup_*`).
We now can safely say that ALL `matrix_postgres_*` variables need to be
renamed.
Fixes https://github.com/spantaleev/matrix-docker-ansible-deploy/issues/2305
More details about the new key type can be found here:
https://eff-certbot.readthedocs.io/en/stable/using.html#rsa-and-ecdsa-keys
Existing RSA-based keys will continue to renew as RSA until manual
action is taken. Example from the documentation above:
> certbot renew --key-type ecdsa --cert-name example.com --force-renewal
In the future, we may add a command which does this automatically for
all domains.
There was no need to add `synapse` to the list manually
and then add all other additional databases.
When the `synapse` database was the main database, this made sense.
Since a long time ago already, ALL databases are "additional" databases,
so the `synapse` database is part of that list.
We could additional add the main (`matrix`) database to this list,
but there's probably no point in backing that one up.
Tests were carried out like this:
- `virtualenv3 env`
- `./env/bin/pip install ansible==4.10.0 ansible-core==2.11.7`
- `./env/bin/ansible-playbook .....`
The lowest version of `ansible-core` available on PyPI right now is
2.11.0. That version has trouble with `ansible==4.0.0` though.
The errors we were hitting seemed to be resolved by others online by
using `ansible==4.10.0` instead, which has a minimum `ansible-core`
requirement of `2.11.7`, so that's what we went with.
Older versions of Ansible may work, but.. I'm having trouble
installing them and don't want to spend too much time on digging through
ancient versions and testing them out. People should just learn to run
up-to-date software.
- forego removing Docker images - it's not effective anyway, because it
only removes the last version.. which is a drop in the bucket, usually
- do not reload systemd - it's none of our business. `--tags=start`,
etc., handle this
- combine all uninstall tasks under a single block, which only runs if
we detect traces (a leftover systemd .service file) of the component.
If no such .service is detected, we skip them all. This may lead to
incorect cleanup in rare cases, but is good enough for the most part.
Ansible recently started showing warnings about `crypt` being
deprecated. If one installs `passlib`, the `password_hash` values that
are generated would be different by default. With this patch, we ensure
consistency regardless of which one is used.
After this patch, password hashes (and UUIDs derived from them) will
change once, but they should be stable after that.
These hashes changing is not a problem, because the playbook
changes all references to the new values. Changes are only a problem if
they're done partially and with different tools.
For example:
- `--tags=setup-COMPONENT` with `passlib`
- `--tags=setup-postgres` with `crypt` (no `passlib`)
If so, the Postgres database password's value will differ for the
configuration generated for `COMPONENT`.
The `rounds=` value is arbitrary. It doesn't matter what it is,
as long as it's different than the default for `crypt` (5000)
and the default for `passlib` for `sha512` (656000).
Source (https://docs.ansible.com/ansible/latest/playbook_guide/playbooks_filters.html):
> To ensure idempotency, specify rounds to be neither crypt’s nor passlib’s default, which is 5000 for crypt and a variable value (535000 for sha256, 656000 for sha512) for passlib
We had checks to avoid stopping/deleting systemd services for workers
that used to exist and will continue to exist, but we were deleting
config files for workers each time.. Only to recreate them again later.
This lead to:
- too many misleading "changed" tasks
- too much unnecessary work
- potential failures during playbook execution possibly leaving the
system in a bad state (no worker config files)
Slavi PantaleevandGitHub