These endpoints should not be proxied to a generic Synapse worker
without other preparation (setting up stream writers, sending traffic
to a specific stream writer, etc.).
Disabling them for now. In the future, we'd like to fix up our awk
script to disable them automatically.
This is a fix up for 058fedff91
This prevented us from keeping our workers reverse-proxying definitions
updated since Synapse v1.54.0.
The last `workers.md` file we could parse is at commit
02632b3504ad4512c5f5a4f859b3fe326b19c788.
Parsing regressed at commit c56bfb08bc071368db23f3b1c593724eb4f205f0,
because the introduction message for `synapse.app.generic_worker` said
"If":
> If a worker is set up to handle a..
.. which made the AWK script think that definitions below were
conditional (which they're not in this case).
This patch fixes up the regex for determining if a line is conditional
or not, so that it doesn't trip up. Hopefully, it doesn't miss something
important.
People often report and ask about these "failures".
More-so previously, when the `docker kill/rm` output was collected,
but it still happens now when people do `systemctl status
matrix-something` and notice that it says "FAILURE".
Suppressing to avoid further time being wasted on saying "this is
expected".
This switches the playbook from devture/rust-synapse-compress-state (a
container image which wraps the upstream-prebuilt amd64 binary of
rust-synapse-compress-state) to registry.gitlab.com/mb-saces/rust-synapse-compress-state
(https://gitlab.com/mb-saces/rust-synapse-compress-state), which builds
rust-synapse-compress-state from source and provides a multi-arch image
that currently works on amd64 and arm64.
Ideally, we'll stop using `:latest` and arm32 support will be made
available upstream as well at some point.
Discussed here: https://gitlab.com/mb-saces/rust-synapse-compress-state/-/issues/1
Changed the commit hash in matrix_synapse_ext_spam_checker_mjolnir_antispam_git_version to latest. Fixes a Synapse ImportError with mjolnir v1.4.1 leading to a Synapse crash-loop.
`roles/matrix-synapse/vars/workers.yml` has not been updated here,
because running `roles/matrix-synapse/files/workers-doc-to-yaml.awk`
seems to cause some trouble with the current `workers.md` file
available at https://github.com/matrix-org/synapse/raw/master/docs/workers.md,
namely lots of:
> FIXME: ADDITIONAL CONDITIONS REQUIRED: to be enabled manually
lines and commented out regex lines.
This is something that remains to be investigated/fixed.
The `to_nice_yaml` helper will by default wrap any string YAML values on
the first space after column 80. This can in worst case yield invalid
YAML syntax. More details in Ansible's documentation here:
https://docs.ansible.com/ansible/latest/user_guide/playbooks_filters.html#formatting-data-yaml-and-json
In short, you need to explicitly provide a custom width argument of a
high number of some kind to avoid the line wrapping.
matrix_synapse_federation_port_enabled can be disabled by users, for
example, when one wants to use the same port for client and federation
requests (docs/configuring-playbook-federation.md).
For now, we disable the new `com.devture.shared_secret_auth` login type
by default, because it causes problems with Element:
https://github.com/vector-im/element-web/issues/19605
This also becomes the first module to use the new Synapse module system
that got introduced in Synapse v1.46.0.
Despite these upgrades, things should remain functionally identical
as far as bridges, matrix-corporal or other consumers are concerned.
This also removes the `matrix_synapse_version_arm64` variable we've
been dragging around for a long time.
Since https://github.com/matrix-org/synapse/pull/11810, a multiarch Synapse
container image (for AMD64 and ARM64) is released at the same time.
Reverts b1b4ba501f, 90c9801c56, a3c84f78ca, ..
I haven't really traced it (yet), but on some servers, I'm observing
`ansible-playbook ... --tags=start` completing very slowly, waiting
to stop services. I can't reproduce this on all Matrix servers I manage.
I suspect that either the systemd version is to blame or that some
specific service is not responding well to some `docker kill/rm` command.
`ExecStop` seems to work great in all cases and it's what we've been
using for a very long time, so I'm reverting to that.
v1.50.0 was found to be buggy for people using a `webclient` listener.
This is fixed in v1.50.1.
We don't use such a listener, so we weren't affected anyway.
Also get rid of `--tags=update-user-password` in the
`matrix-dendrite` role, as what we had doesn't work.
We may be able to do it with some Ansible helper or something else.
For now, we'll omit this feature.
