Fixes a regression introduced in f6097fbba1fb, which was cauing Synapse
to die with this error message:
> ValueError: sender_localpart needs characters which are not URL encoded.
These are just defensive cleanup tasks that we run.
In the good case, there's nothing to kill or remove, so they trigger an
error like this:
> Error response from daemon: Cannot kill container: something: No such container: something
and:
> Error: No such container: something
People often ask us if this is a problem, so instead of always having to
answer with "no, this is to be expected", we'd rather eliminate it now
and make logs cleaner.
In the event that:
- a container is really stuck and needs cleanup using kill/rm
- and cleanup fails, and we fail to report it because of error
suppression (`2>/dev/null`)
.. we'd still get an error when launching ("container name already in use .."),
so it shouldn't be too hard to investigate.
Not specifying bind addresses for the worker resulted in this warning:
> synapse.app - 47 - WARNING - None - Failed to listen on 0.0.0.0, continuing because listening on [::]
Additionally, metrics listening only on 127.0.0.1 seems like a no-op.
Only having it accessible from within the container is likely not what
we intend. Changed that to all interfaces as well.
Whether it actually gets exposed or not depends on the systemd service
and `matrix_synapse_workers_container_host_bind_address`.
This switches the `docker exec` method of spawning
Synapse workers inside the `matrix-synapse` container with
dedicated containers for each worker.
We also have dedicated systemd services for each worker,
so this are now:
- more consistent with everything else (we don't use systemd
instantiated services anywhere)
- we don't need the "parse systemd instance name into worker name +
port" part
- we don't need to keep track of PIDs manually
- we don't need jq (less depenendencies)
- workers dying would be restarted by systemd correctly, like any other
service
- `docker ps` shows each worker separately and we can observe resource
usage
We do this by creating one more layer of indirection.
First we reach some generic vhost handling matrix.DOMAIN.
A bunch of override rules are added there (capturing traffic to send to
ma1sd, etc). nginx-status and similar generic things also live there.
We then proxy to the homeserver on some other vhost (only Synapse being
available right now, but repointing this to Dendrite or other will be
possible in the future).
Then that homeserver-specific vhost does its thing to proxy to the
homeserver. It may or may not use workers, etc.
Without matrix-corporal, the flow is now:
1. matrix.DOMAIN (matrix-nginx-proxy/matrix-domain.conf)
2. matrix-nginx-proxy/matrix-synapse.conf
3. matrix-synapse
With matrix-corporal enabled, it becomes:
1. matrix.DOMAIN (matrix-nginx-proxy/matrix-domain.conf)
2. matrix-corporal
3. matrix-nginx-proxy/matrix-synapse.conf
4. matrix-synapse
(matrix-corporal gets injected at step 2).
This removes some `multi-target.wants` symlinks as well, etc.
But despite systemd saying:
> Removed symlink /etc/systemd/system/matrix-synapse.service.wants/matrix-synapse-worker@appservice:0.service
.. I still see such symlinks tehre for me for some reason, so keeping the
code (below) to find & delete them still seems like a good idea.
There was a `matrix_nginx_proxy_enabled|default(False)` check, but:
- it didn't seem to work reliably for some reason (hmm)
- referring to a `matrix_nginx_proxy_*` variable from within the
`matrix-synapse` role is not ideal
- exposing always happened on `127.0.0.1`, which may not be good enough
for some rarer setups (where the own webserver is external to the host)
I guess it didn't hurt to do it until now, but it's not great serving
federation APIs on the client-server API port, etc.
matrix-corporal doesn't work yet (still something to be solved in the
future), but its firewalling operations will also be sabotaged
by Client-Server APIs being served on the federation port (it's a way to get around its firewalling).
If we load it at runtime, during matrix-synapse role execution,
it's good enough for matrix-synapse and all roles after that,
but.. it breaks when someone uses `--tags=setup-nginx-proxy` alone.
The downside of including this vars file like this in `setup.yml`
is that the variables contained in it cannot be overriden by the user
(in their inventory's `vars.yml`).
... but it's not like overriding these variables was possible anyway
when including them at runtime.
Some people run Coturn or Jitsi, etc., by themselves and disable it
in the playbook.
Because the playbook is trying to be nice and clean up after itself,
it was deleting these Docker images.
However, people wish to pull and use them separately and would rather
they don't get deleted.
We could make this configurable for the sake of this special case, but
it's simpler to just avoid deleting these images.
It's not like this "cleaning things up" thing works anyway.
As time goes on, the playbook gets updated with newer image tags
and we leave so many images behind. If one doesn't run
`docker system prune -a` manually once in a while, they'd get swamped
with images anyway. Whether we leave a few images behind due to the lack
of this cleanup now is pretty much irrelevant.
We log everything in systemd/journald for every service already,
so there's no need for double-logging, bridges rotating log files
manually and other such nonsense.
In short, this makes Synapse a 2nd class citizen,
preparing for a future where it's just one-of-many homeserver software
options.
We also no longer have a default Postgres superuser password,
which improves security.
The changelog explains more as to why this was done
and how to proceed from here.
I had intentionally held it back in 39ea3496a4
until:
- it received more testing (there were a few bugs during the
migration, but now it seems OK)
- this migration guide was written
While administering we will occasionally invoke this script interactively with the "non-interactive" switch still there, yet still sit at the desk waiting for 300 seconds for this timer to run out.
The systemd-timer already uses a 3h randomized delay for automatic renewals, which serves this purpose well.
The `mobile` branch got merged to `master`, which ends up becoming
`:latest`. It's a "rewrite" of the bridge's backend and only
supports a Postgres database.
We'd like to go back (well, forward) to `:latest`, but that will take
a little longer, because:
- we need to handle and document things for people still on SQLite
(especially those with external Postgres, who are likely on SQLite for
bridges)
- I'd rather test the new builds (and migration) a bit before
releasing it to others and possibly breaking their bridge
Brave ones who are already using the bridge with Postgres
can jump on `:latest` and report their experience.
Slavi Pantaleev
Marcel Partap
Panagiotis Georgiadis
throwawayay
pushytoxin