`matrix_synapse_workers_generic_worker_client_server_endpoints` and
`matrix_synapse_workers_generic_worker_federation_endpoints` dropped the
non-matching entries with `difference([none])`. ansible-core implements
`difference` as `list(set(a) - set(b))`, so the result follows Python's
per-process string hash seed and comes out in a different order on every
run.
The reverse-proxy companion renders these lists as nginx `location`
blocks, so its `conf.d/matrix-synapse-reverse-proxy-companion.conf`
changed, and the companion restarted, on every run, even with nothing
else to change.
`reject('none')` drops the same entries and keeps the order of
`matrix_synapse_workers_generic_worker_endpoints`. All these locations
point to the same upstream, so routing is unchanged.
With conditional restart (install-* tags), the homeserver entry for
Synapse fell through to `restart_necessary: true`, and the worker
entries carried no flag at all, which systemd_service_manager also
treats as "restart". Every install-all and install-service synapse
therefore restarted Synapse and all of its workers, even when nothing
had changed. Without workers, systemd also restarted most bridges and
bots, whose units require matrix-synapse.service.
Compute matrix_synapse_restart_necessary from the tasks that install
what Synapse runs: homeserver.yaml, the initial config and signing key,
the log config, labels and systemd service, the container image (pulled,
self-built or customized), each worker's config, labels and systemd
service, and the downloaded or cloned extension modules. Workers follow
the same decision, as they read the same files.
Synapse also mounts the appservice registration files written by other
roles. Add matrix_homeserver_app_service_registration_changed in
group_vars/matrix_servers, which checks the registered result of each
task installing one of the files listed in
matrix_homeserver_container_extra_arguments_auto, so that a changed
registration still restarts Synapse and its workers. The double-puppet
role did not register its result; it does now.
The Synapse image pull and three extension downloads all registered
their result as `result`. The downloads run first, so the pull's result
overwrote theirs before the restart decision could read them; they now
use role-prefixed names.
With Synapse workers enabled, inject_worker.yml appends each worker's
service to devture_systemd_service_manager_services_list_auto with
set_fact. That templates the whole list, including every service's
restart_necessary value, when the matrix-synapse role initializes, and
the resulting fact outranks group_vars. Services whose restart flag is
set after that point are then only started, not restarted, on install-*
runs even when their configuration changed. That includes the
matrix-synapse role's own reverse-proxy companion, Goofys and S3 storage
provider timer, and the services of every later role that sets a restart
flag (Element, Grafana, Prometheus, Traefik and others).
Collect worker services in matrix_synapse_worker_systemd_services_list
and append it to the auto list in group_vars/matrix_servers, like the
other entries there, so the list stays lazy.
Include inject_worker.yml once and loop inside each task instead of
including it once per worker. With 19 workers this cuts the step from
26.2 s to 5.3 s on ansible-core 2.20.1 (controller-only, one run). Also
fix two stream-writer failure messages that referenced the undefined
`item`, and report all missing worker properties in one message.
Containers are created with `--rm`, so Docker removes them asynchronously after they stop.
The pre-start `docker rm` returns immediately while that removal is still in progress
(its error is suppressed), and the following `docker create --name` can fail with
"The container name ... is already in use".
This is easy to hit since systemd_service_manager switched its default restart mode to `all-at-once`:
many containers stop and start at the same time and Docker falls behind on removals.
The playbook then fails at "Restart systemd services needing restart at once",
even though the services recover on their own (`Restart=always`) a little later.
The external roles (MASH, devture) were hardened against this in 2026-03
(see mother-of-all-self-hosting/ansible-role-valkey@0ba79c0), but the roles in `roles/custom` never were.
This ports the same bounded cleanup loop to them: stop, force-remove,
and wait until Docker releases the name, failing explicitly after a timeout.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
