This was causing worker_listener blocks to get generated with
Nonetype resources, crashing things. This was only a problem for
worker types that don't have any configured resources.

If we're supporting multiple instances of individual worker types, then having
a 'run all workers' mode doesn't make too much sense

This required making the shared config a dictionary instead of a large str, so
we could manipulate it easier. We do eventually convert it to a str before writing
though of course.

…apse_worker_docker

The default was just an abstraction for not much gain. Instead we can
just set them explicitly for each worker.

This commit also fixes an issue with how we append to the instance map,
and a log line substitution.

Synapse should enforce this, not us.

… to do so

This required some changes to the way we handle our nginx config. It's
now a dictionary instead of just a str. This was because endpoint
mapping blocks would get duplicated when we configured multiple
instances of the same worker. Now that the nginx config is organised by
a map of endpoints to upstreams, those endpoints de-duplicate
automatically.

Speaking of upstreams, we now handle creating upstreams which load
balance between processes. To route traffic to an upstream, you simply
need to proxy_pass to its name. We use a separate dict to keep track of
upstreams and where they point to. These upstreams do round-robin LBing.

Note that we don't really want other load-balancing types for our tests. For instance,
balancing by request IP makes no sense as the tests will be sending requests
all the from the same IP.

What we really want is for the load-balancing to be deterministic, which
round-robin is (as long as the tests remain constant themselves).

Turns out Complement may run your ENTRYPOINT script multiple times - who
knew!

The script appends to config files, which can break things if the same
content is appended twice (I only noticed this once we started adding
upstreams to the nginx config. You can't have two upstreams with the
same name - but apparently you can have two identical server blocks!)

We could try to be clever and make the script work with being run
multiple times, but that's a maintanence burden. I instead opted to just
place a file down if the script had already been run, and refuse to run
again if that file exists.

…apse_worker_docker

Note that this environment variable only affects writing worker
logs to disk, but it does not affects the main process's log config.

…apse_worker_docker

Co-authored-by: Erik Johnston <erik@matrix.org>

Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>

Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>

…apse_worker_docker

…nt testing

…apse_worker_docker

As we already have a SYNAPSE_WORKER env var.

…apse_worker_docker