Fix #177

Giancarlo Razzolini authored 4 years ago and Jelle van der Waa committed 4 years ago

Change the archweb_db_host variable to point to archlinux.org instead of
apollo. This is of particular importance to gemini.

Zabbix has been replaced by Prometheus for monitoring our services.

sch_fq used to be required by BBR but this is no longer the case after
a fallback to internal pacing was implemented in Linux 4.13. Plus, BBR
was recently disabled on Arch's infra.

Now, we could just set the default qdisc to fq_codel for all machines,
but it seems the default for multiqueue network interfaces is sch_mq +
fq_codel sub-qdiscs (one per tx queue).

I am not sure how to cleanly implement the above multiqueue distinction
in Ansible, so drop the default_qdisc configuration. I will then adjust
affected machines manually to remove any configured qdiscs.

BBR behaves badly when it is not the sole connection. It slows down
other streams (bbr and cubic) and generally doesn't scale well when
deployed widely. Let's disable it so we don't make the internet for
others worse.

https://ripe76.ripe.net/presentations/10-2018-05-15-bbr.pdf

https://kanboard.archlinux.org/public/task/117/7dd7510424e4229247e8e0b90bf43e1553fce86cdf8475b60edc956ed5a8



Signed-off-by: Florian Pritz <bluewind@xinu.at>

This currently deploys the same configuration we used to have apart from
some '127.0.0.1' IPs for the agent IP, but those were incorrect anyways.

Signed-off-by: Florian Pritz <bluewind@xinu.at>

This improves performance for high latency connections like Germany to
India and our machines should have enough memory to support quite a few
connections even with such large windows.

Signed-off-by: Florian Pritz <bluewind@xinu.at>

Signed-off-by: Florian Pritz <bluewind@xinu.at>