This was apparently hosted on the long gone "apollo" server[1], and when
archweb was migrated to a dedicated cloud VM, it was changed to a
redirect to the main site (archlinux.org)[2][3].

It may have made sense at the time, but now four years later there is no
reason for keeping this around.

I guess dev.archlinux.org was something similar to what pkgbuild.com is
today ("Public HTML server" for staff), but only for developers.

[1] f6c3af0e ("Merge branch 'apollo_decomission' into 'master'")
[2] 824fb084 ("tf-stage1/archlinux: Change DNS records for the archweb migration and also increase the machine size")
[3] 9800d023 ("roles/archweb: Create domain redirects for the domains that point to specific archweb sub urls.")

Somehow these changes were not directly applied even though the role
reloads the prometheus config.

Fixes: 10475a62 ("prometheus: Alert if a build hosts is OOM for 12h")
Signed-off-by: Christian Heusel <christian@heusel.eu>

Fixes: 4159a61f ("dbscripts: switch to Git packaging")

Signed-off-by: Christian Heusel <christian@heusel.eu>

There is not much value in knowing when one of our build hosts has no
more memory left as all of them have plenty of swap available.
Additionally these rules trigger quite often even for short spikes.

Signed-off-by: Christian Heusel <christian@heusel.eu>

Link: https://lists.archlinux.org/archives/list/staff@lists.archlinux.org/message/3JNJQJ7O3LBIXST2EFBCAKM2HCHXJZUM/


Signed-off-by: Christian Heusel <christian@heusel.eu>

Signed-off-by: Christian Heusel <christian@heusel.eu>

This was added more than 7 years ago[1] and has not been relevant for a
long time.

[1] d32ce421 ("postfix: Remove compat_maps")

Please see the reject_authenticated_sender_login_mismatch option[1] for
more details.

For now service accounts are not restricted in any way, this should be
improved in the further.

[1] https://www.postfix.org/postconf.5.html#reject_authenticated_sender_login_mismatch

Fix #365

This removes unnecessary parameters, mostly for one of these reasons:
the value is already the default value, the default value is good
enough, or the parameter is not used in our case.

A bit of reordering/"tidying" was also done.

I think it was added to improve the mail reputation (avoid being
filtered as spam), but at lot has changed since it was added (+5 years
ago), so let's remove it.

From chrony FAQ[1]:
"1.2. Should I prefer chrony over timesyncd if I do not need to run a
server?

Generally, yes.

systemd-timesyncd is a very simple NTP client included in the systemd
suite. It lacks almost all features of chrony and other advanced client
implementations listed on the comparison page. One of its main
limitations is that it cannot poll multiple servers at the same time and
detect servers having incorrect time (falsetickers in the NTP
terminology). It should be used only with trusted reliable servers,
ideally in local network.

Using timesyncd with pool.ntp.org is problematic. The pool is very
robust as a whole, but the individual servers run by volunteers cannot
be relied on. Occasionally, servers drift away or make a step to distant
past or future due to misconfiguration, problematic implementation, and
other bugs (e.g. in firmware of a GPS receiver). The pool monitoring
system detects such servers and quickly removes them from the pool DNS,
but clients like timesyncd cannot recover from that. They follow the
server as long as it claims to be synchronised. They need to be
restarted in order to get a new address from the pool DNS.

Note that the complexity of NTP and clock synchronisation is on the
client side. The amount of code in chrony specific to NTP server is very
small and it is disabled by default. If it was removed, it would not
significantly reduce the amount of memory or storage needed."

This commit fixes the issue by switching to a proper NTP client
(chrony), trustworthy time sources from Netnod and
Physikalisch-Technische Bundesanstalt which distributes the official
time for Sweden[2] and Germany[3] respectively, and finally NTS is used
to protect against MITM attacks.

Since most of our servers are in Germany or Finland (close to Sweden),
it makes sense to use these time sources as a low round-trip delay[4] is
preferred for NTP. For the few servers[5] we have outside Europe, the
root delay[4] will be higher than desired, but with the current use-case
for these servers, it should not be a problem.

[1] https://chrony-project.org/faq.html#_should_i_prefer_chrony_over_timesyncd_if_i_do_not_need_to_run_a_server
[2] https://www.netnod.se/swedish-distributed-time-service
[3] https://www.ptb.de/cms/en/ptb/fachabteilungen/abt4/fb-44/ag-442/dissemination-of-legal-time.html
[4] https://blog.meinbergglobal.com/2021/02/25/the-root-of-all-timing-understanding-root-delay-and-root-dispersion-in-ntp/
[5] {america,asia,sydney}.mirror.pkgbuild.com

Robin Candau authored 3 months ago and Robin Candau committed 3 months ago

See https://github.com/ansible/ansible/pull/77644

It has three panels showing "Cache hit ratio", "HTTP version" and "TLS
version".

The metrics are generated with Loki recording rules (see the previous
three commits).

This enables us to calculate the cache hit ratio, which may help
determine whether more caching would be beneficial.

Please note that this only counts requests for which caching is enabled
(e.g. {fastcgi,proxy}_cache is configured), e.g. for static served files
cache_status will be "".

[1] http://nginx.org/en/docs/http/ngx_http_upstream_module.html#var_upstream_cache_status

Mainly because we are curious. The data may also be used to decide if we
want to drop older versions of TLS.

The plan is to use this for creating metrics from the nginx log lines
(e.g. requests per second).

[1] https://grafana.com/docs/loki/latest/alert/#recording-rules

Fixes: bd19c007 ("Add configuration to retain prometheus data for 1 year")

With the 11.6.2 release mariadb has made snapshot isolation a default
setting which recently caused issues on the AUR and the Forums, where
the respective database engines would fail due to the table changing
since the last time they read it:

    PHP Fatal error:  Uncaught mysqli_sql_exception: Record has changed since last read in table 'fluxbb_online' in /srv/http/fluxbb/include/dblayer/mysqli_innodb.php:79
    Stack trace:
    #0 /srv/http/fluxbb/include/dblayer/mysqli_innodb.php(79): mysqli_query()
    #1 /srv/http/fluxbb/include/functions.php(485): DBLayer->query()
    #2 /srv/http/fluxbb/include/common.php(162): update_users_online()
    #3 /srv/http/fluxbb/viewtopic.php(10): require('...')
    #4 {main}

We therefore introduce a config switch to restore the old behavior and
apply the changed setting on the two services.

Link: https://mariadb.com/kb/en/mariadb-11-6-2-release-notes/#innodb
Link: https://bbs.archlinux.org/viewtopic.php?id=301802
Link: aurweb#525


Signed-off-by: Christian Heusel <christian@heusel.eu>

The project membership must also be extended, if not the user is simply
deleted when the membership expires (defeating the purpose of extending
the access tokens).

Fixes: 639101e6 ("gitlab: Add ruby script for continuous extending of bot tokens")

If the cost exceeds $0, it indicates that we have run out of credit
and/or are doing something wrong, in either case we want to be alerted.

With the support for network.wireguard.* credentials[1] in systemd
v256[2], we can now easily avoid storing the credentials centrally in
our ansible vault, which is preferable as it makes the private keys less
exposed. It may also make fine-grained access easier in the future[3] as
there is no longer a vault file for each server.

All the keys have been rotated and the new private keys are only stored
on the servers.

[1] https://github.com/systemd/systemd/pull/30826
[2] https://github.com/systemd/systemd/releases/tag/v256
[3] #64

There is no technical reason for this at the moment, but UEFI is the de
facto firmware for x86-64, so let's be modern.

This should not change anything as the VMs are short-lived (15 minutes
at the most), so it is just added for good measure.

From the kernel patch series[1]:
"This series provides an asynchronous means of reporting free guest
pages to a hypervisor so that the memory associated with those pages can
be dropped and reused by other processes and/or guests on the host.
Using this it is possible to avoid unnecessary I/O to disk and greatly
improve performance in the case of memory overcommit on the host."[1]

The runner hosts may be memory overcommitted if there is too many
running VMs and containers at the same time, which this should help to
avoid.

[1] https://lore.kernel.org/linux-mm/20200211224416.29318.44077.stgit@localhost.localdomain/

It makes more sense to build the image in arch-boxes than building it on
each runner, especially considering that arch-boxes already have all the
necessary infrastructure, so we can avoid maintaining similar code in
two repositories and avoid running losetup, mount, arch-chroot etc. (as
root) on the runners.

The arch-boxes MR[1] has a little more context.

[1] arch-boxes!200

This reverts commit 466230e4.

This has been fixed in pacman[1], so it is no longer unreasonably slow.
Some quick testing at runner1 indicates that this only saves five
seconds at best, so IMO it is not worth the complexity to continue doing
this.

This revert does not revert the timeout back to 60 seconds, but keeps it
at 30 seconds.

[1] pacman/pacman!16

This means that there is no need to make runner-specific changes to the
image, so in theory the image could be build centrally (e.g. in the
arch-boxes project[1]) and then distributed to the runner hosts.

This change also make the SSH keys ephemeral.

[1] https://gitlab.archlinux.org/archlinux/arch-boxes

All libvirt volume management is now handled through virsh instead of
direct file system access. As a volume cannot be uploaded in an atomic
way, the current active volume is now tracked in a file on disk.

This may allow us to run the script with less privileges and use polkit
for libvirt access control[1].

[1] https://libvirt.org/aclpolkit.html

The prepare stage runs "echo "Running on $(hostname)...""[1], resulting
in "bash: line 7: hostname: command not found" and it outputting
"Running on ..." as the hostname command is provided by inetutils, which
is not installed.

Fix it by "monkey patching" it to use "hostnamectl hostname" and inject
the hostname with SMBIOS[2][3]. Injecting creds with SMBIOS may also be
useful in the future, e.g. for injecting an ephemeral SSH public key.

[1] https://gitlab.com/gitlab-org/gitlab-runner/-/blob/v17.5.2/shells/bash.go?ref_type=tags#L452-L456
[2] https://systemd.io/CREDENTIALS/
[3] https://github.com/systemd/systemd/pull/30814

This removes 13 instances of [1] and 1 instance of the IP address from
the job log.

The latter was fixed by no longer waiting for SSH in the "run" stage,
which is unnecessary as we wait for SSH in the "prepare" stage.

[1] Warning: Permanently added '192.168.122.xxx' (ED25519) to the list of known hosts.

It was forgotten once[1] to update it in both places, so avoid that
issue in the future, by moving it to a variable.

[1] c370c9d0 ("gitlab_runner: Update concurreny math to reflect the new VM size")

The project now has a centralized landing page hosted in the root
directory which we can observe instead of the docs for one specific
crate.

Related to signstar#124
Related to signstar!131



Signed-off-by: Christian Heusel <christian@heusel.eu>

The project now has a centralized landing page hosted in the root
directory which we can observe instead of the docs for one specific
crate.

Related to archlinux/alpm/alpm#76
Related to archlinux/alpm/alpm!57



Signed-off-by: Christian Heusel <christian@heusel.eu>