As python-tensorflow-* kills our rebuilderd-workers as they are usually
queue at the same time.

limit port 25 on apollo to mail.archlinux.org only

See merge request !162

Frederik Schwan authored Dec 02, 2020 and Jelle van der Waa committed Dec 12, 2020

While apollo hosts patchwork it needs to receive mail for
patchwork@archlinux.org. Those mails are forwarded from
mail.archlinux.org. This implies apollo being configures for the
archlinux.org domain. Since Patchwork is not maintained anymore,
this is a quick fix to prevent sending of forged mails via
apollo.

Split archive role into archive_web for archive-mirrors

See merge request !175

Jelle van der Waa authored Dec 12, 2020 and Jelle van der Waa committed Dec 12, 2020

To simplify the archive role, split it up in the web serving part for
the archive-mirrors, gemini and keep the archive role for only the
archive operation. This simplifies the new role as only two lines are
required to setup the the archive mirror website.

Add ipv6 addresses to archive mirrors for sync whitelisting

See merge request !177

Add archive mirror DNS entries

See merge request !174

As we want to serve mirros and the archive add a new domain for the
archive so the mirror and archive can be hosted.

Remove old repro2 hostvars

See merge request !172

Create archlinux.org server

See merge request !165

Giancarlo Razzolini authored Dec 04, 2020 and Sven-Hendrik Haase committed Dec 12, 2020

Lower the ttl of master-key, dev, packages and ipxe

Giancarlo Razzolini authored Dec 04, 2020 and Sven-Hendrik Haase committed Dec 12, 2020

tf-stage1/archlinux.tf: Lower TTL for archlinux.org and www.archlinux.org and create a new server for archlinux.org

For the archweb migration, we need to lower the TTL and create a new archlinux.org server to host the site.

Seems we run at about 5GB of cache. Let postgres assume it gets 4GB.

Element uses a statically configured list of homeservers with disabled
presence. Without hosting our own Element this makes for pretty poor UI.

This reverts commit 82ac30f9.

Only changes to comments.

postgres: Set `jit = off` on matrix.archlinux.org

See merge request !169

Bump the mem settings higher again, now that we know they're not the
cause. Also increase the maximum connection count.

This seems to be the cause for our memory leak.

https://www.postgresql.org/message-id/flat/16707-f5df308978a55bf8%40postgresql.org

Mostly changes to comments.

Add Kape server IPv6 addresses (fixes #230)

Closes #230

See merge request !173

Restrict Grafana access to Arch Linux Staff group on Keycloak (fixes #151)

Closes #151

See merge request !171

Fix typo in docs/kape.md

See merge request !170

The hosts file contains repro2.pkgbuild.com in the `[kape_servers]`
group.

This process didn't need any source changes but it added the new Terraform lockfiles.

Still seeing PostgreSQL session processes eating over 600M of private
memory. What's going on?

Add Kape donated servers

See merge request !166

Setup Kape servers as archive mirrors (asia,europe,america), Gitlab
runner and Rebuilderd worker. All machines except runner1 are EFI
machines with grub setup and a EFI parition which is not supported by
our ansible install role and is manually rolled out.

Build artifcats are limited to 10MB per piece and we have
logs/diffoscope and json can become a little over 20M.

Allow the log size to be limited to 10 MiB so we don't hit max body size
limits when posting results.

Due to the massive backlog of packages, disable [testing] until
everything is back to being stable. Also disable long during failing to
build packags such as pytorch and kubernetes

rebuilderd requires an X-Real-IP to be set if configured.