this copies the current mattermost package definition from upstream
nixpkgs into our repo as-is (that definition itself being a modified
version of our definition that I upstreamed recently).
Since apparently no one else is maintaining the nixpkgs package and I am
apparently maintaining a mattermost package mostly on my own anyways,
this should make upstreaming future changes easier.
This jumps Mattermost ESR Versions (see [1] for their release cycle). The
new version makes use of Go's workspace feature, which unfortunately the
buildGoModule function does not (yet?) support [2], and unfortunately this
breaks the previous build process for mattermost.
Further, the new release also makes use of private modules only included
in the (non-free) enterprise version of mattermost which makes it impossible
to build in the usual way even outside of nixpkgs's build abstractions [3].
Both issues can be solved by using Go 1.22, which has added support for
vendoring when using workspaces, and instructing it to ignore errors with
the -e flag. This requires overriding the go-modules derivation's buildPhase.
Finally, this now also build the commands/mmctl subpackage, which contains
a cli utility to administrate mattermost. This currently has its own nixpkgs
package for no reason i can see at all (it also has a version mismatch
between nixpkgs's mattermost and nixpkgs's mmctl).
[1] https://docs.mattermost.com/upgrade/extended-support-release.html
[2] https://github.com/NixOS/nixpkgs/issues/203039
[3] https://github.com/mattermost/mattermost/issues/26221
the bind mount module has been tweaked in a couple ways:
- rename hexchen.* to hacc.*
- rename bindmount to bindMount to make it consistent with usage in
the nixpkgs container module
- add a hacc.bindToPersist option as shorthand for prepending /perist
to a path via bind mount
the nopersist module has been shortened a little by moving
service-specific things which are used once out into the individual
service files, and removing those which we don't need at all (this also
means we get to loose a mkForce or two in case of mismatches between
hexchen's and our current config).
we decided to:
- get rid of unused packages
- simpify the directory layout since we only have one host anyways
- move our docs (such as they are) in-tree
move some options (the nopersist & container profiles + allowUnfree
packages) into the evalConfig used for containers, so we don't have to
repeat ourselves as much.
also removed some no-longer-needed specialArgs.
also made thelounge work with nopersist, which for some reason it didn't
use before.
This reverts commit 90f4971e88d22da6b2a213bbeb1790f456024b36, and resets
the uffd version to the one we are already using, in hopes of making the
update slightly less painfull (haha).
in theory this might be ready to deploy. Potential hazards & things to
know when actually doing so:
1. the mysql version used by mattermost was updated (the old uses an
openssl which is marked insecure). Might have to migrate a database
2. lots of settings now use RFC 42-style settings, which might contain
new typos
3. this updates uffd (& changes the patches we apply). Since version
dependencies of uffd are basically "whatever debian has" we have
never bothered to match them, but afaik have also never updated uffd
since the initial deploy some years ago. No guarantee it still
works.
4. tracktrain depends on haskellPackages.conferer-warp, which is
currently marked broken. There is no reason for this (it builds
fine). Until fixed upstream, build with NIXPKGS_ALLOW_BROKEN=1.
cf. https://github.com/NixOS/nixpkgs/pull/234784; waiting for a
merge of haskell-updates into 23.05
this is currently deployed and appears to be working. please everyone
have a look at it & then decide if we want to use this for the other
secrets as well.
apparently the 7.1.x series is now old enough that even though it
does still get security fixes, the mattermost team no longer mentions
this on their blog, so we missed out on a couple. fun!
upsides:
- we will no longer get confused about which state is currently deployed
downsides:
- deploys get slower, since it has to uploads the entire haccfiles each time
today i woke up to the realisation that there's an extremely obvious way
to make these nicer, & then i did exactly that. For some reason I did
not think of this when originally removing the dependency to nix-hexchen's
evalConfig.
unfortunately, this is not /quite/ a no-op. The only actual change is
different whitespace in some of the semantically-equivalent
coredns-configs that got unified.
