#287
Mar 17, 2024
Books
No books this week! I started reading Perdido Street Station but it is very long,
Roleplaying Games
The Halls of Arden Vul
We had a really great session this week. Not so much action, but a lot of discovery.
Firstly, we have the B-team, those unfortunates who had ventured down to the level below and got attacked by some adventurer-bandits. They turned the situation around by expending one charge of a very valuable, and irreplaceable, magic item. And then the character who got a cursed torch stuck to their hand decided to take a few days off to go back to town and get it removed. I’m fine with this. In a sense, they solved the problem with ease, but they used up some valuable resources to do so—I can imagine them wishing they still had that charge in the future.
Then, we have the rest of the party, who stayed up at the Inn of the Lost to wait for a message from the Lord of the Arena (or, more likely, one of his lieutenants). They passed a couple of days without much luck then, by sheer coincidence, they encountered the Lord while wandering around the Arena!
They begged a moment of his time, explained that they had found what appeared to be high-tech “pods” with more of his kind sealed away in them, and his only question was “do you think this could help the Arena in some way?”
He just did not care at all. The players were floored.
But, pulling back the curtain a little, it is totally in character. The Lord of the Arena is around 3000 years old, preserved by technologies that were never intended to keep someone alive so long. He can’t remember anything of his origins—can’t even remember his own name—all he knows, all he cares about, is the Arena.
The Lord did mention in passing one very interesting tidbit to them, which was that another of his kind, Lord Psalor Ki, lives above and often visits. He said that Psalor Ki had a great interest in the functioning of living creatures. Asking around the Inn of the Lost the players learned that Lord Psalor Ki is widely known by another title—The Vivisectionist—and to run if they ever encounter him outside one of the dungeon’s neutral zones.
Then, we went back to exploring. The players have realised that the Arena is kind of a travel hub, with easy access to the domains of several of the dungeon factions, and they chose to head along a long and twisting series of coridors and stairs leading up to the precincts of the Cult of Set, which will be our focus for the next session.
But all this exploration is really just killing time until Tresti, the party’s magic-user, recovers from her wounds and they can go back to the high-tech pod room armed with a Comprehend Languages spell. I’m looking forward to it.
Miscellaneous
Another good week for thing-doer—now on GitHub! The highlights are:
- An API server, so pods can create new pods (and other types of resources)—and, excitingly, pods can be killed via the API!
- A script to configure the network, automating the previous manual process, making it much quicker to spin up a test cluster. But this pales in comparison to:
- A development VM! So now I can just run
nix build .#vmto generate a qemu VM with etcd, flannel, and thing-doer already installed, and with systemd units to start up the thing-doer services. The VM connects to a bridge interface on the host machine with a random MAC address, so you can just launch multiple instances of the VM to easily test multi-host clusters.
There have been a lot of behind-the-scenes improvements too:
- A lot of refactoring to reduce duplicated code, cut down on unnecessary error handling, simplify things, and so on.
- Changing reaperd and schedulerd to use a single connection to etcd for watching cluster state changes, rather than one connection per key prefix. workerd will come next. This reduces load on etcd, and so makes the cluster scale better.
- Rearranging and renaming the various command-line arguments to improve consistency, allow specifying everything through environment variables, and also allow specifying IP addresses through files (useful for systemd units).
I’ve started looking into integration testing too. I have a basic one right now that builds the development VM and checks that all the services can start up and are visible through the API server, but I’d like to add at least one for testing the container networking in a multi-host cluster.
The NixOS VM tooling is quite impressive.
I think the next big milestone will be implementing some analogue of kubernetes deployments / replicasets, which I hope to implement almost entirely on top of pods running in the cluster—rather than new features in the cluster servers. Off the top of my head, I’ll need to add API endpoints to manage DNS records and to watch resource changes, and I hope that’s all.