Just wait until you defeat Antithesis 😈

RSVP on Meetup: www.meetup.com/papers-we-lo...

Big thanks to @antithesis.com for sponsoring the meetup - after getting to know everyone there, I can't think of a company that loves papers more ❤️

I think Buck2 motivated Meta to write a Starlark implementation in Rust, so it may be just a crate away?

Starlark is my favorite. It’s a real programming language, with the most egregious bad-for-config footguns removed.

The 2014 re:Invent talk from Al Vermeulen and Swami Sivasubramaniam that sparked all this: www.youtube.com/watch?v=QVvF...

And don't even start with Kafka. It's a much higher-level component, the transaction model is painfully under-documented (but kudos to Jack Vanlightly for improving the situation!), and the implementation was fatally flawed until at least Kafka 4.0. issues.apache.org/jira/browse/...

At Seattle Systems, Marc Brooker said that AWS's journal uses a variant of chain replication within regions, and multi-Paxos between regions. Sensible!

Are premium, low-volume brands (Riese & Muller, Stromer, etc.) more financially viable?

Mostly motivated by Google’s decision to have proto3 drop support for features that their own internal systems depend on, plus some optimizations for C++ gencode. More context: buf.build/blog/protobu...

I'm still wondering what the observability best practices are for gossip-based systems. In my experience, observability is especially frustrating when the system is misbehaving and nodes have different views of the world.

There's also a nice shout-out for @antithesis.com! I'm biased, but I think it's pretty sweet and you should check it out for your next commercially-important distributed system.

Gmail could crush this if it:

- Delivered to other ppl within the company instantly
- Displayed intra-company threads like iMessage by default - until some buzzkill sends more than a few sentences

Google has tried 9 zillion chat apps but AFAIK hasn’t let Gmail evolve. (RIP Inbox.)

Wow, that’s…impressive dedication. Feels like a massive miss for the platform that workflows like this are necessary though.

Slack’s default behavior of equating “I’ve seen this” with “I’m done with this, pls consign it to the void” is maddening as soon as anything serious gets discussed.

The only decentralized system like this I've ever operated was an operational nightmare. During outages, each node had a different view of the cluster. Reassembling some global notion of cluster state took obscene observability spend.

Link to the paper: dl.acm.org/doi/pdf/10.1...

We have Hurd your complaints and are working on a free upgrade.