Spec should recommend minimum interval, not enforce. 60 seconds for push, 300 for pull. Implementations can ignore it, but then they're bad neighbors and you drop them.
I have 47 playbooks. One of them is "block abusive peers at nginx layer." Not hard.
I boot everything from the network. My status page would netboot too, obviously. Stateless node, config pulled from TFTP at boot. If it dies, reboot and you're back.
This is compatible with federation. The binary just needs to fit in my initramfs. 12MB is fine.
Same here. Following the thread for when there's something to actually deploy.
Too big. I run Ampere ARM servers. Watts matter. A status page should be under 1MB, no JS, render in lynx if needed. The federation is the point, not the pretty graphs.
I will write an implementation in Rust if someone else does the spec first. Async with tokio, static binary under 800KB.
Member OP
The 12MB includes borrowed upptime UI which is all JS. I agree it's bloated. For my own use I would strip it to a static HTML template. The federation logic is maybe 2MB in go.
But customers want pretty. Tradeoff.
Finally, someone reasonable. My C implementation will output plain HTML, no CSS, readable in anything. The federation protocol is just HTTPS + JSON. The presentation is your problem.
I have not started writing. But I will. This week. Probably.
Member
I have heard this before.
Spec draft now at: https://github.com/YuriDavid/statusth/blob/main/SPEC.md
Name is still bad. Repository name is worse. I will rename before anyone notices.
Current status:
Need testers. Need someone to write a Python implementation so we can interop test. @torontoeng?
I have a 47-line aiohttp sketch that verifies signatures and fetches well-known. Not production. I will clean it up this weekend, probably, and PR an `examples/` directory to your repo.
But I am still not sure about "full incident list exchange." For a peer with years of history, that's megabytes per poll. Delta with sequence numbers is worth the complexity.
Member
You were right. I implemented full list, tested with 2000 incidents, 3MB per poll. Unusable.
Switched to delta: each incident has monotonic `seq` per node. Peer asks `?since_seq=123`. Returns only newer. 200 bytes typical poll.
Spec updated. Breaking change, but nobody else implements yet so who cares.
