Skip to content
Flux
FluxOS 8.24: A Front Door for New Apps

FluxOS 8.24: A Front Door for New Apps

Fluxers! FluxOS 8.24 is out, a week after 8.23 put node moves on the chain’s own record. This release builds a proper front door for new apps, makes every node’s view of the chain faster to build and quicker to follow, and keeps apps on steady machines by holding back nodes that keep failing.

A front door for new apps

When someone registers or updates an app, the request travels the network before its payment lands on chain. In 8.24, that first step goes through the nodes built to guard it: ArcaneOS nodes, talking to each other over authenticated links.

  • Authenticated links. Two ArcaneOS nodes prove to each other who they are with a handshake tied to their node keys and addresses. Every message on the link then carries a check that rules out tampering and replays.
  • Agreed by a quorum. A new registration or update is taken in once three ArcaneOS nodes on different addresses hold it, within ten seconds. Then it spreads to the whole network.
  • Fair allowances. Each client and each admitting operator can have a set number of unpaid requests open at once, set in the network’s signed policy. Retrying the same request simply succeeds.
  • Paid is always welcome. A request whose payment is on chain is never refused and never counted against anyone.

Nodes not on ArcaneOS keep storing, relaying and serving these requests, and now leave taking new ones to ArcaneOS nodes. If you register apps through a node’s API yourself, point it at an ArcaneOS node.

A faster start, and a closer watch on the chain

Every node keeps its own index of app payments on the chain, built by its explorer. In 8.24:

  • Every node builds it the fast way. A new explorer builds from the Flux daemon’s address index in minutes, retries any call that fails on its own, and never falls back to reading the chain block by block from the start.
  • Reorganisations seen at once. The explorer follows the chain by each block’s parent, so a reorganisation shows up at the very next block. A payment in a replaced block keeps counting as paid while it waits to be mined again.
  • Catching up in bulk. A node returning after time away fetches every app message it missed first, then verifies them. That way it quickly hears every renewal and cancellation made while it was gone.

Steady machines for every app

An app deserves a machine that stays up. A healthy node may blip once in a while. In a sample of 28 ArcaneOS nodes, 24 had no episodes at all in a day. 8.24 keeps that occasional blip harmless and gives a node that keeps failing time to recover before it takes new apps.

  • Held longer each time. Within 24 hours, a first or second failure keeps the node from taking new apps for an hour once it is back. A third holds it out of service for 6 hours, and a fourth for 24. Rules a node is under on purpose, such as a version floor, never count as failures.
  • Port checks that say what they found. When a port test fails, the node now says exactly how: no answer over UDP, no answer over TCP, or an answer from something other than the node. Operators know what to fix right away.
  • Silence is not a verdict. When a node asks its peers whether they can reach it, a peer that doesn’t answer at all no longer counts against it.
  • One port, one user. An app install, its port test and the node’s own checks each claim a port before they use it, so they never collide.

Apps where they belong

  • Regions that stay true. An app pinned to a region now follows corrections to the network’s location table. A node the table places outside the region hands the app back, one holder at a time, and the network places it on a node inside the region. The handback rules are modelled in TLA+ and tested on real file sync.
  • A node that moves keeps its place. A node that changes address carries its app records with it and announces itself once at the new address. The network’s view of where apps run is also five to six times faster to build.
  • Commands that reach every instance. A restart, redeploy or removal sent from the dashboard now acts on every instance of the app. That includes the one on the node you are connected through, even behind a home router.

For node operators

  • The address index is required. FluxOS reads the Flux daemon’s address index, so a node needs insightexplorer=1 in flux.conf. A node without it says so clearly through its API and stays out of service until it is set.
  • Hardened inputs. Request bodies, lookups, backup and restore, and peer moves all have stricter input and path handling.

Tested like infrastructure

The 8.24 unit suite counts more than 12,700 tests, and its release gate runs more than 300 integration suites that boot real nodes in a test network.

Get it

FluxOS updates itself on every node, and 8.24 is rolling out across the network now. The full list of changes is in the FluxOS releases on GitHub, and our guide How to Run a FluxNode in 2026 covers the rest.


Posted in Product Updates

by RunonFlux

Tags:

Comments

Leave a Reply

You must be logged in to post a comment.