jschoubben a1d98c2438 mesh-provisioner: judge no retirement from a partial picture (issue 940)
At 2026-11-02 22:55:58 the provisioner's first pass after a new membership read
part of the picture, took the modules missing from it as handing nothing, set
21 route keys of running web apps retiring and asked the operator to remove
them; 3 s later the next pass put every plan back. Only 3 asks reached the
router; the other 18 were refused yet counted asked. Third of its kind (937,
1332), so the kind is fixed, not the instance.

Quick fix: wait for the watched reads' first load before any pass.
Thorough fix, chosen because the kind recurred three times and its obvious
answer would cut running apps off their routes:
- no pass judges before placement, wants, haves, the indexer's modules and the
  published compositions have loaded once; a read error fails the pass
  instead of reading as empty;
- while a module's composition is not published at its build, no key on that
  node turns retiring and no plan is removed; a placed module with no known
  build is held, not counted as handing nothing;
- a key whose consumer still has a want or a have on the node and no plan read
  in this pass is left as it is;
- a pass that would retire most of a node's consumers at once (more than 3, or
  more than half and at least 2) retires and asks nothing and says it as a
  fault through provisioner_health (runtime-health, so a condition names the
  likely cause); provisioner_compose retire-many is the operator's word that
  such a removal is meant;
- at most 3 asks stand open (the router's bound per asker); the rest wait and
  are raised as one ends, and an ask the router holds no record of a minute on
  is raised again.

Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
2026-11-02 23:17:07 +01:00

mesh-catalog

The Novox Mesh catalogue. The modules the mesh builds, provisions and runs — as manifests, one per module under modules/.

This is data, not a control-plane concern. The manifests describe what a module is: what it provides, what it requires, the seats it claims, the resources the host applies for it. The engine that reads them — parsing, eligibility resolution, sealing, declaration emission — lives in the control plane (novox/mesh-controller, internal/catalogue), which consumes this repository as a build source. The host (novox/mesh-host) applies the declarations the control plane emits. Neither is here.

What a module is, and is not

A module is one thing the mesh can run, named once, described completely by its manifest. A manifest names its image (pinned by digest), the resources the host owns for it (directories, files, the container, the private network it joins), what it requires from a provider and what it provides to consumers, and the sealed secrets it needs filled on the machine.

  • Core mesh components are not modules. The node host, the foundation, the control-plane contexts and the surfaces are the mesh itself; they ship as their own repositories (mesh-host, mesh-foundation, mesh-controller, mesh-surfaces, mesh-sdk), not from here.
  • Standalone applications are not here either. A larger application lives in its own repository with its manifest at the root, registered with the mesh as a build source (novox/hq ADR 0010). This repository holds the modules the mesh maintains as its shared catalogue; an application the mesh merely hosts keeps its manifest beside its own code.

So there is one home for the catalogue the mesh owns, and every application that runs on the mesh rather than being of it carries its own — both reach the pipeline the same way, as a registered source.

Layout

modules/<name>.json      one manifest per module

Flat, because the catalogue's shape carries no meaning: a module is found by its name and described by its manifest, and what relates two modules — a shared seat, a claim, a provider/consumer edge — is data inside the manifests, not a directory the tree encodes (novox/hq, the domain-grouping question closed in favour of seats, claims and tags).

The manifest contract

The shape a manifest must satisfy is owned by the control plane's catalogue engine and is what validates a manifest before a machine ever sees it — a stray key, a consumer contributing the wrong provision field, an image that nothing builds. That validation belongs with this repository and is being re-homed here from mesh-controller; until it is, the pipeline is the gate — it builds each module and refuses a manifest it cannot resolve.

Where the reasoning lives

Design and decisions are in novox/hq:

  • 02-DECISIONS/0002-everything-is-a-module.md — one unit, no second mechanism
  • 02-DECISIONS/0010-applications-live-in-their-own-repository.md — why applications are not here
  • 02-DECISIONS/0030-the-repository-structure.md — the repositories, and the open tier-4 question this repository answers
  • 03-DESIGN/00-as-is/10-module-catalogue.md — the catalogue's shape, and what it records
S
Description
Novox Mesh — the catalogue. Module manifests the mesh builds, provisions and runs. Data, not a control-plane concern.
Readme
452 MiB
Languages
Go 83.2%
TypeScript 9.3%
Shell 6.5%
C++ 0.9%