jschoubben
5b7b280e3a
Jochen: "I thought we did not want to run the host under a systemd/openrc/init loop, but instead had our own host-init program?" -- and that was right. I had moved the give-up logic out of unit files and left RESTART in them, with the launcher exec'ing the host and disappearing. So init still decided when the host came back, which is the arrangement 0061 exists to remove. The launcher now stays and supervises: starts the host as a child, waits, decides. Init is asked for one thing, run this at boot. There is an OpenRC script beside the systemd unit now, four lines each, which is the point -- a second init is transcription rather than a port. The cost of not exec'ing is signals. A supervisor that exits while its child runs leaves the host to be killed rather than to stop, and an apply interrupted that way is the half-configured machine this project is about. So SIGTERM is trapped, passed down, and waited on. Two bugs, both found by the tests rather than by review: A clean exit was counted as a failure. The host exits cleanly to stand aside for a new binary after an upgrade (0057), so a host that upgraded itself three times rolled itself back having worked perfectly every time. The counter now counts CONSECUTIVE FAILURES, incremented after the wait rather than before the start. And when rolling back I reset the counter file but not the variable, so the next failure counted from the old value -- the rolled-back version got one attempt instead of three. Also: the host now clears the counter when it completes a reconcile, at the same moment it records known-good and for the same reason. Without it the count only climbs, and a node up for months rolls itself back on its third ordinary restart -- a healthy machine undone by its own recovery. One test expectation was tightened rather than fixed: "resets the counter after rolling back" asserted exactly 0, which was true only under the old count-before-start semantics. It now asserts the property -- below the limit -- since 1 is correct after a rollback plus one failure. 32 launcher tests, all confirmed to bite.
mesh-host
Tier 0 of the Novox Mesh. The one thing ever installed by hand, and the only thing that changes a machine.
scp mesh-host root@machine:/usr/local/bin/
mesh-host profile
That is the whole installation. One statically linked binary, nothing else present, no runtime
to install first (novox/hq ADR 0041).
What it is for
Apply declared state on this machine. Overlay membership, packet filtering, packages, services, containers and filesystems are not six concerns it carries; they are six instances of the one.
It does not decide. Anything needing knowledge of another node is the control plane's, and the host never queries the mesh database. It receives declarations and applies them.
What exists today
Stages 1 and 2. It reports what a machine is, and it applies a declaration to one. It connects to nothing and listens on nothing — what it applies comes from a file.
mesh-host profile what this machine can be asked to do
mesh-host inventory what this machine is, and what it holds
mesh-host apply FILE make this machine match a declaration from a file
mesh-host reconcile make this machine match the declaration this host carries
mesh-host bundle show what this host carries
mesh-host owned what this host has applied and still owns
--json machine-readable
--state where this node keeps what it knows
--dry-run read and check the declaration, change nothing
$ mesh-host profile
linux/amd64
yes container-runtime 29.7.2
no firewall nft exited 1: Operation not permitted (you must be root)
yes graphical-session x11: :1
yes overlay wg0
yes package-manager pacman 7.1.0
no privileged effective uid 1000, not 0
yes service-manager degraded
cannot be asked to: [firewall privileged]
Applying
A declaration is JSON, versioned, and an ordered list of resources — the order is stated
rather than derived, because deriving it would be the host deciding
(novox/hq ADR 0043). The vocabulary is directory, file
and service, and anything outside it refuses the whole declaration: a host that skipped
what it did not understand would apply most of a declaration and report success.
{"declaration":1,"resources":[
{"id":"mesh-etc","type":"directory","path":"/etc/mesh","mode":"0755"},
{"id":"node-conf","type":"file","path":"/etc/mesh/node.conf","content":"role = anchor\n","mode":"0640"},
{"id":"journal","type":"service","unit":"systemd-journald.service","state":"running"}
]}
It converges rather than executes. Applying twice changes nothing the second time; applying to a drifted machine returns it. A mode is maintained, not merely set — a permission applied at creation is not a permission held.
It owns a footprint, and only that. What it applied and is no longer declared is removed; what it did not create is never touched. It knows which is which because it recorded what it did, after each thing worked.
A failed step fails the apply. No step runs after a failure, and the error carries what had already been done — the machine is in whatever state that left it, and pretending otherwise is the fault this exists to prevent.
The bundle a host carries
A host built for a machine carries its declaration inside the binary:
make host BUNDLE=path/to/substrate.lock
mesh-host reconcile then applies it. That is the first node's path — no mesh present, nothing
fetched, nothing else copied onto the machine. copy it and run it stops being true the moment
a second file has to arrive with it, which is why the bundle is embedded rather than beside it.
A default build carries nothing and refuses to reconcile, saying so. A host that applied nothing and reported success would look exactly like one that raised a first node, and the difference would surface later as a mesh that never came up with nothing to point at.
Stages 3 and 4 — the link, and enrolment — are designed and not built.
What stage 2 does not yet prove
The design defines stage 2 as the host applies substrate.lock with no mesh present, and
calls out the claim underneath it: that one host can raise the substrate alone.
The mechanism is proved — a sealed machine, one binary, and it configures itself from what it carries. The claim is not. The substrate is four container services, and:
- the vocabulary has no container type, because a container needs an image and where images
come from is open (
novox/hqresearch 012); - what belongs in a substrate is not known — the closure for a one-node mesh is what research 011 and 012 exist to answer;
- and the machine used to test this has no container runtime, because a sealed network cannot install one.
So substrate.lock here is a real bundle with a placeholder's content. Saying that plainly
beats shipping a host that claims a substrate it has never raised.
A capability is detected, never assumed
The reason this is the first thing built rather than a detail of it.
An installed package is not a capability. A container client on disk with its daemon down looks exactly like a working runtime, and a node assigned work on that basis fails at the moment the work arrives. So every detector runs something that only succeeds if the thing is functioning — the daemon is asked for its version, the package database is queried, the firewall is asked to list a ruleset, which needs the privilege as well as the tool.
Every verdict says how it knows. A capability reported absent with no reason is a fault nobody can act on. The reason is what a person reads when a node will not take work they expected it to take.
A unit that does not exist is not a unit that is stopped. systemctl is-active says
inactive for both, so declaring a unit stopped reported success for a unit the host cannot
manage at all. LoadState separates them. Found by applying inside a raised machine, not by
reasoning — and its sibling: removing an orphaned service whose unit has since been uninstalled
used to fail the whole apply, which left a node able to apply nothing, ever.
Exit codes are not the whole answer. Found by running against a real machine rather than by
reasoning: systemctl is-system-running exits non-zero for every state except running —
including degraded, which means some units failed and the init is emphatically there. Reading
the exit code reported no service manager on a machine whose init it was. That is the same
fault in the mirror — installed-but-broken reported present, working-but-imperfect reported
absent — and both place work wrongly.
Building
go test ./... structure and logic, and the same checks against this machine
CGO_ENABLED=0 go build -ldflags="-s -w" -o mesh-host ./cmd/mesh-host
Roughly 3 MB, static, no dynamic dependencies. Cross-compiles with GOOS/GOARCH; a host is
built once per architecture and copied, never built on the machine it runs on.
Mocking the boundary is forbidden (novox/hq ADR 0034).
Every detector is exercised against a fake runner for its logic and against this machine for
its behaviour. The tests do not assert which capabilities a machine has — that varies, and is
the point of detecting — they assert that detection tells the truth about whatever is there.
Where the reasoning lives
Design and decisions are in novox/hq, not here. This
repository carries implementation and does not carry decisions.
03-DESIGN/01-to-be/05-the-node-host.md— what this is and the order it is built in02-DECISIONS/0037-the-host-applies-it-does-not-decide.md— the one concern02-DECISIONS/0038-a-node-joins-by-linking-first.md— one behaviour, two sources02-DECISIONS/0039-the-link-is-the-security-boundary.md— a node owns no password02-DECISIONS/0041-the-host-depends-on-nothing.md— why this is a static binary, and Go04-ISSUES/007-an-installed-package-is-not-a-capability— why detection works this way