novox/hq
3
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request 'The scenario model, the lifecycle, and what the lab actually costs' (#6) from design/scenario-underlay-detail into main

Browse Source
This commit was merged in pull request #6.
This commit is contained in:
jschoubben
2026-08-24 00:40:54 +02:00
parent d64b386d03 e88b448145
commit 132f6a0626
7 changed files with 1074 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files
01-RESEARCH/010-lab-inner-loop-cost/00-overview.md
+60
View File
@@ -0,0 +1,60 @@
---
status: active
initiated: 2026-08-24
touches:
- 03-DESIGN/01-to-be/03-scenario-lifecycle.md
- 02-DECISIONS/0016-a-lab-node-is-a-virtual-machine.md
- 02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md
became: []
---
# 010 — What the lab's inner loop actually costs
## What is being investigated
[The lifecycle design](../../03-DESIGN/01-to-be/03-scenario-lifecycle.md) closes on an open
question that is measurable rather than arguable:
> **What a snapshot costs.** Whole-scenario snapshots of several machines are the operation the
> inner loop repeats most, so their cost sets the loop's speed. If restoring is slow, the loop
> is slow, and everything above is theory.
Measured on a workstation, 2026-08-24. Numbers in [`measurements.md`](measurements.md).
## Why it matters
[ADR 0029](../../02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md) makes the
bootstrap scenario the inner development loop for tiers 0 and 1 — the argument being that
raising a node from nothing stops being the least-exercised path and becomes the most-exercised
one. **That argument is only true if raising and resetting are cheap.** A loop that costs
minutes is a loop people avoid, and the least-exercised path stays least-exercised.
## Status
Measured, and the finding is a blocker rather than a data point.
**A snapshot on the current host is a full copy of the machine's disk.** 1.6 GB and ten seconds
for one small virtual machine at best — and over two minutes when observed a second time. Cost
scales with the number of machines and the size of their disks, not with what changed.
**With copy-on-write it is 0.13 seconds and costs the delta.** Verified, not assumed. The
projected four-machine reset cycle falls from roughly ninety seconds to roughly fifteen, of
which almost all is a boot that cannot be avoided.
The cause is not virtual machines and not incus. It is that the host offers incus exactly one
storage driver, `dir`, which has no copy-on-write and therefore no cheap snapshot. The kernel
supports btrfs; the userspace tool that would let incus use it is simply not installed.
So the question *"is the lab's inner loop fast enough"* currently answers itself the wrong way,
for a reason that is one declared package away from being fixed — and declaring packages is
something the mesh already does.
## Open questions
| Question | Why it matters |
|---|---|
| ~~How much does a copy-on-write pool actually improve it?~~ **Measured: snapshot 9.9 s → 0.13 s, restore 10.4 s → 0.80 s, three snapshots sharing 1.36 GB rather than costing 4.8 GB.** The projected four-machine cycle falls from ~90 s to ~15 s. | Answered. The inner-loop argument holds *with* copy-on-write and did not without it. |
| How does the lab install its own prerequisites on a clean machine? | The lab needs a virtualisation daemon, copy-on-write tooling and a pool before it can do anything — and it cannot depend on the mesh for them, since it is where the mesh is built. |
| Why was the second snapshot more than twelve times slower than the first? | If snapshot cost is unpredictable rather than merely high, that is worse — a loop with a variable multi-minute step is one nobody trusts. |
| Does a scenario snapshot need the machines stopped? | Stateless snapshots of a running virtual machine capture the disk but not memory. Whether a mesh restored that way is coherent is not established. |
| What is the cost at scenario scale — four machines rather than one? | Only single-machine numbers were taken. If the operation is serial, four machines is four times the wait. |
01-RESEARCH/010-lab-inner-loop-cost/measurements.md
+141
View File
@@ -0,0 +1,141 @@
---
effort: 010-lab-inner-loop-cost
updated: 2026-08-24
---
# Measurements
Taken 2026-08-24 on a workstation with hardware virtualisation available, an NVMe-backed ext4
root, and 300 GB free. One virtual machine, 1 GiB memory, 2 CPUs, from a cached distribution
image.
## The environment, before anything ran
| Fact | Value | Consequence |
|---|---|---|
| Hardware virtualisation | present | virtual machines run at native speed; the choice in [ADR 0016](../../02-DECISIONS/0016-a-lab-node-is-a-virtual-machine.md) is not paying an emulation penalty |
| Storage drivers the daemon offers | **`dir` only** | no copy-on-write, therefore no cheap snapshot |
| Host filesystems | ext4 throughout | nothing copy-on-write to put a pool on |
| btrfs kernel module | **available** | the kernel can do it |
| `btrfs-progs` | **not installed** | which is the entire reason the driver is absent |
The last two rows are the finding. The daemon advertises only `dir` because the userspace tool
for anything better is missing — not because the host cannot do better.
## Raising a machine
| Step | Time |
|---|---|
| launch call returns | **3.4 s** |
| machine actually usable — a command executes on it | **14.3 s** |
The gap matters for the lifecycle design: `raise` returning is not the same as the scenario
being ready, so the verb has to wait for the second number, not report the first. Reporting
the first would be the mesh's own recurring failure — transport reported as effect.
## Snapshot and restore
| Operation | Time | Disk |
|---|---|---|
| snapshot, first | **9.9 s** | **+1.6 GB** |
| snapshot, second | **> 120 s — did not complete** | — |
| restore call returns | **10.4 s** | — |
| machine usable again | **20.1 s** total | — |
Instance on disk before snapshotting: 1.5 GB. Snapshot directory afterwards: 1.6 GB. **A `dir`
snapshot is a full copy** — the storage cost equals the instance, and nothing is shared.
Implied copy throughput on the first snapshot is roughly 160 MB/s, which is far below what the
underlying NVMe can do and is consistent with a real, durable copy rather than a metadata
operation.
**The second snapshot is the more troubling number.** It exceeded two minutes and was still
running when the observation was cut off; only the first snapshot exists. Whatever the cause —
page cache exhausted by the preceding restore, writeback contention — the practical
consequence is that snapshot cost here is **not merely high, it is unpredictable**.
## What this projects to
A four-machine scenario, taking the optimistic single-machine numbers and assuming the
operations are serial:
| | one machine | four machines |
|---|---|---|
| raise, to usable | 14 s | ~57 s |
| snapshot | 10 s, 1.6 GB | ~40 s, 6.4 GB |
| restore, to usable | 20 s | ~80 s |
A reset-and-rerun cycle is therefore **around a minute and a half at best**, and unbounded at
worst, before any of the mesh's own work begins.
## The judgement
**This is too slow for an inner loop**, and the reason is not the design.
[ADR 0029](../../02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md) argues that
making the bootstrap path the inner development loop turns the least-exercised code in the
system into the most-exercised. That argument holds only while resetting is cheap. At a minute
and a half a cycle, with occasional multi-minute stalls, the loop is one a person works around
— and the path stays under-exercised for exactly the reason it always was.
Nothing about virtual machines causes this. Hardware virtualisation is present and the machines
boot in fourteen seconds. **The cost is entirely the storage driver**, and the driver is absent
because one userspace package is not installed on the host.
The mesh already has the mechanism for that: a module declares a package, and a hook makes it a
working capability — which is precisely what was just done for the virtualisation daemon
itself, and what [`04-ISSUES/007`](../../04-ISSUES/007-an-installed-package-is-not-a-capability/00-report.md)
is about.
## The fix, measured
The comparison was subsequently run. One package — `btrfs-progs`, no dependencies — installed by
hand, the daemon restarted so it re-detected drivers, a copy-on-write pool created on a loop
file, and the identical image launched onto it.
| Operation | `dir` | copy-on-write | |
|---|---|---|---|
| snapshot | 9.9 s, then **> 120 s** | **0.13 s** | ~76× faster, and *consistent* |
| snapshot again | — | 0.12 s | |
| snapshot a third time | — | 0.13 s | |
| restore call | 10.4 s | **0.80 s** | ~13× faster |
| restore, to usable | 20.1 s | **10.5 s** | the remainder is boot, which is irreducible |
| three snapshots, storage | ~4.8 GB | **1.36 GB total, shared** | cost is the delta, not the disk |
**The fix is real, and larger than expected.** Snapshot goes from ten seconds to a tenth of a
second, and — more importantly — from *wildly variable* to *flat*. Three consecutive snapshots
took 0.13, 0.12 and 0.13 seconds. On `dir` the second snapshot never finished.
Storage stops scaling with the machine and starts scaling with what changed: three snapshots of
a 1.5 GB instance occupied 1.36 GB in total, because they share.
### What it projects to
A four-machine reset-and-rerun cycle, the operation the inner loop repeats most:
| | `dir` | copy-on-write |
|---|---|---|
| snapshot the scenario | ~40 s, 6.4 GB | **~0.5 s**, delta-sized |
| restore it | ~40 s + boot | **~3 s** + boot |
| **cycle** | **~90 s, unbounded at worst** | **~15 s, dominated by boot** |
At fifteen seconds, dominated by a boot that cannot be avoided, the inner loop is viable and
[ADR 0029](../../02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md)'s argument holds.
At ninety it did not.
### One honest counter-observation
Launching onto the fresh copy-on-write pool was **slower** — 20.2 s to usable against 14.3 s —
because the image had to be unpacked into a pool that had never seen it. That cost is paid once
per pool, not per scenario, and it is dwarfed by what snapshotting saves. But it is a real
number and it went the other way.
### State this left behind
Recorded because hand-made state is exactly what the mesh's rules exist to prevent, and it must
be declared properly rather than left as an artefact of a measurement:
- `btrfs-progs` installed by hand. Its installation regenerated the boot initramfs, a side
effect worth knowing about.
- The daemon restarted once, to re-detect drivers.
- The test pool and instance were **removed**; the pool the lab actually needs does not exist.
02-DECISIONS/0032-a-scenario-is-an-isolated-address-space.md
+71
View File
@@ -0,0 +1,71 @@
---
status: accepted
date: 2026-08-23
deciders: jochen
reconstructed: false
---
# 32. A scenario is an isolated address space, and the lab never reaches into it over IP
## Context
A scenario declares literal addresses —
[the declaration](../03-DESIGN/01-to-be/02-scenario-declaration.md) is full of them, and it has
to be, because reproducing *published but behind NAT* means saying which address the world sees.
That raises a question the declaration left open: **two scenarios at once.** Several agents
working means several scenarios, and the lab design already calls that a requirement. But two
scenarios built from the same declaration want the same addresses, and there are only three
documentation ranges in existence.
## Considered options
1. **Allocate addresses from a pool at raise time**, rewriting the declaration's literals.
Rejected. It makes the addresses in a declaration a fiction, so a scenario reproducing a
specific topology no longer reproduces it; it breaks the RFC-range validation, since
allocated addresses would have to come from somewhere real; and the numbers a person reads
in the file stop being the numbers they will see in a capture.
2. **One scenario at a time.** Rejected — it is the requirement, not an inconvenience. A gate
an agent has to queue for is a gate that gets bypassed.
3. **Give each scenario its own network stack, so the addresses do not collide.** Chosen.
## Decision
**A scenario is a closed address space.** Every segment materialises as its own isolated link,
belonging to one scenario instance. Two scenarios raised from the same declaration hold the same
addresses and never meet, because nothing joins their links.
The declaration therefore keeps its literal addresses, and they mean exactly what they say.
**The consequence that constrains everything else: the lab never reaches into a scenario over
IP.** It talks to a machine through the virtualisation layer's own channel — the same way one
executes a command in a container without the container being routable.
That is not a preference. If the lab reached machines by address, the workstation running it
would need a route into each scenario, and two scenarios carrying the same prefix would give it
two routes to the same destination. Concurrency would be impossible, and it would fail in the
worst available way: not with an error, but by one scenario's traffic arriving in another.
## Consequences
- Scenarios are concurrent by construction, with no allocation, no bookkeeping and no limit
beyond the machine's capacity.
- The three documentation ranges stop being a scarce resource. Every scenario may use all of
them, because no two scenarios share a link.
- **The lab cannot use IP to check anything**, which is more of a constraint than it first
appears: *"can this machine reach that one"* has to be asked **from inside the scenario**, by
executing on a machine, rather than probed from outside. That is the honest way to ask it
anyway — reachability from the workstation is not the question.
- A scenario is a unit that can be paused, snapshotted and destroyed whole, because nothing
outside holds a reference into it.
- The lab needs a scenario **instance** identity distinct from the scenario name in the
declaration: the declaration is a kind, and several instances of one kind may exist.
- **The workstation is not on the scenario's network, so it is not a node in it.** Anything a
developer wants to reach — a web interface, a database — needs an explicit, deliberate
forward out of the scenario, which is a feature rather than a gap: nothing leaks by default.
## References
- [ADR 0031](0031-the-lab-provides-the-underlay.md) — the declaration whose literal addresses
this preserves.
- [ADR 0029](0029-the-labs-first-scenario-has-no-pipeline.md) — the lifecycle jobs this shapes.
03-DESIGN/01-to-be/02-scenario-declaration.md
+544 -46
View File
@@ -17,31 +17,191 @@ everything in the lab hangs off, so it is worth getting small.
It states what a hosting provider and a home router would provide, and nothing the mesh is
responsible for ([ADR 0031](../../02-DECISIONS/0031-the-lab-provides-the-underlay.md)).
## Public networks are unrelated, and routed rather than bridged
The internet is not a network. It is a very large number of unrelated networks that route to
each other, and a machine in one is **many hops** from a machine in another with no shared
broadcast domain between them.
So a scenario does not have *an* internet segment. It has **one public segment per public
network**, each with its own unrelated prefix, and the lab wires them together **through a
router, never onto a shared bridge**.
That distinction is load-bearing, and putting several public addresses in one prefix would
quietly make four things true that are false in reality:
| If public addresses share a segment | Reality |
|---|---|
| machines resolve each other by ARP and talk directly | they are routed, many hops apart |
| TTL never decrements | every hop decrements it |
| broadcast and multicast reach across | neither crosses a router |
| any two are adjacent | adjacency is the exception, not the rule |
The third is not hypothetical here. The mesh has already been bitten by multicast name
resolution — [`00-as-is/01-mesh-and-transport.md`](../00-as-is/01-mesh-and-transport.md)
records that mesh names are deliberately not multicast names, after a delay and a
one-node-only failure mode. A lab where "the internet" is one broadcast domain would let a node
discover a peer by multicast that it could never discover in production, and report success.
**The rule: `kind: public` segments are routed to one another and never bridged.** It is a
property of how the lab wires them, not a field anyone sets, because there is no correct
scenario in which two public networks are adjacent.
### Which addresses to use
RFC 5737 reserves three ranges and RFC 3849 reserves one v6 prefix. Each public network takes
its own, and they are chosen to look nothing like each other — because in reality they would
not:
| Public network | v4 | v6 |
|---|---|---|
| a hosting provider | `192.0.2.0/24` | `2001:db8:a::/48` |
| a household ISP | `198.51.100.0/24` | `2001:db8:b::/48` |
| a mobile or foreign network | `203.0.113.0/24` | `2001:db8:c::/48` |
Three is enough for the topologies that matter, and a fourth public network can subnet one of
them — an ISP handing out `198.51.100.0/25` and `198.51.100.128/25` to two customers is exactly
what really happens.
Private segments still use RFC 1918 and stay byte-identical to production.
## What NAT does, and why the design turns on it
A household or office has **one** address the outside world can see, and **many** machines
behind it. Network address translation is what reconciles those.
When a machine inside dials out, the gateway rewrites the packet's source from the private
address to the public one, **remembers the mapping**, and rewrites the replies on the way back.
Four consequences follow, and every one of them shapes this design:
1. **Outbound works; inbound does not.** A mapping exists only because something inside started
a conversation. Nothing outside can start one — there is no mapping to look up, and no way
to know which internal machine was meant.
2. **A forwarded port is a permanent mapping made by hand**, in the inbound direction:
*anything arriving at the public address on 443 goes to this machine.* That is the only way
a machine behind NAT becomes reachable, and it requires control of the gateway.
3. **Mappings expire.** A gateway forgets one that goes unused. This is why anything holding a
connection through NAT sends keepalives, and why a mesh that does not is fine until it is
idle.
4. **From outside, every machine behind the gateway looks like one address.** Identity and
address stop corresponding.
This is why the mesh dials outward and never inward
([ADR 0001](../../02-DECISIONS/0001-nodes-communicate-over-a-broker.md)), why a hub exists at
all, and why a node's endpoint is something a peer **learns** from arriving packets rather than
something anyone configures.
**Carrier-grade NAT** is the same mechanism applied by an ISP: your own gateway gets a private
address too, and the public one is shared with strangers. Nothing can be forwarded, because the
rule would have to live on equipment you do not own. Common on mobile connections and
increasingly on fixed ones.
## Three positions a machine can be in
The underlay's whole job is to reproduce **where a machine sits relative to the internet**,
because that is what the mesh has to cope with and what only production currently exercises.
There are three positions, and they are genuinely different:
| Position | Reachable from outside | Apparent address | Example |
|---|---|---|---|
| **Attached** | yes, at its own address | its own | a hosted server |
| **Behind a forwardable gateway** | only through a forwarded port, at the *gateway's* address | the gateway's | a machine at home |
| **Behind an unforwardable gateway** | **no** | someone else's, and it changes | a laptop on a café network; anything behind carrier-grade NAT |
The axis is **forwardability, not ownership** — which is worth stating because the obvious
framing gets it wrong. Carrier-grade NAT is *your* connection and is still unforwardable, so it
belongs in the third row alongside the café. What the mesh has to cope with is whether an
inbound mapping can be made, not who owns the equipment.
The third position is the hard one. A machine there can dial out and nothing more: it cannot be
published, its apparent address belongs to a router it does not control, and that address
changes when it moves. Every assumption a mesh makes about reachability breaks there first.
A declaration has to be able to say all three, and to move a machine between them.
## Reachability is per address family, not per machine
Adding IPv6 is not a field. It changes the position model, and the reason is worth stating
before the syntax.
**IPv6 usually has no NAT.** A machine behind a household gateway can hold a *globally routable*
v6 address while its v4 address is private and unforwardable. The same machine, at the same
moment, is in **two different positions at once**:
| | IPv4 | IPv6 |
|---|---|---|
| a typical machine at home | behind an unforwardable-or-forwardable gateway | **attached**, directly reachable |
| a machine on mobile data | behind carrier-grade NAT | often attached, sometimes absent entirely |
| a machine on an older network | attached or behind NAT | **no address at all** |
So the three positions apply **per family**, and a machine's reachability is a property of
*(machine, family)* rather than of the machine. A mesh that treats reachability as one fact per
node will reach a peer over one family, fail over the other, and report whichever it tried.
That has a direct consequence for what the lab is for: *"can these two nodes reach each other"*
stops being a yes/no question. It is asked once per family, and the interesting answers are the
asymmetric ones.
The v6 documentation prefix is `2001:db8::/32` (RFC 3849) — the exact counterpart of the RFC
5737 rule, and load-bearing for the same reason.
## The shape
```yaml
scenario: published-behind-nat
scenario: roaming-and-published
segments:
wan:
cidr: 203.0.113.0/24 # RFC 5737 — never routes on the real internet
lan:
cidr: 192.168.1.0/24
behind: wan # NAT; the lab materialises a router
hosting: # one public network
kind: public
cidr: [192.0.2.0/24, 2001:db8:a::/48]
isp-home: # another, unrelated — routed to it, not bridged
kind: public
cidr: [198.51.100.0/24, 2001:db8:b::/48]
home:
kind: private
cidr: [192.168.1.0/24, 2001:db8:b:1::/64]
mtu: 1500
gateway:
to: isp-home
address: [198.51.100.7, 2001:db8:b::7] # what the world sees this network as
nat: [v4] # v4 is translated; v6 is routed, not translated
forwardable: true
mapping_ttl: 120s # an unused inbound mapping is forgotten after this
cafe: # a network we do not control
kind: private
cidr: [10.50.0.0/16] # v4 only — no v6 offered here at all
mtu: 1400 # a tunnelled path, smaller than standard
gateway:
to: hosting # it reaches the world via a different public network
address: [192.0.2.200]
nat: [v4]
forwardable: false # café wifi, or carrier-grade NAT
mapping_ttl: 30s # aggressive, as carrier NAT tends to be
machines:
anchor:
segment: wan
address: 203.0.113.10
at: { segment: hosting, address: [192.0.2.10, 2001:db8:a::10] }
home-server:
segment: lan
address: 192.168.1.135
forwarded: [443] # reachable from wan through the router
at: { segment: home, address: [192.168.1.135, 2001:db8:b:1::135] }
published:
- { port: 443, on: home } # v4 only: 198.51.100.7:443 → 192.168.1.135:443
inbound: allow # v6 is routable here, so this decides whether it is reachable
workstation:
segment: lan
address: 192.168.1.250
at: { segment: home, address: [192.168.1.250, 2001:db8:b:1::250] }
inbound: deny # a host firewall: dials out, accepts nothing
laptop:
segment: detached # reachable by nothing until attached
at: { segment: home, address: [192.168.1.98, 2001:db8:b:1::98] }
border: # a machine on two segments at once
at:
- { segment: home, address: [192.168.1.2] }
- { segment: isp-home, address: [198.51.100.60] }
place:
all: [host]
@@ -50,41 +210,113 @@ place:
snapshot: raised
```
That is a complete bootstrap scenario. Nothing in it mentions the overlay, a hub, peering,
names or certificates — all of which are outcomes to be observed.
## What each part means, precisely
## The four parts
**`segments`** — a broadcast domain with an address range, and a `kind:`. A segment is a single
broadcast domain, which is precisely why several public networks cannot be one segment.
**`segments`** — the networks that exist. `behind:` declares NAT, and is the only place a
router comes from: the lab materialises one without being asked, because NAT has to run
somewhere. This is the one implicit machine in an otherwise explicit declaration.
`kind: public` marks a segment that stands in for a public network — and there is normally more
than one, unrelated to each other. `kind: private` is everything else. This is stated rather than inferred, and the earlier version inferred it — *a segment with
no gateway is the internet* — which made an **isolated network inexpressible**: a LAN with no
route out is a private segment with no gateway, and would have been read as public and forced
to use documentation addresses. A mesh spanning a site with no internet access is a real
topology, and the model has to be able to say it.
**`machines`** — what sits where. A machine has a segment and an address, and that is nearly
all. `forwarded:` opens a port through the router, which is what makes *published but behind
NAT* reproducible — the case that exists only in production today. `segment: detached` is a
machine on no network, which is how a roaming node is expressed at rest.
**`gateway:`** — how a segment reaches its parent, and this is where the previous version was
too thin. It carries three facts, and all three are load-bearing:
**`place`** — what goes inside. `all:` applies to every machine; a machine name overrides for
that machine. This is the only part that differs between the two scenario classes.
- `to:` — the parent segment.
- `address:` — **the address the outside world sees this network as.** For a household
connection this is the public address the ISP hands out. It is not decoration: it is what a
peer records as the endpoint when a machine here dials out, and what a public name for a
published machine here resolves to.
- `nat:` — **which families are translated**, as a list. `[v4]` is the ordinary modern case:
v4 translated, v6 routed. `[v4, v6]` describes a gateway that translates both, which exists
and is worth being able to reproduce. `[]` is a routed range, where machines keep their own
addresses and the gateway only forwards.
- `forwardable:` — whether an inbound mapping can be created. Independent of `nat:`, and the
field that separates a home gateway from carrier-grade NAT. Publishing through a gateway with
`forwardable: false` is a declaration error, because that is exactly the constraint being
reproduced.
- `mapping_ttl:` — how long an unused inbound mapping survives. This is what makes keepalive
behaviour testable: a mesh that holds a connection through NAT without refreshing it works
perfectly until the far side goes quiet for longer than this. Aggressive values reproduce
carrier NAT; omitting it means mappings never expire, which no real gateway does.
**`snapshot`** — names the state once placement finishes, so a run can return to it without
raising everything again. Snapshots are what make repetition cheap, and cheap repetition is
what makes the bootstrap path the inner development loop rather than a ceremony.
**`segments[].mtu`** — the largest packet the segment carries, defaulting to 1500. Lower values
reproduce tunnelled and PPPoE paths. This matters because an overlay adds its own header: a
tunnel over a 1400-byte path establishes a connection and then silently drops large packets,
which is the shape of fault this whole effort exists to stop shipping.
**`machines[].inbound`** — `allow` or `deny`, a host firewall. Distinct from NAT and behaves
differently: a machine can be perfectly routable and still refuse everything unsolicited, which
is the normal state of a v6-addressed machine. Without this, v6 addressing would imply
reachability, and it does not.
The lab materialises a machine to be the gateway. That is the one implicit machine in an
otherwise explicit declaration, and it exists because NAT has to run somewhere.
**`machines[].at`** — segment and addresses, or a **list** of them for a machine on several
segments at once. Multi-homing is not exotic: it is what a border machine is, and what any node
with both a LAN and a WAN interface is. Each entry carries the addresses that machine holds on
that segment, one per family.
Position follows from the pair, per family: on a `kind: public` segment a machine is attached;
on a private one it is behind that segment's gateway, unless the gateway does not translate
that family — in which case it is attached on that family and behind a gateway on the other.
**`machines[].published`** — a destination-NAT rule on a named gateway, stated as an outcome
rather than a port list. `{ port: 443, on: home }` means the `home` gateway forwards its own
`203.0.113.50:443` to this machine's `443`. The resulting public endpoint is derivable, which is
the point: a scenario never writes an endpoint down, and the mesh has to discover it.
A machine may be published on **any gateway between it and a public network** — which is how *"our
LAN also has a public IP"* is expressed, and why `on:` names the gateway rather than being
implied. It cannot be published at all on a gateway the scenario models as foreign; attempting
it is a declaration error, because that is precisely the constraint being reproduced.
**`at: detached`** — on no segment. A machine that exists and can reach nothing.
## Moving a machine is a lifecycle operation
`at:` states where a machine *starts*. Moving it is something a run does:
```
move laptop → { segment: elsewhere, address: 198.51.100.23 }
move laptop → detached
move laptop → { segment: home, address: 192.168.1.98 }
```
This is the roaming case made testable, and it is the one that finds the interesting faults.
The same machine, the same identity, three positions in one run: at home where its peers can
reach it directly, on a foreign network where it can only dial out and its apparent address
belongs to a router it does not control, and asleep.
Whether the overlay survives that, re-forms, and is noticed to have changed endpoint is
**observed**, never arranged
([ADR 0031](../../02-DECISIONS/0031-the-lab-provides-the-underlay.md)).
## Why the addresses are load-bearing
The routable segment uses RFC 5737 documentation space, and this is not a stylistic choice.
The internet segment uses RFC 5737 documentation space, and this is not a stylistic choice.
The mesh decides *public versus private* by matching the address. A private range on the
segment meant to be routable makes the hub test as unreachable, and **the mesh silently never
forms** — no error, no failed step, just a mesh that does not exist. Research 004 calls this
the single most important fact in its analysis.
segment meant to be routable makes a would-be hub test as unreachable, and **the mesh silently
never forms** — no error, no failed step, just a mesh that does not exist. Research 004 calls
this the single most important fact in its analysis.
So the format should make this hard to get wrong rather than merely documented: a segment
without `behind:` is a routable segment, and an address in it that is not documentation space
is a declaration error, refused before anything is raised. That is
Which range goes where is covered above, under *public networks are unrelated*: one range per
public network, chosen to look nothing like each other.
Private segments use RFC 1918 and can be **byte-identical to production**, because those
addresses mean the same thing everywhere. Only the public side is substituted, and only because
it must be.
The format should make getting this wrong hard rather than merely documented: a segment without
a `gateway:` is a public segment, and an address in it — including a gateway's `address:` — that
is not documentation space is a declaration error, refused before anything is raised. That is
[ADR 0008](../../02-DECISIONS/0008-a-failed-step-fails-the-job.md) applied to a configuration
file — the failure it prevents is silent, so the check has to be loud.
file: the failure it prevents is silent, so the check has to be loud.
## The same declaration serves both classes
@@ -124,18 +356,284 @@ not first.
- **Steps.** A scenario is a desired state. Anything expressed as an ordered list of actions
belongs in the lifecycle, not the declaration.
## What the model deliberately does not contain
A real network is full of equipment: a modem, a router, switches, access points, controllers.
Almost none of it appears here, and the omission is deliberate rather than an oversight.
**The test is whether a device changes what an IP packet can do.** If two machines can exchange
packets, at the same addresses, with the same reachability and the same MTU, whether or not the
device exists — then the device is invisible to the mesh, and modelling it would add a fixture
with no fault to catch.
| Equipment | Modelled? | Why |
|---|---|---|
| **Switch** | no | Moves frames within a segment. Two machines on a switch are two machines on a segment. |
| **Access point** | no | Bridges wireless clients onto a segment. A machine on wifi and a machine on cable are the same machine to IP. |
| **Network controller** | no | Configures equipment. Its effects appear as segments and policy; it has no packets of its own. |
| **Cabling, PoE, uplink speed** | no | Change performance and availability, not reachability. |
| **Router / security gateway** | **yes — it *is* the gateway** | Translation, forwarding and inter-segment policy all live here. |
| **ISP modem** | **only in router mode** | In bridge mode it is a media converter and invisible. Doing its own NAT, it is a second gateway — and that is double NAT. |
| **VLANs** | **yes — they are segments** | Machines on separate VLANs cannot reach each other except through the router, which is the definition of a separate segment. |
| **Inter-VLAN firewall rules** | **yes** — see below | A rule stopping one segment reaching another is a reachability fact, and the mesh will hit it. |
The two entries worth dwelling on are the ones where a common household setup produces
something the mesh has to survive.
**A modem in router mode gives you double NAT.** Your gateway holds a private address from the
modem, which holds the public one. Forwarding then requires a rule on *both*, and one of them
may not be configurable. This is expressible as nested segments — `to:` chains — but publishing
through two gateways is not, and remains open.
**Segmented networks are the common case, not the exotic one.** A router with separate networks
for trusted machines, guests and devices is ordinary, and the rules between them are ordinary
too. A mesh node on one segment and a mesh node on another are, as far as reachability goes, on
different networks that happen to share a gateway.
## Segments may share a gateway
Several segments can name the same parent and the same external address. That is one router
with several networks behind it, which is what a VLAN-capable gateway is:
```yaml
segments:
trusted:
kind: private
cidr: [192.168.1.0/24]
gateway: { to: isp-home, address: [198.51.100.7], nat: [v4], forwardable: true }
devices:
kind: private
cidr: [192.168.30.0/24]
gateway: { to: isp-home, address: [198.51.100.7], nat: [v4], forwardable: true }
```
Identical gateway declarations mean **one gateway machine**, not two. The lab materialises a
single router serving both, because that is what the topology being reproduced is — and two
routers sharing one address would not work anyway.
## Policy between segments
Sharing a gateway does not mean segments can reach each other. What they may do is stated
separately, because it is a fact about a pair rather than a property of either:
```yaml
policy:
- { from: devices, to: trusted, allow: false } # devices may not initiate to trusted
- { from: trusted, to: devices, allow: true } # the reverse is fine
```
Default is `true` between segments behind the same gateway, matching a router with no rules
configured. Asymmetry is the normal case and the reason this cannot be a single flag: the
useful configuration is almost always one-directional.
This is **not** the same as `machines[].inbound`, and conflating them loses a real distinction:
| | Enforced by | Blocks |
|---|---|---|
| `policy` | the gateway, between segments | everything crossing, regardless of what the destination thinks |
| `inbound` | the machine itself | unsolicited traffic that already reached it |
A mesh node behind a `policy` deny cannot be reached even by a peer that knows exactly where it
is — and that is a real topology, not a contrived one.
## Worked example — a whole mesh of the ordinary kind
The shape research 004 identified: one machine with a routable address, one publicly named but
behind a household connection, one stationary machine on that network, one that roams. Written
out completely, with every field the model has.
```yaml
scenario: the-ordinary-shape
segments:
# ---- three unrelated public networks. Routed to each other, never bridged. ----
hosting: # where the always-on machine lives
kind: public
cidr: [192.0.2.0/24, 2001:db8:a::/48]
isp-home: # the household's uplink
kind: public
cidr: [198.51.100.0/24, 2001:db8:b::/48]
isp-mobile: # wherever the roaming machine happens to be
kind: public
cidr: [203.0.113.0/24, 2001:db8:c::/48]
# ---- private networks behind them ----
home: # the household network
kind: private
cidr: [192.168.1.0/24, 2001:db8:b:1::/64]
mtu: 1492 # PPPoE on the uplink; 1500 if the line is not PPPoE
gateway:
to: isp-home
address: [198.51.100.7, 2001:db8:b::7]
nat: [v4] # v4 translated, v6 routed — the modern default
forwardable: true # the household router is ours to configure
mapping_ttl: 120s
devices: # optional: a segmented network on the same router
kind: private
cidr: [192.168.30.0/24]
gateway:
to: isp-home
address: [198.51.100.7, 2001:db8:b::7] # identical → the SAME gateway machine
nat: [v4]
forwardable: true
mapping_ttl: 120s
cafe: # a network we do not control
kind: private
cidr: [10.50.0.0/16] # RFC 1918 — someone else's private range
mtu: 1400
gateway:
to: isp-mobile
address: [203.0.113.129]
nat: [v4]
forwardable: false # carrier-grade NAT, or simply not ours
mapping_ttl: 30s
policy:
- { from: devices, to: home, allow: false }
- { from: home, to: devices, allow: true }
machines:
anchor: # routable, nothing in front of it
at: { segment: hosting, address: [192.0.2.10, 2001:db8:a::10] }
inbound: allow
home-server: # publicly named, behind the household connection
at: { segment: home, address: [192.168.1.135, 2001:db8:b:1::135] }
published:
- { port: 443, on: home } # v4 reaches it only through the forward
inbound: allow # and v6 reaches it directly, so this matters
workstation: # on the household network, not published
at: { segment: home, address: [192.168.1.250, 2001:db8:b:1::250] }
inbound: deny
laptop: # starts at home; moves during the run
at: { segment: home, address: [192.168.1.98, 2001:db8:b:1::98] }
inbound: deny
place:
all: [host]
anchor: [substrate]
snapshot: raised
```
### What the ISP modem is doing here
**Nothing, if it is in bridge mode** — which is the ordinary arrangement when the household has
its own router. A bridged modem is a media converter: it changes the physical medium and leaves
the packets alone, so it creates no IP-level fact and appears nowhere above.
Were it in router mode it would be a second gateway, `home` would sit behind it rather than
behind `internet` directly, and publishing would need a rule on both — the case the model
cannot yet express.
### What is deliberately absent
Nothing here mentions overlay addresses, which node is the hub, who peers with whom, any name,
or any certificate. Research 004 recorded all of those for this topology, and **a scenario must
not state them** ([ADR 0031](../../02-DECISIONS/0031-the-lab-provides-the-underlay.md)): they
are what the mesh does, and a scenario that supplied them would be certifying its own work.
The absence is the point. Given the declaration above, whether a hub is elected, whether the
NATed machine's endpoint is learned, whether the roaming machine re-forms after moving — all of
it is observed.
### Running it
```
move laptop → { segment: cafe, address: [10.50.3.23] } # it leaves the house
move laptop → detached # it sleeps
move laptop → { segment: home, address: [192.168.1.98, 2001:db8:b:1::98] }
```
One identity, four positions, one run. The `mapping_ttl: 30s` on `cafe` means a connection held
without refreshing dies while it is out — which is the point of putting a number there.
Note that the laptop's three positions are on **three different public networks**: at home it
appears as `198.51.100.7`, at the café as `203.0.113.129`, and the machine it is trying to reach
is on a fourth. None of them are adjacent, all of them are routed. That is the situation the
mesh actually faces.
## Is this general? — the axes a setup can vary along
The question that matters is not *does this cover our mesh*, but **can it express any mesh**.
The standard applied is not "every property a network has". It is **every property that changes
how the mesh behaves**. Bandwidth does not change correctness; MTU does.
| Axis | Values | Expressible | |
|---|---|---|---|
| **Reachability** | attached · forwardable gateway · unforwardable gateway · isolated | yes | the core of the model, and **per family** |
| **Address family** | IPv4 · IPv6 · dual-stack · neither | yes | `cidr:` and `address:` take both; `nat:` names which families are translated |
| **Interfaces per machine** | one · several | yes | `at:` takes a list |
| **Reachability policy, host** | symmetric · asymmetric | yes | `inbound:` — a routable machine that refuses everything |
| **Reachability policy, network** | open · segmented · asymmetric between segments | yes | `policy:` — inter-segment rules, as a segmented router enforces |
| **Segments per gateway** | one · several behind one router | yes | identical gateway declarations mean one gateway machine |
| **Gateway state** | permanent · expiring mappings | yes | `mapping_ttl:` |
| **Path MTU** | standard · reduced | yes | `segments[].mtu` |
| **Overlapping ranges** | distinct · two sites both on `192.168.1.0/24` | yes | segments may carry the same range |
| **Segment count** | one · many · isolated island | yes | `kind:` distinguishes an island from a public network |
| **Adjacency** | same segment · routed · unrelated public networks | yes | public segments are routed, never bridged — so no two are adjacent unless declared so |
| **Gateway depth** | direct · one gateway · nested | **partly** | `to:` chains, so nesting exists; `published:` names one gateway, so forwarding through two does not |
| **Address stability** | static · dynamic · changes mid-run | **partly** | a machine can be *moved*; an address changing under it in place cannot be stated |
| **Path quality** | latency · loss · bandwidth | **no**, deliberately | changes performance, not correctness — modelling it makes a network simulator, not a fixture |
### What closing the gaps changed
**Address family was not a field.** It changed the position model: a machine behind a household
gateway is typically *unforwardable on v4 and directly attached on v6, simultaneously*. So
reachability is a property of *(machine, family)*, and *"can these two nodes reach each other"*
is no longer a yes/no question — it is asked once per family, and the asymmetric answers are the
interesting ones. That distinction did not exist in the model an hour ago and would have been
discovered by a mesh failing over one family while reporting the other.
**`inbound:` became necessary because of v6.** With NAT, unreachability was implied by the
topology. With a globally routable v6 address, a machine is reachable unless something refuses —
so refusing has to be sayable, or v6 addressing would silently imply reachability.
**`nat:` became a list rather than a boolean** for the same reason: a real gateway translates v4
and routes v6, and a boolean cannot say that.
### What remains open, and whether it matters
Two partial axes, both extensible when something needs them, neither blocking: **nested
forwarding** and **an address changing in place**. A machine can already be moved, which covers
the roaming case; what is missing is a lease expiring underneath a machine that stays put.
One deliberate exclusion: **path quality**. Latency and loss change how fast the mesh is, not
whether it is correct. If a timeout turns out to be load-bearing that judgement should be
revisited — and it would be revisited by a real failure, which is the right trigger.
### The honest summary
The model now covers **where a machine sits**, **what the path between machines is like**, and
**what is permitted between them** — across both address families. Together those are what the
mesh's reachability logic turns on.
It contains almost no equipment, by the test above: a device that does not change what an IP
packet can do has no fault for a scenario to catch. What it does contain is every device that
does — which turns out to be the router, and a modem only when the modem is also a router.
What it still does not model is *change over time* beyond moving a machine, *degradation* short
of failure, and *publishing through two gateways at once*. The first two are chosen; the third
is the one real absence, and it is exactly the double-NAT case.
## Open
- **`user` and `edge` profiles have no scenario.** A lab machine is always privileged, so the
two profiles that exist for unprivileged and phone-like participation cannot currently be
exercised. Either the lab grows a way to run the host unprivileged, or those profiles are
developed against something that is not a virtual machine.
- **Attaching and detaching during a run.** `segment: detached` covers a machine at rest;
moving one between segments while a scenario is live is what makes a roaming node
interesting, and that is lifecycle rather than declaration.
two profiles that exist for unprivileged and phone-like participation cannot be exercised.
Either the lab grows a way to run the host unprivileged, or those profiles are developed
against something that is not a virtual machine. This is the largest gap.
- **Where `place:` gets its artifacts from.** Before the mesh is self-hosting these come from
outside; afterwards from the mesh itself. The declaration should not have to care, which
suggests a named source rather than a path.
- **Multiple scenarios at once.** Each needs its own segments and addresses. Whether the
declaration carries absolute addresses, as above, or a template the lab allocates from,
decides whether two scenarios can run side by side.
- **Nested forwarding** — `published:` names one gateway, so a machine behind two cannot be
published through both.
- **An address changing in place**, as a DHCP lease expiring under a machine that has not moved.
03-DESIGN/01-to-be/03-scenario-lifecycle.md
+140
View File
@@ -0,0 +1,140 @@
---
layer: to-be
status: designed
code: [mesh-lab]
updated: 2026-08-23
decisions:
- 02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md
- 02-DECISIONS/0031-the-lab-provides-the-underlay.md
- 02-DECISIONS/0032-a-scenario-is-an-isolated-address-space.md
---
# Scenario lifecycle
The first thing the lab must do, and the only thing it must do before anything else can be
written: **materialise a mesh, return it to a known state, and destroy it**
([ADR 0029](../../02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md)).
A [declaration](02-scenario-declaration.md) describes a scenario. This describes what happens
to one.
## The verbs
| Verb | Does |
|---|---|
| `raise` | materialise a declaration into a running scenario instance |
| `snapshot` | name the current state of the whole scenario |
| `restore` | return the whole scenario to a named state |
| `move` | change a machine's position while the scenario runs |
| `exec` | run something on a machine, and get its output |
| `destroy` | tear the instance down |
Six verbs, and `raise` plus `destroy` are enough to be useful. The rest are what make repetition
cheap, and cheap repetition is what turns the bootstrap path into an inner development loop
rather than a ceremony.
## Raising, in order
The order is not arbitrary — each step needs the one before it to exist:
1. **Segments.** Isolated links, one per declared segment, belonging to this instance and
joined to nothing outside it
([ADR 0032](../../02-DECISIONS/0032-a-scenario-is-an-isolated-address-space.md)).
2. **Gateways.** Derived, never declared as machines: a gateway is materialised for each
distinct `gateway:` declaration, sitting on both its segment and its parent, carrying the
translation, forwarding and mapping-expiry the declaration asked for.
3. **Machines.** Each on its segments, holding its addresses.
4. **Policy.** Rules between segments, applied on the gateways that route between them.
5. **Placement.** Artifacts onto machines.
6. **Snapshot**, if the declaration named one.
`raise` must wait for the machines to be **usable**, not for the call to return. Measured, those
are 14 seconds apart for a single machine — and reporting the earlier number would be the mesh's
own recurring failure, transport reported as effect.
Raising is **convergent, not incremental**: raising an instance that already exists brings it to
the declared state rather than failing or duplicating. That is the same model the mesh itself
uses, and a lab that behaved differently from the thing it tests would be teaching the wrong
habit.
## A failed raise leaves the wreckage
A step that fails stops the raise
([ADR 0008](../../02-DECISIONS/0008-a-failed-step-fails-the-job.md)) — and **does not tear
down**.
Tearing down on failure destroys the only evidence of what went wrong, which is precisely
backwards: a scenario that failed to raise is more interesting than one that succeeded. The
instance stays, marked failed, with the step that failed named.
The caller decides what happens next, and the two callers want different things
— the coordinator captures and destroys, a person opens a shell. That is the same
one-runner-two-callers split the lab design already makes, applied to failure.
## Snapshots are whole-scenario
A snapshot captures **every machine and the state of the network between them**, as one thing.
Restoring returns all of it.
Per-machine snapshots would be cheaper and are wrong. The mesh keeps state that spans nodes —
what is assigned where, which grants exist, what has been delivered — so restoring one machine
to an earlier moment while its peers move on produces a mesh that has never existed and could
not. The faults found there would be artefacts of the lab.
This is what makes *fresh* and *upgrade* both cheap and both default: one snapshot of a mesh
that has never seen a change, another of a mesh running the previous version, and a restore
between runs.
## Moving a machine
`move` changes a machine's position while the scenario runs: to another segment, with different
addresses, or to `detached`.
It is the roaming case, and it is a **lifecycle** operation rather than a declaration because
the interesting part is the transition, not the destination. A mesh that forms correctly with a
node at home and correctly with it away may still fail to notice it moved.
Moving does not invalidate a snapshot. A snapshot is a state to return to; a move is a change
made after it, and returning undoes it like any other change.
## Reaching in
Everything the lab does to a machine goes through the virtualisation layer, never over IP
([ADR 0032](../../02-DECISIONS/0032-a-scenario-is-an-isolated-address-space.md)). `exec` runs a
command on a machine and returns its output.
This has one consequence worth stating plainly: **a reachability question is asked from inside**.
*Can this machine reach that one* is `exec` on the first, testing the second — not a probe from
the workstation. The workstation is not on the scenario's network and its opinion of
reachability would be a different question with a misleadingly similar answer.
Anything a person wants to open in a browser needs a deliberate forward out of the instance.
Nothing leaks by default.
## The two callers
The lab design already establishes that the runner serves the coordinator and a person, and
that anything only one of them can do will drift. Applied here:
| | the coordinator | someone working on the mesh |
|---|---|---|
| wants | structured results, clean teardown, no prompts | readable output, the scenario **left standing**, one thing re-run without repeating the rest |
| on failure | capture, then destroy | leave it, open a shell |
Both use the same verbs. The difference is what happens after the verdict, which is a caller's
decision rather than a second implementation.
## Open
- **What a snapshot costs — measured, and the answer is a blocker.** On the current host a
snapshot is a full copy: 1.6 GB and ten seconds for one small machine, over two minutes when
observed again, projecting to roughly a minute and a half for a four-machine reset-and-rerun
cycle. The cause is the storage driver, not virtual machines. See
[research 010](../../01-RESEARCH/010-lab-inner-loop-cost/measurements.md).
- **Instance naming.** A declaration is a kind and instances are many; how they are named
decides whether a person can find the one they left standing yesterday.
- **What survives `destroy`.** Logs and captures are the output of a failed run, so destroying
the instance must not destroy them.
- **Placement before the mesh is self-hosting.** `place:` needs artifacts from somewhere, and
before the mesh builds itself that somewhere is outside it — the open question from
[research 006](../../01-RESEARCH/006-mesh-from-scratch/code-skeleton.md).
03-DESIGN/01-to-be/04-lab-installation.md
+116
View File
@@ -0,0 +1,116 @@
---
layer: to-be
status: designed
code: [mesh-lab]
updated: 2026-08-24
decisions:
- 02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md
- 02-DECISIONS/0008-a-failed-step-fails-the-job.md
---
# Installing the lab on a clean machine
The lab has prerequisites — a virtualisation daemon, copy-on-write storage, a pool, an identity
permitted to talk to it — and it cannot get them from the mesh, because it is where the mesh is
built ([ADR 0029](../../02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md)).
So the lab needs an install path of its own. This describes it, and the shape it has to take is
determined by two failures observed while measuring
([research 010](../../01-RESEARCH/010-lab-inner-loop-cost/measurements.md)).
## The two failures that shape this
**One: installed is not available.** The virtualisation package was present and explicitly
installed. Both its units were disabled, the operator was in no group, and the client reported
the server unreachable. Nothing had failed — the declaration was satisfied exactly as written
([`04-ISSUES/007`](../../04-ISSUES/007-an-installed-package-is-not-a-capability/00-report.md)).
**Two, and worse: available is not adequate.** With the storage tooling absent, the daemon
offered one driver, everything worked, and snapshots took **seventy-six times longer** than they
needed to. Nothing failed. Nothing warned. A lab in that state runs correctly and is simply too
slow to use — and the inner loop it exists to provide quietly does not exist.
The second is the more dangerous shape and it is a variant the mesh has not catalogued before.
Its usual failure is *reported success and did nothing*. This is **reported success and did it
seventy-six times slower**, which no error surface catches because nothing is wrong.
## What follows: the lab verifies capability, never installation
The install path may differ. **The verification does not.**
Before the lab will raise anything, it asserts the outcomes it needs — not that packages are
present, but that the machine can actually do the work:
| Assertion | Failing means |
|---|---|
| the daemon answers **as the invoking user**, not as root | a group membership that was granted but never took effect |
| a copy-on-write storage driver is offered | the userspace tooling is missing; snapshots will be full copies |
| **the pool the lab will use is on that driver** | a pool exists, and is the slow kind — the failure that has no symptom |
| hardware virtualisation is present | machines will be emulated and unusably slow |
| an image can be fetched or is cached | the first raise will fail late instead of early |
Each check states **why it matters**, in the terms of what it costs. *"The pool uses the `dir`
driver"* means nothing to someone who does not already know it means seventy-six times slower
and unbounded at worst.
**The lab refuses to run degraded.** It does not warn and continue: a warning about a slow inner
loop is read once and ignored forever, and the loop stays slow. This is
[ADR 0008](../../02-DECISIONS/0008-a-failed-step-fails-the-job.md) applied where the failure is
performance rather than an error.
## Two ways the prerequisites arrive
**On a machine the mesh manages** — a module declares them, and a hook turns them into
capabilities: the units enabled, the group granted, the pool created on the right driver. This
already works; it is what was done for the virtualisation daemon itself.
**On a machine the mesh does not manage** — the lab's own bootstrap does it. One command, on a
clean machine, that installs what is missing and configures it.
The second path is not a convenience. It is **required**, because the lab must work before the
mesh does, and a lab that could only be installed by a mesh would be unable to host the
development of the mesh that installs it.
Both paths end at the same verification. Whoever satisfied the prerequisites, the lab checks
them itself — because the lesson of the first failure is precisely that *something else said it
was done* is not evidence.
## The lab is the second thing installed by hand
Worth stating, because it looks like an exception and is not.
The node host is the one thing installed by hand on a machine
([research 006](../../01-RESEARCH/006-mesh-from-scratch/code-skeleton.md)): everything else
arrives through it. The lab is the same shape on a workstation — installed once, by hand, and
then everything about the mesh is developed inside it.
Two bootstraps, at two levels, for the same reason: **something has to be first, and pretending
otherwise produces a circularity that gets papered over with a script nobody exercises.**
## What a clean install actually needs
In order, on a machine with nothing:
1. **A virtualisation daemon**, running, with its socket enabled.
2. **Copy-on-write storage tooling** — the kernel side is usually already present; it is the
userspace half that is missing and that decides whether the driver is offered at all.
3. **A pool on that driver.** A loop-backed file is sufficient and needs no partitioning, which
matters: requiring a dedicated filesystem would make the lab uninstallable on a machine
already in use.
4. **Group membership** for the operator — which does **not** apply to sessions that already
existed. Observed directly: a shell whose process tree predated the grant could not reach
the daemon while a fresh lookup showed the membership present. The bootstrap has to say so,
or the first thing a person meets is a permission error that looks like a broken install.
5. **Verification**, as above, before anything is raised.
## Open
- **Whether the lab's bootstrap may install packages at all**, given that the mesh's rules
forbid installing by hand. The resolution is probably that the lab's bootstrap *is* the
sanctioned mechanism on an unmanaged machine, in the way the mesh's own first-node script is —
but that is an argument to record, not to assume.
- **What "adequate" means numerically.** The checks above are qualitative. A snapshot-time
threshold would catch a copy-on-write pool that is slow for some other reason, and would be a
real assertion rather than a proxy.
- **Whether the lab should own its pool** rather than using an existing one. Owning it makes the
driver guaranteed; sharing it avoids duplicating storage on a machine that already has a pool.
03-DESIGN/01-to-be/README.md
+2
View File
@@ -12,6 +12,8 @@ document is written and this one's status becomes `implemented`.
| [`00-work-breakdown.md`](00-work-breakdown.md) | How the decomposition gets built, in what order, and where a human must look | [ADR 0015](../../02-DECISIONS/0015-mesh-brokers-nodes-host-agents-think.md) |
| [`01-end-to-end-testing.md`](01-end-to-end-testing.md) | The lab: a real mesh a change can be run against before it reaches nodes | [ADR 0016](../../02-DECISIONS/0016-a-lab-node-is-a-virtual-machine.md), [0029](../../02-DECISIONS/0029-the-labs-first-scenario-has-no-pipeline.md) |
| [`02-scenario-declaration.md`](02-scenario-declaration.md) | What a scenario declares — the underlay, and what to place on it | [ADR 0031](../../02-DECISIONS/0031-the-lab-provides-the-underlay.md) |
| [`03-scenario-lifecycle.md`](03-scenario-lifecycle.md) | What happens to a scenario — raise, snapshot, restore, move, destroy | [ADR 0032](../../02-DECISIONS/0032-a-scenario-is-an-isolated-address-space.md) |
| [`04-lab-installation.md`](04-lab-installation.md) | Getting the lab onto a clean machine, and why it verifies capability rather than installation | [ADR 0008](../../02-DECISIONS/0008-a-failed-step-fails-the-job.md) |
## Not yet written