Every scenario that places a container runtime gives each of its machines `egress: true` — a node that runs modules pulls images from the internet, which is what a node does. The underlay-only scenarios (bootstrap-single, behind-nat, segmented-and-unforwardable, the-ordinary-shape, two-on-a-segment) stay sealed on purpose: an extra NIC would change the very reachability they are asserting about. `images:` keeps only the mesh's own — 55 third-party entries leave whole-mesh-full alone, and the machine fetches them itself by the digest its module.json already pins. The whole-mesh beds also say per machine which of ours they get: novox the substrate control plane and its own fifteen runtimes, ace its twenty-four, the two workstations one each. That is not a lab economy. An operator's workstation holds the images its own modules need, and giving these two the union would put some thirty gigabytes onto a thirty-gigabyte disk. bootstrap-with-registry.yml is deleted. It existed only to demonstrate the lab's registry, nothing referenced it, and there is nothing left for it to demonstrate. Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
57 lines
2.7 KiB
YAML
57 lines
2.7 KiB
YAML
# The DB-consumer chain a single node cannot host, proved across two machines.
|
||
#
|
||
# The app-postgres provider and the mesh's own substrate store both want host port 5432, so they
|
||
# cannot share a machine — the collision that blocked this chain single-node. Here the substrate
|
||
# (store, broker, control) lives on `anchor` and NOTHING else; `laptop` runs the whole chain —
|
||
# postgres and redis PROVIDERS plus the baserow and letta CONSUMERS that require them. Both
|
||
# machines sit on one shared segment and enrol into the one mesh; only enrolment crosses to anchor,
|
||
# over the underlay both machines already share. Provider and consumers are co-located on laptop, so
|
||
# no cross-node module comms and no overlay are needed — and the 5432-vs-substrate conflict is gone
|
||
# because the substrate store is on the OTHER node.
|
||
scenario: two-node-db
|
||
|
||
segments:
|
||
hosting:
|
||
kind: public
|
||
cidr: [192.0.2.0/24]
|
||
|
||
machines:
|
||
# The substrate ONLY: store, broker, control — three containers. Four gigabytes is plenty for a
|
||
# node that hosts no modules; the thrash the two-nodes bed warns of comes from stacking eleven
|
||
# containers on a node, which this one never does.
|
||
anchor:
|
||
at: { segment: hosting, address: [192.0.2.10] }
|
||
egress: true
|
||
inbound: allow
|
||
memory: 4GiB
|
||
cpus: 4
|
||
# The whole DB-consumer chain: postgres + redis providers, each a server and a broker-bound
|
||
# runtime, plus the baserow and letta consumer services and their tools runtimes — a dozen
|
||
# containers, two of them memory-hungry app servers (the Baserow all-in-one and the Letta server).
|
||
# At the 2GiB the two-nodes bed gives this machine it would thrash — its own anchor comment says
|
||
# so — and convergence would present as "the mesh hangs". Six gigabytes gives it room.
|
||
laptop:
|
||
at: { segment: hosting, address: [192.0.2.20] }
|
||
egress: true
|
||
inbound: allow
|
||
memory: 6GiB
|
||
cpus: 4
|
||
# The runtime images (280MB–600MB each) plus the service images — two of them heavy app images
|
||
# (Baserow ~1.5GB, Letta ~1.8GB) — are pulled from the internet over the uplink, so the
|
||
# same bytes land on this node twice. The pool default root disk exhausts mid-apply ("no space
|
||
# left on device"); sixty gigabytes holds the whole chain.
|
||
disk: 60GiB
|
||
|
||
images:
|
||
- mesh-control:development
|
||
# The per-module runtimes, built by scripts/build-module-runtime.sh and stocked here. Each carries
|
||
# its module's provisioner, so no separate mesh-provision-* image is listed — the runtime is the
|
||
# provisioner (ADR 0048).
|
||
- mesh-runtime-postgres:development
|
||
- mesh-runtime-redis:development
|
||
- mesh-runtime-baserow:development
|
||
- mesh-runtime-letta:development
|
||
|
||
place:
|
||
all: [host, runtime]
|