The lab raises machines on the virtualisation daemon, and a module may
mount a machine's socket only through the capability that grants it
(novox/hq ADR 0172). Also brings the resolver's tests to the setting
dnsmasq's listen addresses now come from, and to a module left out
rather than refused.
The nftables module's runtime mounts the file filtering.into names, to reload
the mesh's table; the mount check knew every other declaration of a path and
not this one, and the module's first build was refused for it.
What a person asks a machine's packet filter whatever filter answers: the
rules as enforced, reload the mesh's own, remove one rule set the mesh did
not write — named as the host reports it under ADR 0168. Every holder serves
all three; a running mesh widens its seat row at the next controller start.
Per-machine and mesh-wide settings could be set only from the
controller's command line. The settings verb runs settings set|clear,
passing the values inline, which the command now accepts as well as a
file (novox/hq issue 198).
The forward chain judged a packet relayed from one machine of the mesh
to another by this machine's own published ports, so two machines behind
the hub reached each other only on ports the hub published for itself
(novox/hq issue 196). In and out on the tunnel is now accepted, and the
machine it is for filters it.
take ends its preview with a digest and --yes names it, as the flip does; a
changed preview or an account older than the flip allows is refused. A module
the machine holds nothing for has nothing to compare, and --yes suffices. A
published port's reach is said as the machine reported it. Every secret the
module holds on the machine is listed with where it came from, and one the
mesh minted for a service whose data was found refuses unless --mint names it.
One judgement of a module's settings against its definition, in the catalogue:
settings set refuses what cannot compose or reaches nothing, naming node,
module, layer and key; Compose leaves out a module whose definition moved
under a stored setting, the envelope says so (left_out), plan and push say it
by name, and the machine is told everything else. A stray setting no longer
refuses the whole machine where it is read (issue 096).
The per-machine setting networks keeps a found network for a taken container,
on an adopted machine only; the container's declaration carries it and the
preview names it (rule 4).
A module's account was mintable only from the controller's command line,
so a rollout that gave a module one could not be finished through the
mesh's own tools (novox/hq issue 191). The issue verb runs module issue
for a module on a machine; the caller pushes the machine after.
The proxy answered every routed name to any request carrying it, so an
internal-only route would have been public under its internal name. Each
membership now carries what its module receives, from the same
composition as its received file, and every machine's private-network
address, the list the packet filter's "from the mesh" is. The proxy
follows its membership, serves internal names only to those machines and
itself, and keeps the file until the bus has spoken (novox/hq ADR 0167,
issue 191).
Given a repository and the files a branch changes (or the modules by name), plans answers with the tiers
the merge handler would produce — the modules those files touch, what packages their source, everything
reachable along the dependency relation — and what each tier does: built and sent to its machines, or
built and left because its policy records. Saved nowhere. On the seat as plans {repository, paths|modules}.
The first live plan took seventy-five modules along for a controller change: the builder packages the
controller's source, everything is built by the builder, so everything was reachable. A module built by
the build machine is not changed by a new build machine. Reachability now follows the code and build
edges only; built-by still orders a tier after the build machine and gates it on the machine's roll-out.
plans stop <id> ends a plan by hand: what was asked still builds and registers, nothing further is asked.
The mesh's seat answers plans — the recent plans with their tier, what each waits for and since when,
or one plan whole given its id — so the console reads a merge's progress where it reads everything
else, instead of a person reading the daemon's log.
#195 refused two modules beside a consumer that both answer a bound
provision — in every pass. The first pass exists only to answer what a node
offers, and a refusal there makes the machine vanish from every other node's
world (resolve.go says so for its sibling case): with novox refused for its
own acme-ca, ace's plan lost the vault and the identity provider and read
"nothing in this mesh provides secret". Seen live within minutes of the
rollout. The second pass refuses it, where it is asked, as before.
A provider is a (node, module) pair (design 23), and the pin — the one way a
consumer names its provider — named only the node. Two modules on one node
can both answer a provision (public-acme and step-ca both offer acme-ca on
novox), and then the resolver, given a pin naming that node, took the last
provider listed: a coin flip. The same ambiguity beside the consumer was
settled by a map walk — random per plan — which is how novox's own
route-proxy got its issuer (novox/hq #258).
- `pin <node> <provision> <from-node> <module>`: both halves, always. The
console gains `pin` and `unpin`. The provider may be on the consumer's own
node, since two modules beside it can both answer.
- The resolver refuses ambiguity instead of picking, across machines and
beside the consumer alike, naming every candidate as node/module and the
form of the pin that settles it. A plain capability that grants nothing
and serves nothing (three shells beside an editor) is not a choice to put
to anybody and stays as it was.
- provision_pin gains a nullable module (0050); records made before are
completed where the node they name answers once, and left for a person
where it answers twice (0051).
- The provider of something already satisfied is looked for among what was
assigned, not only what the walk has reached — a consumer reached before
the provider beside it no longer loses its binding.
- The start-time check that every declared verb is runnable samples each
verb's required arguments from its schema instead of three guessed keys.
Live consequence: a node that has two providers of one bound provision
assigned (novox: acme-ca) resolves only once pinned —
`pin novox acme-ca novox public-acme`.
mesh-vault joins the mesh's own set — mesh-scoped, delivering secret — because the controller seals
every minted credential with it, which is the test for a seat of the mesh's own; a second provider is
a second claimant, refused by name (issue 106). A report may carry the machine's profile, detected
again by the apply that reports, and the latest replaces the enrolled one: a machine that switched
its network manager is a machine whose uplink holder lacks a capability at its next push (issue 138).
The store's databases is not postgres's postgres_list_databases, and a holder may serve both. A
claim's serves names the seat's verbs the module implements for the role; absent, the module's own
tools must list every verb the seat promises, which is how a module named like its seat says they
are one and the same. Registration refuses a claim naming a verb the seat never promised, and the
credential's claims carry the claim's own verbs to the runtime.
For every module on every machine the controller composes what that instance serves — its machine's
address always, the module's plain address in a queue when it is alone or its definition says its
instances are interchangeable — the verbs of the seats it holds at the seats' subjects, where its
events land, and what it may reach, resolved the same way for the modules it invokes. Published
beside the node's declaration on `mesh.assignment.<node>.<module>`, last per subject in a stream
that allows direct reads, and the account may read exactly its own. Composed from the same records
the bus's accounts are, so what a runtime serves and what its account may are one composition.
`instances: interchangeable` is the one fact a definition states for it.
The shape issued is the shape the mesh already had, so nothing moves when the membership arrives;
the runtime that reads it instead of deriving it is the next piece.
Seeded additively into the seat's row at the controller's next start; a holder must list tools of
these names (design 33 §3), so this merges after the database engine's definition does.
An offer may say `"credential": {"own": "<secret>"}`: the provider's own secret is the credential
every consumer of that provision receives, in the shape of a pair credential. The vault keeps one
value, sealed to the provider, to every consumer that holds the provision and to the operator, all
under one generation stamp; a consumer binding later, or `secret rotate` on the provider's secret,
makes a fresh value and seals it to every holder in one act, and the rotate command sends every
holding machine together. An accepted value is sealed to the consumers of the moment and never
remade: a consumer binding after it is refused with the way out (ADR 0113). The named own secret
must say how it is taken (issue 180), so the provider's start applies the file.
A need carries the shared secret's name from either side of the machine boundary; the plan mints a
consumer's copy from the provider's value. Registered manifests keep their bytes.
`secret rotate <node> <module> <name>` makes the secret anew the way the first mint did, seals it
to the machine and the operator, and sends the machine, so the module starts again on the new value
— said in the log with who asked and when, never the value. Only for a secret whose definition says
`"taken": "at-start"`: an own secret is now a path, or {path, taken}, and a definition that says
nothing of how a secret is taken is refused with the word to write, because a credential rotated
under software that never reads it again is worse than one left alone (issue 179). `applied` is
refused by name until the staged form ADR 0114 decided is built; a value given to the mesh is
refused as ADR 0113 says. `rotate` is a verb on the controller's seat with two shapes — a pair
credential by provision, an own secret by machine, module and name — so the console can ask.
Registered manifests keep their bytes: a path alone is written back as a path.
The names region attributed a composed name to whichever labelled contribution a map yielded last.
A dashboard contributes its label to its route and to the identity provider, which must know the
public name for a redirect; so on one plan grafana.<domain> pointed at the proxy's machine and on
the next at the identity provider's, and the whole region flipped with it (forge issue 227). And a
module routed several times contributed no name at all, because the single-value reading of its
contributions is empty for the many shape.
Now every node's resolution is read first and the names are attributed across them at once: the
terminus serves the name — the provider that is not itself published under a labelled name through
another — walked in order, so one mesh yields one region. Name-agnostic, structural, deterministic.
The console's `build` tool answered "no build machine answered within 0s", handed a forge path to
git as written, and a build heard afterwards was recorded and never registered: recording and
registration lived only in the waiting caller, and the tool did not wait.
Now one function takes a build's outcome in — records it, parses the manifest, refuses a definition
naming an installation, registers the module with its source as the seat and path the request
carried — and both the waiting command and the daemon that follows the role's `built` event call
it. `build --wait 0` asks and returns with the id; `builds --log <id>` follows it. The seat verb
says `--self` for a repository given without a scheme.
The build-machine seat emits `started` and `log.<build id>` beside `built`. Every line the builder
speaks — each step, each command with its duration, and on failure the command's own output — goes
to stderr as before and onto the bus under the build's id, one subject per build, kept a week in
EVENTS with every other event. `builds --log <id>` reads it back from the stream with a consumer
that is gone when the reading is done, on the command line and as the controller's seat verb;
`builds` lists each build's id and `build` says the id it asked with.
Lines are core publishes with a sequence number, so a build is not slowed by an ack per line and a
gap is visible; `started` and `built` are awaited into the stream. The seat protocol widens
additively at the controller's next start; the holder's grant follows on the broker node's next
composition.
An access named by id (issue 153) resolves to the path the definition still carries when the
assignment says nothing, and the proof compares that — the same rule as a placed directory.
A definition names no host path (ADR 0112); an adopted machine keeps its
data where the predecessor put it. Two settings, validated like endpoints:
places: {<directory id>: <path> | {path, owner}}
accesses: {<access id>: <path>}
An access may now be declared by id (`{"id": "series", "mode": "read-write"}`)
and named in mounts, env and content as ${access:<id>}; the assignment
says where it is on this node, and an access nobody placed is refused by
name. A definition still carrying a path keeps it as the default the
assignment replaces. A placed directory takes the assignment's owner
where it says one. Resolved in composition, so the host receives paths
and owners exactly as before.
Issue 173's rule needs it: a mail provider serves its domain and an identity provider its issuer,
and neither is the definition's to state. Declared as ${setting:<key>}, filled from the layers after
the overrides; a key so asked for is not stray.
Three of novox/hq's group-4 leftovers, one branch.
Issue 173: a module's settings reached every route it contributed, every database it asked for and
every served fact its consumers read — a mail server's site name arrived at the proxy as a route
fact. A setting now overrides a key a contribution or served fact declares and adds none; a file
still merges any key, and a key nothing takes is named as stray instead of dropped silently.
Issue 174: the mesh's own files for a module — its bus credential, its merged config, its bindings —
were placed by the definition under /var/lib/mesh/<module>, 232 host paths in 50 definitions. A
directory may now say `place: "mesh"` and resolves to <root>/mesh/<module>; a directory beneath a
placed one may state its path as `${dir:<id>}/<rest>` and moves with it. The proof test resolves
both catalogues and compares: 48 definitions, no path moved. The controller's own manifest is
converted here; the catalogue in mesh-catalog.
ADR 0155: the installation check moves to registration. `module add` and a build's result both
refuse a definition that names an installation, in the check's words, with the way out; the build
stays recorded.
A check that only ran when somebody remembered a variable was a check nobody ran (novox/hq issue
134). MESH_CATALOGUE still overrides; the checks skip only when no catalogue can be found.
ADR 0121 decided it and deferred it as a delivering-seat migration; ADR 0122's aliases made it one
update and one alias (migration 0048). The former name resolves to it forever (novox/hq ADR 0156,
issue 123).
The check novox/hq issue 119 asks for before a definition stops naming where its data lives: two
catalogue checkouts, every module in both resolved with the controller's own rule and compared whole.
InstallationProblems judges every value the mesh acts on for a name under a public top-level domain
or a public address, with prose, the world's registries, resolvers and certificate authorities
exempt, and a name a resource means on purpose declared with its reason (names-on-purpose). Run by
module check and a catalogue-wide test, not yet at registration, while the declared list shrinks.
${setting:<key>} fills a file from the assignment's settings and is refused when nothing set it.
A build context may live on the git seat; the request carries the seat's clone base (novox/hq ADR
0112, ADR 0155, issues 122 and 134).
A seat's protocol lives in the store (migration 0047; seeded additively), a served verb carries its
description and schema, holding a mesh seat requires serving its verbs, a node-scoped seat's tool
carries the node, and the control plane serves status, nodes, node, modules, seats, builds, plan,
assign, unassign, push, build and tools on its seat by running the same commands (novox/hq ADR 0132,
ADR 0154, design 33). A grant of * reaches a role's tools; seat:<seat>.<verb> grants one.
A module contributes a label; the mesh joins it with the node's domains and
the provider serves the result — and the module itself was never told.
Software that must know its own address (a login redirect, a canonical URL,
an issuer) had it written into the manifest as a literal: a domain in a
definition, wrong on every other machine (ADR 0112). Found converting
grafana's keycloak login for ace, where it forced GF_SERVER_ROOT_URL and
keycloak's issuer back into manifests.
The binding for a requirement a module contributes to now carries `name`
and `internal-name` (or `names` by local name for several contributions),
and `${bound:<requirement>:name}` / `:internal-name` (`:name-<local>`) fill
files from it. Both come from the one function the provider's received
file is composed by, so the proxy and the module cannot disagree about the
name. Absent when nothing was composed, so a file asking for a name on a
node with no public domain is refused, not rendered empty.
Also: `${bound:…}` could not name a requirement answered by a node-scoped
provider on the same machine — its binding file was written (from `here`)
but the placeholders only looked at the mesh's needs. Filled from the same
answer now.
"3478/udp" read as one token was not a port, so it passed through and the
runtime published it wherever it liked: on ace, unifi's STUN and discovery
landed on random machine ports while every TCP pin beside them held. The
protocol is split off, the number is assigned as for any short form, and
the suffix rides along on the outside.
invokes: a manifest word that becomes exactly the publish grant a person's account gets (ADR 0152),
derived by the same composition; refused at parse when it names no tool. module check <file|dir>...
runs what registration runs with no store, for a manifest in any repository (hq issue 148).
novox/hq ADR 0151 (issues 139, 157). <label>.<node>.internal is answered
by every resolver as 'anything under that node goes to that node', so
the node in a route's internal name must be the one whose proxy answers
it; composed under the consumer's own name it sent a client to a machine
with nothing listening whenever the proxy ran elsewhere. Composed under
the serving node now — the same machine wherever the proxy runs beside
the module, so nothing changes on a mesh with one hub.
A routed public name gets no .internal alias any more: the roster
publishes it as itself, once. The alias resolved and nothing served it.
`assign ace postgres` made the control plane's own store unresolvable:
postgres's manifest claims mesh-store, nobody was ever recorded as its
holder, and with two eligible assignments and nothing on record both
claimed and both were refused — novox's included. ADR 0110 says the
assignment holds, by a deliberate act; ADR 0131 gave the record its force
but left a seat nobody handed over held by whichever assignment happened
to be alone.
Before acting on an assignment the controller now writes the derived
answer down: every mesh-scoped seat the store knows, resolved to exactly
one holder with nothing on record, gets that holder recorded — the same
record `seat <name> --to <node>/<module>` makes by hand. The next
assignment able to hold the seat then stands beside the holder, eligible
and silent. A seat with two derived claimants is left for a person; a
seat on record is never rewritten; seats the store does not list stay
held by derivation as before. And the refusal, when it still happens,
names the handover that records the holder.
Verified: catalogue tests against the real catalogue; the cmd suite
against a store (the only failure, TestConvergingPreviewsThenChanges…,
fails identically on main).
novox/hq ADR 0148, step 3. Every container got the whole roster as
--add-host entries at creation and nothing re-read them (issues 109,
135); once the roster was in the digest so that could be caught, one
name moving anywhere replaced every container in the mesh (issue 151).
A container resolves through its machine's resolver, which the resolver
module tells the runtime about once per machine. A module's own hosts
entries stay exactly as declared.
Also brings the resolver tests up to the catalogue as it now is: the
runtime is reloaded (never restarted) and given live-restore, and the
resolver answers by address, not by interface (issue 110).
novox/hq 04-ISSUES/142. The name a machine runs a binary by is not always
the name of the package that built it. The host's command is cmd/mesh-host
and every machine runs it as nox-mesh-host — the path it is installed at,
the name in its unit, and the name its launcher looks for inside a
delivered version.
So the first delivered host version landed as `mesh-host`, the host
reported "created /usr/lib/nox-mesh-host/versions/2681d936b949: 1 file(s)",
everything said success, and the launcher would never have seen it. Found
by listing the directory instead of believing the line.
An artifact may now say what its executable is called. Saying nothing
keeps what the compiler would have chosen — the package's name — so
nothing that worked before changes.
novox/hq 04-ISSUES/142, and the second of the two things ADR 0141's own
insight named: "a version cannot reach the path". A component is unpacked
into a directory named for its version so it can read its own version from
its path — and an archive named a fixed path in the manifest with nothing
interpolating the build into it, so nothing could ask for
.../versions/<version>/ and every machine took a hand-placed fallback.
A resource using an archive or a bundle may now say ${version} in any of
its values. No artifact name in the reference: the resource already says
which artifact it is for, and a second name is a second thing to keep in
step.
The version is the artifact's digest, short, and not the commit. Two
builds of one commit are meant to be the same bytes — every toolchain here
is -trimpath for that reason — so a content-addressed version means an
unchanged build resolves to the path it already had. A commit-named path
would move for an identical binary and recreate everything that reads it.
An image is refused one, with a reason: an image is not unpacked, so it
has no versioned place. Left alone it would reach a machine as literal
text and be created as a directory called ${version}.
novox/hq 04-ISSUES/142. A bundle is refused if it names what it is built
from, because a bundle is the module's own directory compiled whole and
naming a source would be describing its own build. That reason holds for
an interpreted language and cannot hold for a compiled one.
A repository written in Go carries several commands — the host and its
bootstrap live in one — and "the module's own directory" is then not a
package at all. So a compiled bundle may say which package, and says the
module root by saying nothing. The refusal stands for every interpreted
bundle, which is what it was written for.
Found by writing the host's manifest, which is the first bundle in a
compiled language this mesh has had.
novox/hq ADR 0147. ca-trust carries a script and a unit; the one thing
neither can state is where the authority is, because that is a fact about
the mesh. This checks the rendering — the script fetches from the bound
address and is executable, and the unit runs it both ways, install and
remove. The verification itself is the lab's.
Local is not a boundary this mesh draws. A service here is callable by everything
else here, whatever form either takes — a package with a unit, a binary, a
container. Whether a caller sits in a container was never meant to change the
answer, and the only reason it did was that this chain asked about addresses: a
caller on the machine carries the machine's address, a caller in one of its
containers carries a bridge address, and a rule naming the former silently refused
the latter.
One rule for every service here, replacing the line-per-port added an hour ago,
which only ever covered the ports somebody remembered to think about. The three
reaches are now three lines: on this machine, over the private network, from
anywhere.
The tests assert per chain body, because the forward chain carries the same line in
the same words and an assertion on the whole file passed with the input chain's copy
deleted — which is what ADR 0137's own tests say to do and this file was not doing.
A port declared from the mesh admitted the machines' own addresses on the private
network. A container reaching a port on the machine it runs on comes from a bridge,
matching none of them — and where the container runtime routes directly, that packet
is delivered to this machine rather than forwarded, so the forward chain's allowance
never saw it either.
ADR 0100 requires this to work: the store is reachable 'from a container on the node
itself'. It was, through a rule the predecessor left, which allowed the private
ranges wholesale. Converging the machine replaced that with the four overlay
addresses and closed it.
Measured, and it was an outage: every module reaching another by its machine's own
name timed out for eleven hours while the mesh reported the machine healthy. A web
application logged 'connection to server at novox.internal (10.10.0.1), port 6852
failed: timeout expired' throughout.
Asked for by the link it arrives on, for the reason the forward chain no longer
names an address: a range describes one machine and goes stale in silence. A port
open to everything needs no such line.
Everything downstream reads the port: the provider is told where to reach the
consumer, and the redirection that turns a declared port into the number the
machine published is keyed on it. A route naming only its endpoint left the proxy
with no port at all, and a proxy with no port has nothing to dial.
Caught after the catalogue had already been changed to name endpoints and before
the mesh picked those manifests up, which is the only reason nothing broke: every
module's manifest is behind its source right now, so the plan still renders from
the old shape.
The declared port, not the machine one — the redirection happens later and is keyed
on the declared number, so filling in the machine port here would be redirected
twice or not at all. A route that repeats a port keeps it.
A test that reads every manifest in a catalogue checkout and parses it with the
control plane's own parser, rather than asserting against a fixture: whether the
manifests as written are accepted is the question, and a copy of one proves nothing
about the other seventy-one.
Skipped unless MESH_CATALOGUE names a checkout, so it costs nothing in ordinary
runs and is there when the catalogue changes shape. It also refuses to pass if no
endpoint is named, because a run that validated nothing would otherwise read as
success.
jschoubben
mesh-admin