Follow-up to the review of #186 (hq issue 353, hq #253). Do not merge without review.
The four review points
The real judgement is tested. The asker's wiring moved into servingAsker. TestTheServingAskerJudgesTheGrantForReal builds the serving asker and calls its grantHeld: with nothing to read it must say not held. These mutations were each run and each fails a test: removing grantHeld: grantHeldIn(open, conn); forcing held = true in grantHeldBy; making controllerMayAsk always yes; reading an answer without in_force as a yes; and listBehind ignoring a compose error.
The controller's own grant, not any change to the user list.controllerMayAsk composes the users from the records (broker.Users) and asks broker.MayPublish of the controller's own permissions for accept.ask.mesh-controller. Changes to other users no longer hold asking back. Test: TestTheControllersOwnGrantToAskIsReadFromTheComposition.
Waits for the bus's reload, not only the send. The second fact is asked of the bus's own module on the machine holding the bus: nats_user_can for user controller, publish, the ask subject. It must answer in_force: true, which means the user list there allows the publish and the server's config_load_time is after the list's last write. This needs mesh-catalog #166. A bus module older than that answer is not a yes. Test: TestTheBusModulesAnswerIsInForceOnlyWhenItSaysSo.
Fails closed.judgeGrant treats every error as not held and says why (the conditions that need the operator are raised as undelivered, as in #186). brokerBehind treats a list that cannot be composed as behind (listBehind). Before, a plan that could not be worked out read as not behind.
Behaviour change to note (point 4, push): when the bus machine's plan cannot be worked out, a push now puts that machine first, and it fails there with the plan's own reason. A rollout through sendRollout (all or nothing) now refuses instead of sending code whose grants the bus may not hold. Before, both went ahead.
Checks. The branch is on main's head (7477346), so the head is the merge result. sh merge-check.sh passed locally with mesh-host at main b0d03be and a throwaway PostgreSQL: gofmt, vet, go test -race, 26 packages ok.
Order. mesh-catalog #166 first, live on the machine holding the bus. Then this. If this merges first, the controller asks the operator nothing and says why ("the bus module does not say whether a grant is in force"), raised as undelivered.
Follow-up to the review of #186 (hq issue 353, hq #253). Do not merge without review.
**The four review points**
1. **The real judgement is tested.** The asker's wiring moved into `servingAsker`. `TestTheServingAskerJudgesTheGrantForReal` builds the serving asker and calls its `grantHeld`: with nothing to read it must say not held. These mutations were each run and each fails a test: removing `grantHeld: grantHeldIn(open, conn)`; forcing `held = true` in `grantHeldBy`; making `controllerMayAsk` always yes; reading an answer without `in_force` as a yes; and `listBehind` ignoring a compose error.
2. **The controller's own grant, not any change to the user list.** `controllerMayAsk` composes the users from the records (`broker.Users`) and asks `broker.MayPublish` of the controller's own permissions for `accept.ask.mesh-controller`. Changes to other users no longer hold asking back. Test: `TestTheControllersOwnGrantToAskIsReadFromTheComposition`.
3. **Waits for the bus's reload, not only the send.** The second fact is asked of the bus's own module on the machine holding the bus: `nats_user_can` for user `controller`, publish, the ask subject. It must answer `in_force: true`, which means the user list there allows the publish and the server's `config_load_time` is after the list's last write. This needs **mesh-catalog #166**. A bus module older than that answer is not a yes. Test: `TestTheBusModulesAnswerIsInForceOnlyWhenItSaysSo`.
4. **Fails closed.** `judgeGrant` treats every error as not held and says why (the conditions that need the operator are raised as undelivered, as in #186). `brokerBehind` treats a list that cannot be composed as behind (`listBehind`). Before, a plan that could not be worked out read as not behind.
**Behaviour change to note (point 4, push):** when the bus machine's plan cannot be worked out, a push now puts that machine first, and it fails there with the plan's own reason. A rollout through `sendRollout` (all or nothing) now refuses instead of sending code whose grants the bus may not hold. Before, both went ahead.
**Checks.** The branch is on main's head (7477346), so the head is the merge result. `sh merge-check.sh` passed locally with mesh-host at main b0d03be and a throwaway PostgreSQL: gofmt, vet, `go test -race`, 26 packages ok.
**Order.** mesh-catalog #166 first, live on the machine holding the bus. Then this. If this merges first, the controller asks the operator nothing and says why ("the bus module does not say whether a grant is in force"), raised as undelivered.
Deliverynovox/mesh-controller@f996cf7b6b9b — delivered since 2026-10-12T11:43:08Z
Delivery plan — builds mesh-controller → novox; no bus step
tier 0: mesh-controller
novox: receives mesh-controller
Machines
mesh-controller on novox: passed — healthy 3 times over 3m40s
mesh-controller on the rest running it: sent
Transitions
2026-10-12 11:25 (new) → proposed (announced): novox/mesh-controller#188's head announced
2026-10-12 11:32 proposed → checked (checked): the verdict names this commit
2026-10-12 11:32 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
2026-10-12 11:36 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
2026-10-12 11:36 published → delivering (go): its walk started: let go by this owner in its group's order, by a person, or on the controller's own path — not a failed walk that carried a later commit for it, after which it waits for a walk of its own commit
2026-10-12 11:43 delivering → delivered (done): its walk is done, carrying its own commit: every machine of its deploy plan passed, or was left as its policy says
The commit's note under refs/notes/mesh-plan keeps every transition: git log --notes=mesh-plan.
<!-- mesh-delivery:view -->
**Delivery** `novox/mesh-controller@f996cf7b6b9b` — **delivered** since 2026-10-12T11:43:08Z
**Delivery plan** — builds mesh-controller → novox; no bus step
- tier 0: mesh-controller
- novox: receives mesh-controller
**Machines**
- mesh-controller on novox: passed — healthy 3 times over 3m40s
- mesh-controller on the rest running it: sent
**Transitions**
- 2026-10-12 11:25 (new) → proposed (announced): novox/mesh-controller#188's head announced
- 2026-10-12 11:32 proposed → checked (checked): the verdict names this commit
- 2026-10-12 11:32 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
- 2026-10-12 11:36 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
- 2026-10-12 11:36 published → delivering (go): its walk started: let go by this owner in its group's order, by a person, or on the controller's own path — not a failed walk that carried a later commit for it, after which it waits for a walk of its own commit
- 2026-10-12 11:43 delivering → delivered (done): its walk is done, carrying its own commit: every machine of its deploy plan passed, or was left as its policy says
The commit's note under `refs/notes/mesh-plan` keeps every transition: `git log --notes=mesh-plan`.
build-agent keeps data (ADR 0233): what it holds is backed up before it moves
route-proxy provides route: its consumers are sent again after it
route-proxy keeps data (ADR 0233): what it holds is backed up before it moves
Run by the build seat on novox as build-1791568027996482179 (builds --log build-1791568027996482179).
--- the gate: build-agent, mesh-controller, route-proxy
mesh-controller: ok, 54 tool(s); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-09T17:45:02Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-09T17:45:02Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 3 module(s) in 3 tier(s): build-agent, mesh-controller, route-proxy
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.048s
ok github.com/novox/mesh-controller/cmd/mesh-controller 256.705s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.016s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.030s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.019s
ok github.com/novox/mesh-controller/examples/route-proxy 1.134s
ok github.com/novox/mesh-controller/internal/artifacts 1.066s
ok github.com/novox/mesh-controller/internal/broker 6.841s
ok github.com/novox/mesh-controller/internal/builder 18.596s
ok github.com/novox/mesh-controller/internal/catalogue 4.981s
ok github.com/novox/mesh-controller/internal/conditions 2.691s
ok github.com/novox/mesh-controller/internal/envfile 1.032s
ok github.com/novox/mesh-controller/internal/facts 1.035s
ok github.com/novox/mesh-controller/internal/identity 17.555s
ok github.com/novox/mesh-controller/internal/inventory 106.986s
ok github.com/novox/mesh-controller/internal/lease 17.872s
ok github.com/novox/mesh-controller/internal/licences 16.985s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.013s
ok github.com/novox/mesh-controller/internal/link 57.356s
ok github.com/novox/mesh-controller/internal/lint 2.111s
ok github.com/novox/mesh-controller/internal/outward 1.036s
ok github.com/novox/mesh-controller/internal/overlay 1.026s
ok github.com/novox/mesh-controller/internal/secrets 1.404s
ok github.com/novox/mesh-controller/internal/store 10.057s
ok github.com/novox/mesh-controller/internal/testbus 1.176s
ok github.com/novox/mesh-controller/internal/token 1.019s
**Merge check** at `00bcdccd`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller, route-proxy)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step
- tier 0: mesh-controller
- tier 1: build-agent
- tier 2: route-proxy
- ace: receives build-agent, route-proxy
- g14: receives build-agent
- novox: receives mesh-controller, build-agent, route-proxy
- shanks: receives build-agent
- build-agent keeps data (ADR 0233): what it holds is backed up before it moves
- route-proxy provides route: its consumers are sent again after it
- route-proxy keeps data (ADR 0233): what it holds is backed up before it moves
Run by the build seat on novox as `build-1791568027996482179` (`builds --log build-1791568027996482179`).
```
--- the gate: build-agent, mesh-controller, route-proxy
mesh-controller: ok, 54 tool(s); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-09T17:45:02Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-09T17:45:02Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 3 module(s) in 3 tier(s): build-agent, mesh-controller, route-proxy
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.048s
ok github.com/novox/mesh-controller/cmd/mesh-controller 256.705s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.016s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.030s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.019s
ok github.com/novox/mesh-controller/examples/route-proxy 1.134s
ok github.com/novox/mesh-controller/internal/artifacts 1.066s
ok github.com/novox/mesh-controller/internal/broker 6.841s
ok github.com/novox/mesh-controller/internal/builder 18.596s
ok github.com/novox/mesh-controller/internal/catalogue 4.981s
ok github.com/novox/mesh-controller/internal/conditions 2.691s
ok github.com/novox/mesh-controller/internal/envfile 1.032s
ok github.com/novox/mesh-controller/internal/facts 1.035s
ok github.com/novox/mesh-controller/internal/identity 17.555s
ok github.com/novox/mesh-controller/internal/inventory 106.986s
ok github.com/novox/mesh-controller/internal/lease 17.872s
ok github.com/novox/mesh-controller/internal/licences 16.985s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.013s
ok github.com/novox/mesh-controller/internal/link 57.356s
ok github.com/novox/mesh-controller/internal/lint 2.111s
ok github.com/novox/mesh-controller/internal/outward 1.036s
ok github.com/novox/mesh-controller/internal/overlay 1.026s
ok github.com/novox/mesh-controller/internal/secrets 1.404s
ok github.com/novox/mesh-controller/internal/store 10.057s
ok github.com/novox/mesh-controller/internal/testbus 1.176s
ok github.com/novox/mesh-controller/internal/token 1.019s
```
mesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-controller)
mesh/repo-check: PASS — its merge-check.sh passed
Delivery plan — builds mesh-controller → novox; no bus step
tier 0: mesh-controller
novox: receives mesh-controller
read by no module's build: cmd/mesh-controller/asker_test.go, cmd/mesh-controller/delivery_order_test.go
Run by the build seat on shanks as build-1791800029376562746 (builds --log build-1791800029376562746).
--- the gate: mesh-controller
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T10:08:43Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T10:08:43Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-controller
notes:
- cmd/mesh-controller/asker_test.go and cmd/mesh-controller/delivery_order_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.042s
ok github.com/novox/mesh-controller/cmd/mesh-controller 253.837s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.053s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.068s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.040s
ok github.com/novox/mesh-controller/examples/route-proxy 1.226s
ok github.com/novox/mesh-controller/internal/artifacts 1.112s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.489s
ok github.com/novox/mesh-controller/internal/builder 20.074s
ok github.com/novox/mesh-controller/internal/catalogue 8.192s
ok github.com/novox/mesh-controller/internal/conditions 2.634s
ok github.com/novox/mesh-controller/internal/envfile 1.053s
ok github.com/novox/mesh-controller/internal/facts 1.053s
ok github.com/novox/mesh-controller/internal/identity 5.109s
ok github.com/novox/mesh-controller/internal/inventory 101.397s
ok github.com/novox/mesh-controller/internal/lease 17.729s
ok github.com/novox/mesh-controller/internal/licences 4.647s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.054s
ok github.com/novox/mesh-controller/internal/link 62.760s
ok github.com/novox/mesh-controller/internal/lint 2.468s
ok github.com/novox/mesh-controller/internal/outward 1.093s
ok github.com/novox/mesh-controller/internal/overlay 1.076s
ok github.com/novox/mesh-controller/internal/secrets 1.580s
ok github.com/novox/mesh-controller/internal/store 3.540s
ok github.com/novox/mesh-controller/internal/testbus 1.259s
ok github.com/novox/mesh-controller/internal/token 1.017s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.47s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.46s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.43s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.40s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.60s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.77s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.30s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.31s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.32s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.28s)
**Merge check** at `473d597f`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-controller)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds mesh-controller → novox; no bus step
- tier 0: mesh-controller
- novox: receives mesh-controller
- read by no module's build: cmd/mesh-controller/asker_test.go, cmd/mesh-controller/delivery_order_test.go
Run by the build seat on shanks as `build-1791800029376562746` (`builds --log build-1791800029376562746`).
```
--- the gate: mesh-controller
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T10:08:43Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T10:08:43Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-controller
notes:
- cmd/mesh-controller/asker_test.go and cmd/mesh-controller/delivery_order_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.042s
ok github.com/novox/mesh-controller/cmd/mesh-controller 253.837s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.053s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.068s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.040s
ok github.com/novox/mesh-controller/examples/route-proxy 1.226s
ok github.com/novox/mesh-controller/internal/artifacts 1.112s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.489s
ok github.com/novox/mesh-controller/internal/builder 20.074s
ok github.com/novox/mesh-controller/internal/catalogue 8.192s
ok github.com/novox/mesh-controller/internal/conditions 2.634s
ok github.com/novox/mesh-controller/internal/envfile 1.053s
ok github.com/novox/mesh-controller/internal/facts 1.053s
ok github.com/novox/mesh-controller/internal/identity 5.109s
ok github.com/novox/mesh-controller/internal/inventory 101.397s
ok github.com/novox/mesh-controller/internal/lease 17.729s
ok github.com/novox/mesh-controller/internal/licences 4.647s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.054s
ok github.com/novox/mesh-controller/internal/link 62.760s
ok github.com/novox/mesh-controller/internal/lint 2.468s
ok github.com/novox/mesh-controller/internal/outward 1.093s
ok github.com/novox/mesh-controller/internal/overlay 1.076s
ok github.com/novox/mesh-controller/internal/secrets 1.580s
ok github.com/novox/mesh-controller/internal/store 3.540s
ok github.com/novox/mesh-controller/internal/testbus 1.259s
ok github.com/novox/mesh-controller/internal/token 1.017s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.47s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.46s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.43s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.40s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.60s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.77s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.30s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.31s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.32s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.28s)
```
Verdict: request changes, at head 473d597 (rebased onto main 22a2f73; the old head was 00bcdcc).
Independent review for hq issue 353.
Rebase: asker_test.go conflicted with main's new test for issue 369 (TestARefusalIsAskedAgainAndTheUndeliveredConditionClearsOnceTaken). Both sides only added code, so I kept both. Main had also added a second caller, the settings proposer (proposals.go:594, ADR 0277), that called grantHeldIn(open). I changed it to grantHeldIn(open, conn), so the proposer now judges the grant the same way.
What holds:
TestNothingIsAskedWhileTheBusLacksTheControllersGrant exists. It fails without the gate: with if a.grantHeld != nil disabled in asker.go it fails three ways (asked while the bus lacks the grant, said 0 times, no undelivered condition).
The four new tests pass. They check the two facts, the composed grant, the bus module's answer (an answer without in_force is not a yes) and that the serving asker is wired for real.
listBehind fails closed and is tested.
go test -race ./... passes on all 26 packages locally (the store tests skip without PostgreSQL; mesh/repo-check ran them).
Checks at 473d597: mesh/merge-gate pass (4 of 4 compose, no bus step), mesh/repo-check pass, mesh/delivery ready.
Why I ask for changes: the order with mesh-catalog #166.
This head is mergeable now, and nothing stops it landing before #166 is live on novox. If it does, the bus module there answers nats_user_can without in_force. grantInForce reads that as not held, and it stays not held until #166 is live. In that order:
The asker asks nothing at all. Every condition that needs the operator is withheld, including the Approve/Decline asks that reach Telegram. They are raised only as asks-undelivered ("Questions for you not delivered").
settings propose refuses every proposal ("the operator cannot be asked yet: …") and points to the terminal.
Only rehearse still publishes, because it is not gated.
This is not a short window between two merges. mesh-delivery.what-if says #166 needs a person's bus upgrade (mesh-controller.bus upgrade), and that holds every delivery to novox until it runs. So the asks stay silent until the operator runs that upgrade.
The condition names the wrong action. busGrantIn wraps every not-in-force reason as "…; push novox carries it, and the bus reloads it in place". For an older bus module that is false. A push changes nothing; only #166 delivered by the bus upgrade does. This breaks the issue's acceptance ("says which push carries the grant") in exactly this order.
A guard is needed. Merge order alone is not enough here, because the head is green and mergeable on its own. I suggest:
When the bus module's answer has no in_force, fall back to the judgement on main today (#186: brokerBehind, whether the list sent to the bus's machine is the one composed now). Say in why that the reload is not confirmed because the bus module is older. Today's behaviour then stays the floor, and the 9-second reload gap is closed once #166 is live. Errors and unreachable modules keep failing closed as they do now.
Give a reason that names the actual remedy for each case. For an older module, that is the nats bus upgrade carrying mesh-catalog #166, not push. For a list not allowed on the machine, it is push <node>. For a reload not yet done, it is to wait about 5 s.
Add a test where the asker meets an answer without in_force and the list sent is current: it asks. Add a second case where the list is behind: it does not ask, and the reason names the push.
If you would rather keep the strict reading (an older module is never a yes), the alternative is a hard merge rule: #188 merges only after nats_user_can on novox answers with in_force. Then the wrong push text in point 4 still has to be fixed. I recommend the fallback.
Verdict: request changes, at head 473d597 (rebased onto main 22a2f73; the old head was 00bcdcc).
Independent review for hq issue 353.
Rebase: asker_test.go conflicted with main's new test for issue 369 (TestARefusalIsAskedAgainAndTheUndeliveredConditionClearsOnceTaken). Both sides only added code, so I kept both. Main had also added a second caller, the settings proposer (proposals.go:594, ADR 0277), that called `grantHeldIn(open)`. I changed it to `grantHeldIn(open, conn)`, so the proposer now judges the grant the same way.
What holds:
- TestNothingIsAskedWhileTheBusLacksTheControllersGrant exists. It fails without the gate: with `if a.grantHeld != nil` disabled in asker.go it fails three ways (asked while the bus lacks the grant, said 0 times, no undelivered condition).
- The four new tests pass. They check the two facts, the composed grant, the bus module's answer (an answer without in_force is not a yes) and that the serving asker is wired for real.
- listBehind fails closed and is tested.
- `go test -race ./...` passes on all 26 packages locally (the store tests skip without PostgreSQL; mesh/repo-check ran them).
- Checks at 473d597: mesh/merge-gate pass (4 of 4 compose, no bus step), mesh/repo-check pass, mesh/delivery ready.
Why I ask for changes: the order with mesh-catalog #166.
This head is mergeable now, and nothing stops it landing before #166 is live on novox. If it does, the bus module there answers nats_user_can without in_force. grantInForce reads that as not held, and it stays not held until #166 is live. In that order:
1. The asker asks nothing at all. Every condition that needs the operator is withheld, including the Approve/Decline asks that reach Telegram. They are raised only as asks-undelivered ("Questions for you not delivered").
2. `settings propose` refuses every proposal ("the operator cannot be asked yet: …") and points to the terminal.
3. Only `rehearse` still publishes, because it is not gated.
This is not a short window between two merges. mesh-delivery.what-if says #166 needs a person's bus upgrade (`mesh-controller.bus upgrade`), and that holds every delivery to novox until it runs. So the asks stay silent until the operator runs that upgrade.
4. The condition names the wrong action. busGrantIn wraps every not-in-force reason as "…; `push novox` carries it, and the bus reloads it in place". For an older bus module that is false. A push changes nothing; only #166 delivered by the bus upgrade does. This breaks the issue's acceptance ("says which push carries the grant") in exactly this order.
A guard is needed. Merge order alone is not enough here, because the head is green and mergeable on its own. I suggest:
- When the bus module's answer has no in_force, fall back to the judgement on main today (#186: brokerBehind, whether the list sent to the bus's machine is the one composed now). Say in why that the reload is not confirmed because the bus module is older. Today's behaviour then stays the floor, and the 9-second reload gap is closed once #166 is live. Errors and unreachable modules keep failing closed as they do now.
- Give a reason that names the actual remedy for each case. For an older module, that is the nats bus upgrade carrying mesh-catalog #166, not `push`. For a list not allowed on the machine, it is `push <node>`. For a reload not yet done, it is to wait about 5 s.
- Add a test where the asker meets an answer without in_force and the list sent is current: it asks. Add a second case where the list is behind: it does not ask, and the reason names the push.
If you would rather keep the strict reading (an older module is never a yes), the alternative is a hard merge rule: #188 merges only after `nats_user_can` on novox answers with in_force. Then the wrong `push` text in point 4 still has to be fixed. I recommend the fallback.
build-agent keeps data (ADR 0233): what it holds is backed up before it moves
route-proxy provides route: its consumers are sent again after it
route-proxy keeps data (ADR 0233): what it holds is backed up before it moves
grants step: if this changes what the bus's user list says (a new tool, subject or user), novox is sent that list alone, every build there kept, before build-agent, route-proxy is judged on its first machine
Run by the build seat on g14 as build-1791801278187283201 (builds --log build-1791801278187283201).
--- the gate: build-agent, mesh-controller, route-proxy
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T10:31:44Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T10:31:44Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 3 module(s) in 3 tier(s): build-agent, mesh-controller, route-proxy
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.088s
ok github.com/novox/mesh-controller/cmd/mesh-controller 319.627s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.022s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.029s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.022s
ok github.com/novox/mesh-controller/examples/route-proxy 1.194s
ok github.com/novox/mesh-controller/internal/artifacts 1.062s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 8.758s
ok github.com/novox/mesh-controller/internal/builder 23.039s
ok github.com/novox/mesh-controller/internal/catalogue 9.270s
ok github.com/novox/mesh-controller/internal/conditions 4.932s
ok github.com/novox/mesh-controller/internal/envfile 1.023s
ok github.com/novox/mesh-controller/internal/facts 1.029s
ok github.com/novox/mesh-controller/internal/identity 17.450s
ok github.com/novox/mesh-controller/internal/inventory 165.504s
ok github.com/novox/mesh-controller/internal/lease 17.738s
ok github.com/novox/mesh-controller/internal/licences 15.950s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.015s
ok github.com/novox/mesh-controller/internal/link 87.829s
ok github.com/novox/mesh-controller/internal/lint 2.321s
ok github.com/novox/mesh-controller/internal/outward 1.043s
ok github.com/novox/mesh-controller/internal/overlay 1.033s
ok github.com/novox/mesh-controller/internal/secrets 1.466s
ok github.com/novox/mesh-controller/internal/store 6.000s
ok github.com/novox/mesh-controller/internal/testbus 1.356s
ok github.com/novox/mesh-controller/internal/token 1.023s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.34s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.37s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.36s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.35s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.33s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.31s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.36s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.22s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.21s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.23s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.21s)
**Merge check** at `c3371ae6`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller, route-proxy)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step
- tier 0: mesh-controller
- tier 1: build-agent
- tier 2: route-proxy
- ace: receives build-agent, route-proxy
- g14: receives build-agent
- novox: receives mesh-controller, build-agent, route-proxy
- shanks: receives build-agent
- build-agent keeps data (ADR 0233): what it holds is backed up before it moves
- route-proxy provides route: its consumers are sent again after it
- route-proxy keeps data (ADR 0233): what it holds is backed up before it moves
- grants step: if this changes what the bus's user list says (a new tool, subject or user), novox is sent that list alone, every build there kept, before build-agent, route-proxy is judged on its first machine
Run by the build seat on g14 as `build-1791801278187283201` (`builds --log build-1791801278187283201`).
```
--- the gate: build-agent, mesh-controller, route-proxy
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T10:31:44Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T10:31:44Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 3 module(s) in 3 tier(s): build-agent, mesh-controller, route-proxy
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.088s
ok github.com/novox/mesh-controller/cmd/mesh-controller 319.627s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.022s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.029s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.022s
ok github.com/novox/mesh-controller/examples/route-proxy 1.194s
ok github.com/novox/mesh-controller/internal/artifacts 1.062s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 8.758s
ok github.com/novox/mesh-controller/internal/builder 23.039s
ok github.com/novox/mesh-controller/internal/catalogue 9.270s
ok github.com/novox/mesh-controller/internal/conditions 4.932s
ok github.com/novox/mesh-controller/internal/envfile 1.023s
ok github.com/novox/mesh-controller/internal/facts 1.029s
ok github.com/novox/mesh-controller/internal/identity 17.450s
ok github.com/novox/mesh-controller/internal/inventory 165.504s
ok github.com/novox/mesh-controller/internal/lease 17.738s
ok github.com/novox/mesh-controller/internal/licences 15.950s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.015s
ok github.com/novox/mesh-controller/internal/link 87.829s
ok github.com/novox/mesh-controller/internal/lint 2.321s
ok github.com/novox/mesh-controller/internal/outward 1.043s
ok github.com/novox/mesh-controller/internal/overlay 1.033s
ok github.com/novox/mesh-controller/internal/secrets 1.466s
ok github.com/novox/mesh-controller/internal/store 6.000s
ok github.com/novox/mesh-controller/internal/testbus 1.356s
ok github.com/novox/mesh-controller/internal/token 1.023s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.34s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.37s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.36s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.35s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.33s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.31s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.36s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.22s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.21s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.23s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.21s)
```
Independent re-review for hq issue 353, after review 22. One gap is left in one merge order; the rest holds.
What holds
Review 22's three asks are built. An answer without in_force falls back to main's list check (asker_wire.go:405-426). Each case names its own remedy: push <node> for a list that does not allow it (asker_wire.go:400-403, 408-410, 418-423, 434-435), wait about 5 s for a list written but not reloaded (asker_wire.go:427-433), and bus upgrade when the list check cannot be worked out (asker_wire.go:414-417). The two tests the review asked for exist (asker_test.go:877, 922).
The caveat is logged once and re-armed when it clears (asker.go:322-328).
settings propose uses the same judgement (proposals.go:594). The decisions coming back from Telegram (Approve/Decline) are not gated: only publishing an ask is.
Errors fail closed and are said. An unreachable module, a module error, an unreadable answer and records that are not open all come back as not held with the error in the reason (asker_wire.go:277-300, 386-398). The asker raises them as asks-undelivered. A list that cannot be composed counts as behind (push.go:1109-1131).
Live today: novox/nats.nats_user_can for user controller, publish, mesh.seat.operator-channel.accept.ask.mesh-controller answers allowed: true with no in_force (the module before #166). So on merge the fallback path runs. While the list sent to novox is current, the controller keeps asking, as main does.
The tests fail without the fix. I copied c3371ae's asker_test.go onto 473d597 and added a one-function adapter for busGrantBy, which reproduced 473d597's judgement (grantInForce alone). Both new tests then fail: "older module, list current" is not asked, "written, not reloaded" names push anchor, and "older module, list unknown" names push. Without the adapter the file does not compile at 473d597.
sh merge-check.sh passed on c3371ae with a throwaway PostgreSQL 18 (MESH_TEST_POSTGRES): gofmt, go vet, go test -race -count=1 ./... with every package ok, and all 11 listed store tests PASS.
Checks at c3371ae, read at 12:49 when they had finished: mesh/merge-gate pass ("builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step; every machine composes"), mesh/repo-check pass ("its merge-check.sh passed"), mesh/delivery ready.
What I ask to change: the window between #166's merge and the bus upgrade
Once mesh-catalog #166 is merged, every send to novox is refused until the operator runs the bus upgrade (push.go:1170-1181, errBusWaits, ADR 0236). The old nats module answers on novox throughout that window, so this PR takes the fallback. In that window:
The asks are silenced while the bus holds the grant.brokerBehind (push.go:1086) is true after any change to the composed user list: an assign, an enrolment, a person's access. Pushes to novox are held, so it stays true until the bus upgrade. The bus module meanwhile answers allowed: true from a file the server reloaded long ago, so the bus does hold the controller's grant. The fallback still withholds every ask, including the Approve/Decline asks to Telegram, and settings propose refuses. This brings back for an older module the "any change to the user list holds asking back" behaviour that point 2 of this PR removed.
The remedy it names is not true. The reason says "push novox carries it" (asker_wire.go:418-423). A push of novox is refused in that window with "a new bus build waits for its planned step". Only bus upgrade sends the list.
This window exists in both merge orders (#188 before #166, and #166 merged with its upgrade still pending). Main (#186) behaves the same way today, so this is not a regression. It is still the case the review was about: the asks go silent until a person's bus upgrade, and the condition names the wrong action.
The smallest fix I see is one of these:
(a) Recommended. For an answer without in_force and allowed: true, treat the grant as held with the caveat ("reload not confirmed; the bus upgrade carries a module that confirms it"), and do not consult brokerBehind. The file allows the publish, and the entrypoint reloads within 5 s of any content change (mesh-catalog modules/nats/entrypoint.sh, POLL=5). The worst case is one ask refused in those 5 s, which issue 369's re-ask already takes back. allowed: false keeps push <node>.
(b) Keep the list check, but when the bus machine's sends are held for a bus step (busHeld for that node), name bus upgrade and not push. This keeps the silence and fixes only the text.
For either option, add a test case "older module, list behind, bus step pending". It asks under (a), or names bus upgrade and not push under (b).
Smaller points (not blocking)
With #166 live, in_force: false with allowed: true also covers the cases #166 calls "not known": the write time cannot be stat'ed, /varz cannot be read, or the server gives no load time. The controller then says "written but not in force yet (); ... nothing is to be done but wait" (asker_wire.go:427-433). The why comes through, but "wait" is false for those cases. I suggest saying "wait" only when both file_written and server_loaded are present.
With #166 live, a list whose content changes and changes back within one 5 s poll gets a newer mtime but no reload (the digest equals the last one). in_force then stays false until the next real change. This is rare; it belongs to #166 (compare against the digest, or reload on mtime), not to this PR.
listBehind now counts a list that cannot be composed as behind. The asker then says "push <node> carries it", but that push fails with the plan's reason. The plan's reason would be the truer remedy there.
The branch is two merges behind main (#246, #249). It merges cleanly (git merge-tree).
Merge order I recommend: mesh-catalog #166 first, then the operator's bus upgrade (which delivers it to novox), confirmed by nats_user_can on novox answering with in_force, then this PR with the fix above. If this PR merges first, it is no worse than main, but the window described above remains until (a) or (b) lands.
Verdict: request changes at c3371ae
Independent re-review for hq issue 353, after review 22. One gap is left in one merge order; the rest holds.
**What holds**
- Review 22's three asks are built. An answer without `in_force` falls back to main's list check (asker_wire.go:405-426). Each case names its own remedy: `push <node>` for a list that does not allow it (asker_wire.go:400-403, 408-410, 418-423, 434-435), wait about 5 s for a list written but not reloaded (asker_wire.go:427-433), and `bus upgrade` when the list check cannot be worked out (asker_wire.go:414-417). The two tests the review asked for exist (asker_test.go:877, 922).
- The caveat is logged once and re-armed when it clears (asker.go:322-328).
- `settings propose` uses the same judgement (proposals.go:594). The decisions coming back from Telegram (Approve/Decline) are not gated: only publishing an ask is.
- Errors fail closed and are said. An unreachable module, a module error, an unreadable answer and records that are not open all come back as not held with the error in the reason (asker_wire.go:277-300, 386-398). The asker raises them as asks-undelivered. A list that cannot be composed counts as behind (push.go:1109-1131).
- Live today: `novox/nats.nats_user_can` for user `controller`, publish, `mesh.seat.operator-channel.accept.ask.mesh-controller` answers `allowed: true` with no `in_force` (the module before #166). So on merge the fallback path runs. While the list sent to novox is current, the controller keeps asking, as main does.
- The tests fail without the fix. I copied c3371ae's asker_test.go onto 473d597 and added a one-function adapter for `busGrantBy`, which reproduced 473d597's judgement (`grantInForce` alone). Both new tests then fail: "older module, list current" is not asked, "written, not reloaded" names `push anchor`, and "older module, list unknown" names push. Without the adapter the file does not compile at 473d597.
- `sh merge-check.sh` passed on c3371ae with a throwaway PostgreSQL 18 (MESH_TEST_POSTGRES): gofmt, `go vet`, `go test -race -count=1 ./...` with every package ok, and all 11 listed store tests PASS.
- Checks at c3371ae, read at 12:49 when they had finished: mesh/merge-gate pass ("builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step; every machine composes"), mesh/repo-check pass ("its merge-check.sh passed"), mesh/delivery ready.
**What I ask to change: the window between #166's merge and the bus upgrade**
Once mesh-catalog #166 is merged, every send to novox is refused until the operator runs the bus upgrade (push.go:1170-1181, `errBusWaits`, ADR 0236). The old nats module answers on novox throughout that window, so this PR takes the fallback. In that window:
1. **The asks are silenced while the bus holds the grant.** `brokerBehind` (push.go:1086) is true after any change to the composed user list: an assign, an enrolment, a person's access. Pushes to novox are held, so it stays true until the bus upgrade. The bus module meanwhile answers `allowed: true` from a file the server reloaded long ago, so the bus does hold the controller's grant. The fallback still withholds every ask, including the Approve/Decline asks to Telegram, and `settings propose` refuses. This brings back for an older module the "any change to the user list holds asking back" behaviour that point 2 of this PR removed.
2. **The remedy it names is not true.** The reason says "`push novox` carries it" (asker_wire.go:418-423). A push of novox is refused in that window with "a new bus build waits for its planned step". Only `bus upgrade` sends the list.
This window exists in both merge orders (#188 before #166, and #166 merged with its upgrade still pending). Main (#186) behaves the same way today, so this is not a regression. It is still the case the review was about: the asks go silent until a person's bus upgrade, and the condition names the wrong action.
The smallest fix I see is one of these:
- (a) **Recommended.** For an answer without `in_force` and `allowed: true`, treat the grant as held with the caveat ("reload not confirmed; the bus upgrade carries a module that confirms it"), and do not consult `brokerBehind`. The file allows the publish, and the entrypoint reloads within 5 s of any content change (mesh-catalog modules/nats/entrypoint.sh, POLL=5). The worst case is one ask refused in those 5 s, which issue 369's re-ask already takes back. `allowed: false` keeps `push <node>`.
- (b) Keep the list check, but when the bus machine's sends are held for a bus step (`busHeld` for that node), name `bus upgrade` and not `push`. This keeps the silence and fixes only the text.
For either option, add a test case "older module, list behind, bus step pending". It asks under (a), or names `bus upgrade` and not push under (b).
**Smaller points (not blocking)**
- With #166 live, `in_force: false` with `allowed: true` also covers the cases #166 calls "not known": the write time cannot be stat'ed, `/varz` cannot be read, or the server gives no load time. The controller then says "written but not in force yet (<why>); ... nothing is to be done but wait" (asker_wire.go:427-433). The why comes through, but "wait" is false for those cases. I suggest saying "wait" only when both `file_written` and `server_loaded` are present.
- With #166 live, a list whose content changes and changes back within one 5 s poll gets a newer mtime but no reload (the digest equals the last one). `in_force` then stays false until the next real change. This is rare; it belongs to #166 (compare against the digest, or reload on mtime), not to this PR.
- `listBehind` now counts a list that cannot be composed as behind. The asker then says "`push <node>` carries it", but that push fails with the plan's reason. The plan's reason would be the truer remedy there.
- The branch is two merges behind main (#246, #249). It merges cleanly (`git merge-tree`).
**Merge order I recommend:** mesh-catalog #166 first, then the operator's `bus upgrade` (which delivers it to novox), confirmed by `nats_user_can` on novox answering with `in_force`, then this PR with the fix above. If this PR merges first, it is no worse than main, but the window described above remains until (a) or (b) lands.
mesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller)
mesh/repo-check: PASS — its merge-check.sh passed
Delivery plan — builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step
tier 0: mesh-controller
tier 1: build-agent
ace: receives build-agent
g14: receives build-agent
novox: receives mesh-controller, build-agent
shanks: receives build-agent
build-agent keeps data (ADR 0233): what it holds is backed up before it moves
grants step: if this changes what the bus's user list says (a new tool, subject or user), novox is sent that list alone, every build there kept, before build-agent is judged on its first machine
Run by the build seat on ace as build-1791802571702797812 (builds --log build-1791802571702797812).
--- the gate: build-agent, mesh-controller
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T10:50:44Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T10:50:44Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-controller
notes:
- cmd/mesh-controller/asker_test.go and cmd/mesh-controller/delivery_order_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.042s
ok github.com/novox/mesh-controller/cmd/mesh-controller 231.888s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.019s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.033s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.023s
ok github.com/novox/mesh-controller/examples/route-proxy 1.150s
ok github.com/novox/mesh-controller/internal/artifacts 1.075s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.234s
ok github.com/novox/mesh-controller/internal/builder 16.397s
ok github.com/novox/mesh-controller/internal/catalogue 7.927s
ok github.com/novox/mesh-controller/internal/conditions 2.708s
ok github.com/novox/mesh-controller/internal/envfile 1.079s
ok github.com/novox/mesh-controller/internal/facts 1.084s
ok github.com/novox/mesh-controller/internal/identity 4.767s
ok github.com/novox/mesh-controller/internal/inventory 90.491s
ok github.com/novox/mesh-controller/internal/lease 17.747s
ok github.com/novox/mesh-controller/internal/licences 4.435s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.072s
ok github.com/novox/mesh-controller/internal/link 59.685s
ok github.com/novox/mesh-controller/internal/lint 2.427s
ok github.com/novox/mesh-controller/internal/outward 1.116s
ok github.com/novox/mesh-controller/internal/overlay 1.088s
ok github.com/novox/mesh-controller/internal/secrets 1.498s
ok github.com/novox/mesh-controller/internal/store 3.565s
ok github.com/novox/mesh-controller/internal/testbus 1.258s
ok github.com/novox/mesh-controller/internal/token 1.021s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.42s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.41s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.39s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.39s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.40s)
./cmd/mesh-controller --- PASS: TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds (0.29s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.23s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.22s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.23s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.23s)
./internal/inventory --- PASS: TestAConditionsLooksAreKeptBesideTheBucket (0.22s)
**Merge check** at `218b378d`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step
- tier 0: mesh-controller
- tier 1: build-agent
- ace: receives build-agent
- g14: receives build-agent
- novox: receives mesh-controller, build-agent
- shanks: receives build-agent
- build-agent keeps data (ADR 0233): what it holds is backed up before it moves
- grants step: if this changes what the bus's user list says (a new tool, subject or user), novox is sent that list alone, every build there kept, before build-agent is judged on its first machine
Run by the build seat on ace as `build-1791802571702797812` (`builds --log build-1791802571702797812`).
```
--- the gate: build-agent, mesh-controller
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T10:50:44Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T10:50:44Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-controller
notes:
- cmd/mesh-controller/asker_test.go and cmd/mesh-controller/delivery_order_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.042s
ok github.com/novox/mesh-controller/cmd/mesh-controller 231.888s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.019s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.033s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.023s
ok github.com/novox/mesh-controller/examples/route-proxy 1.150s
ok github.com/novox/mesh-controller/internal/artifacts 1.075s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.234s
ok github.com/novox/mesh-controller/internal/builder 16.397s
ok github.com/novox/mesh-controller/internal/catalogue 7.927s
ok github.com/novox/mesh-controller/internal/conditions 2.708s
ok github.com/novox/mesh-controller/internal/envfile 1.079s
ok github.com/novox/mesh-controller/internal/facts 1.084s
ok github.com/novox/mesh-controller/internal/identity 4.767s
ok github.com/novox/mesh-controller/internal/inventory 90.491s
ok github.com/novox/mesh-controller/internal/lease 17.747s
ok github.com/novox/mesh-controller/internal/licences 4.435s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.072s
ok github.com/novox/mesh-controller/internal/link 59.685s
ok github.com/novox/mesh-controller/internal/lint 2.427s
ok github.com/novox/mesh-controller/internal/outward 1.116s
ok github.com/novox/mesh-controller/internal/overlay 1.088s
ok github.com/novox/mesh-controller/internal/secrets 1.498s
ok github.com/novox/mesh-controller/internal/store 3.565s
ok github.com/novox/mesh-controller/internal/testbus 1.258s
ok github.com/novox/mesh-controller/internal/token 1.021s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.42s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.41s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.39s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.39s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.40s)
./cmd/mesh-controller --- PASS: TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds (0.29s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.23s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.22s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.23s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.23s)
./internal/inventory --- PASS: TestAConditionsLooksAreKeptBesideTheBucket (0.22s)
```
Independent third review for hq issue 353, after reviews 22 and 31. Fresh clone; the head is three commits on main's d881f47 (main has since gained #248, fafa3fd).
Review 31's ask (a) is built, and it closes the window.
An answer without in_force and with allowed: true is held, with the caveat, and the list check is not consulted (asker_wire.go:468-471). In the window between mesh-catalog #166's merge and the operator's bus upgrade, the old nats module keeps answering allowed: true, so the asks keep going out even though brokerBehind reads behind for good and novox cannot be pushed. The test case "older module, list behind, bus step pending" covers this (asker_test.go:946-947).
The list check now runs only for allowed: false (asker_wire.go:454-466, 472-475) and for a module that cannot answer or gives an unreadable answer (asker_wire.go:438-453). That check is main's floor: held while the list sent is current.
The caveat is said once over two reconciles; the test counts it (asker_test.go:979-986, asker.go:322-328).
"Wait about 5 s" is said only when the answer carries both file_written and server_loaded (asker_wire.go:478-481). Otherwise the reason says what is not known, with the module's why (asker_wire.go:483-488). The test checks that the word "wait" is absent in both of #166's not-known cases.
Every order is safe from silenced asks and from asks published into a refusal.
#188 before #166: the old module answers allowed: true (review 31 read this live), so the controller asks, with the caveat.
#166 merged, upgrade pending: same as above. Pushes to novox are held, which no longer matters to the asker.
After the upgrade: in_force: true means held. During a reload gap, in_force: false with both times means not held, for at most one 5 s poll.
Refusal window: only the older module's allowed: true can be early. The file allows it, and the entrypoint reloads within one 5 s poll of a content change (mesh-catalog modules/nats/entrypoint.sh, POLL=5). Issue 369's re-ask takes back an ask refused in that time.
I checked one more way in_force could stay false for good: the node-engine rewriting an unchanged file, which bumps the mtime with no reload. It does not happen. mesh-host's applyFile writes only when the content differs (internal/apply/apply.go, contentSame), so the mtime moves only with a real change.
Every remedy text is true.
busListState.remedy (asker_wire.go:395-406) names plan <node> when the list cannot be composed. Review 31's smaller point is fixed: a push would fail there on the plan's reason.
It names the bus step when sends wait on it. That text comes from busHeld (bus_step.go:130), the same function push.go:1183 uses to refuse with errBusWaits. The remedy and the refusal cannot drift apart.
Otherwise it names push <node>.
bus upgrade and plan exist as commands (bus_step.go busCommand).
Push behaviour is unchanged.brokerBehind (push.go:1086-1089) now only projects brokerListState (push.go:1093-1133). The body is the same line for line except the return values. The Uncomposed field is new and read only by the asker. busHeld is called from the asker read-only, and only when the list is behind.
The tests fail without the fix. I copied 218b378's asker_test.go onto 010d7ef (the rebased c3371ae; git range-diff shows them identical) with a one-function adapter that maps the new busGrantBy signature onto the old one. Both new tests fail:
"older module, allowed, list behind" is not held;
"older module, list behind, bus step pending" is not asked;
the two not-known cases say "wait";
"not allowed, bus step pending" and "list uncomposed" name push anchor;
"module cannot answer" is not held.
Suite.sh merge-check.sh with a throwaway PostgreSQL (MESH_TEST_POSTGRES) passed at 218b378:
gofmt and go vet passed;
go test -race -count=1 ./...: 26 packages ok;
all 13 listed store tests PASS.
Checks at 218b378, read at 13:07 once they had finished:
mesh/merge-gate: pass ("builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without");
In the not-allowed case with a current list, the reason names "push <node> carries it again". BusStep is looked up only when the list is behind (asker_wire.go:370), so a pending bus step is not named here. The case is narrow: the composition grants, the sent digest matches, and the file still disallows, which means the send was not applied. In that case the push named would be refused with errBusWaits. Looking up busHeld whenever the answer is allowed: false would make the text true there too.
A module that cannot answer, with the list current, is now held (asker_wire.go:451-452). At 010d7ef it failed closed. This is main's judgement, as intended. Its refusal window is longer than 5 s, though: the sent digest is recorded at send time (push.go:1045), before the node-engine writes the file. It applies only while the nats module's tool is unreachable and the bus itself is up.
The older-module caveat says "the bus upgrade (bus upgrade) carries one that confirms it". Before #166 merges, a bus upgrade has nothing to carry. Wording such as "the nats module that answers in_force (mesh-catalog #166), delivered by a bus upgrade" would be true in every order.
grantLookEvery is 30 s (asker_wire.go:512). A not-held judgement such as "wait about 5 s" is therefore kept up to 30 s, and asks arriving then go to asks-undelivered until the next look. This is unchanged from review 31's head. It delays the asks and loses none, but it is longer than the 5 s the text promises.
saidGrantCaveat is reset only by a clean yes (asker.go:326-327), not by a no. A caveat, then a no, then the caveat again is said only once.
As review 31 noted, #166 has its own flip-back case: the content changes and changes back within one poll, the mtime is newer and no reload follows. That belongs to #166.
Recommended order. Any order is now safe for the asks. I recommend:
Merge this PR, on its own controller rollout, once its plan is done.
Confirm nats_user_can on novox answers with in_force: true, and that the asker's caveat line stops.
Verdict: approve at 218b378
Independent third review for hq issue 353, after reviews 22 and 31. Fresh clone; the head is three commits on main's d881f47 (main has since gained #248, fafa3fd).
**Review 31's ask (a) is built, and it closes the window.**
- An answer without `in_force` and with `allowed: true` is held, with the caveat, and the list check is not consulted (asker_wire.go:468-471). In the window between mesh-catalog #166's merge and the operator's bus upgrade, the old nats module keeps answering `allowed: true`, so the asks keep going out even though `brokerBehind` reads behind for good and novox cannot be pushed. The test case "older module, list behind, bus step pending" covers this (asker_test.go:946-947).
- The list check now runs only for `allowed: false` (asker_wire.go:454-466, 472-475) and for a module that cannot answer or gives an unreadable answer (asker_wire.go:438-453). That check is main's floor: held while the list sent is current.
- The caveat is said once over two reconciles; the test counts it (asker_test.go:979-986, asker.go:322-328).
- "Wait about 5 s" is said only when the answer carries both `file_written` and `server_loaded` (asker_wire.go:478-481). Otherwise the reason says what is not known, with the module's why (asker_wire.go:483-488). The test checks that the word "wait" is absent in both of #166's not-known cases.
**Every order is safe from silenced asks and from asks published into a refusal.**
- #188 before #166: the old module answers `allowed: true` (review 31 read this live), so the controller asks, with the caveat.
- #166 merged, upgrade pending: same as above. Pushes to novox are held, which no longer matters to the asker.
- After the upgrade: `in_force: true` means held. During a reload gap, `in_force: false` with both times means not held, for at most one 5 s poll.
- Refusal window: only the older module's `allowed: true` can be early. The file allows it, and the entrypoint reloads within one 5 s poll of a content change (mesh-catalog modules/nats/entrypoint.sh, POLL=5). Issue 369's re-ask takes back an ask refused in that time.
- I checked one more way `in_force` could stay false for good: the node-engine rewriting an unchanged file, which bumps the mtime with no reload. It does not happen. mesh-host's applyFile writes only when the content differs (internal/apply/apply.go, `contentSame`), so the mtime moves only with a real change.
**Every remedy text is true.**
- `busListState.remedy` (asker_wire.go:395-406) names `plan <node>` when the list cannot be composed. Review 31's smaller point is fixed: a push would fail there on the plan's reason.
- It names the bus step when sends wait on it. That text comes from `busHeld` (bus_step.go:130), the same function push.go:1183 uses to refuse with `errBusWaits`. The remedy and the refusal cannot drift apart.
- Otherwise it names `push <node>`.
- `bus upgrade` and `plan` exist as commands (bus_step.go `busCommand`).
**Push behaviour is unchanged.** `brokerBehind` (push.go:1086-1089) now only projects `brokerListState` (push.go:1093-1133). The body is the same line for line except the return values. The `Uncomposed` field is new and read only by the asker. `busHeld` is called from the asker read-only, and only when the list is behind.
**The tests fail without the fix.** I copied 218b378's asker_test.go onto 010d7ef (the rebased c3371ae; `git range-diff` shows them identical) with a one-function adapter that maps the new `busGrantBy` signature onto the old one. Both new tests fail:
- "older module, allowed, list behind" is not held;
- "older module, list behind, bus step pending" is not asked;
- the two not-known cases say "wait";
- "not allowed, bus step pending" and "list uncomposed" name `push anchor`;
- "module cannot answer" is not held.
**Suite.** `sh merge-check.sh` with a throwaway PostgreSQL (MESH_TEST_POSTGRES) passed at 218b378:
- gofmt and `go vet` passed;
- `go test -race -count=1 ./...`: 26 packages ok;
- all 13 listed store tests PASS.
**Checks at 218b378**, read at 13:07 once they had finished:
- mesh/merge-gate: pass ("builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without");
- mesh/repo-check: pass ("its merge-check.sh passed");
- mesh/delivery: ready.
**Smaller points (not blocking)**
1. In the not-allowed case with a current list, the reason names "`push <node>` carries it again". `BusStep` is looked up only when the list is behind (asker_wire.go:370), so a pending bus step is not named here. The case is narrow: the composition grants, the sent digest matches, and the file still disallows, which means the send was not applied. In that case the push named would be refused with `errBusWaits`. Looking up `busHeld` whenever the answer is `allowed: false` would make the text true there too.
2. A module that cannot answer, with the list current, is now held (asker_wire.go:451-452). At 010d7ef it failed closed. This is main's judgement, as intended. Its refusal window is longer than 5 s, though: the sent digest is recorded at send time (push.go:1045), before the node-engine writes the file. It applies only while the nats module's tool is unreachable and the bus itself is up.
3. The older-module caveat says "the bus upgrade (`bus upgrade`) carries one that confirms it". Before #166 merges, a bus upgrade has nothing to carry. Wording such as "the nats module that answers `in_force` (mesh-catalog #166), delivered by a bus upgrade" would be true in every order.
4. `grantLookEvery` is 30 s (asker_wire.go:512). A not-held judgement such as "wait about 5 s" is therefore kept up to 30 s, and asks arriving then go to asks-undelivered until the next look. This is unchanged from review 31's head. It delays the asks and loses none, but it is longer than the 5 s the text promises.
5. `saidGrantCaveat` is reset only by a clean yes (asker.go:326-327), not by a no. A caveat, then a no, then the caveat again is said only once.
6. As review 31 noted, #166 has its own flip-back case: the content changes and changes back within one poll, the mtime is newer and no reload follows. That belongs to #166.
**Recommended order.** Any order is now safe for the asks. I recommend:
1. Merge this PR, on its own controller rollout, once its plan is done.
2. Then mesh-catalog #166.
3. Then the operator's `bus upgrade`.
4. Confirm `nats_user_can` on novox answers with `in_force: true`, and that the asker's caveat line stops.
change held asking back, a list sent was taken for a list the bus enforces, a plan that could not be
worked out read as not behind, and no test held the real judgement. Now the asker asks only when the
composition grants the controller itself to publish its ask, and the bus's own module says that grant
is in force on the running server (nats_user_can's in_force, mesh-catalog); each fact not known is not
held, said with why. brokerBehind takes a list nobody could compose as behind.
A bus module older than its in_force answer read as not held, so a merge
before mesh-catalog #166 is live would silence every ask until a person's
bus upgrade. Such an answer now falls back to the check main makes (is the
list sent to the bus's machine the one composed now), says the reload is
not confirmed, and each case names its own remedy: push <node> for a list
behind or not allowing it, a ~5 s wait for a reload, the bus upgrade when
an older module's list cannot be judged.
After mesh-catalog #166 merges and before the bus upgrade, the bus's
machine cannot be pushed, so any change to the composed list made the
list check read behind for good: every ask was withheld and the condition
named a push that is refused. An older module's allowed now counts as
held, the unconfirmed reload said once; the worst case is one ask refused
in the seconds before a reload, which is asked again (issue 369).
The list check stays for a module that says not allowed or cannot answer,
and its remedy is the plan's own reason when the list cannot be composed,
the bus upgrade while sends there wait on it, and a push otherwise. An
in_force false is a wait only when the module gives both times; otherwise
what is not known is named.
mesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-controller)
mesh/repo-check: PASS — its merge-check.sh passed
Delivery plan — builds mesh-controller → novox; no bus step
tier 0: mesh-controller
novox: receives mesh-controller
read by no module's build: cmd/mesh-controller/asker_test.go, cmd/mesh-controller/delivery_order_test.go
Run by the build seat on shanks as build-1791804306213787464 (builds --log build-1791804306213787464).
--- the gate: mesh-controller
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T11:22:09Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T11:22:09Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-controller
notes:
- cmd/mesh-controller/asker_test.go and cmd/mesh-controller/delivery_order_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.043s
ok github.com/novox/mesh-controller/cmd/mesh-controller 259.694s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.058s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.063s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.045s
ok github.com/novox/mesh-controller/examples/route-proxy 1.210s
ok github.com/novox/mesh-controller/internal/artifacts 1.100s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.454s
ok github.com/novox/mesh-controller/internal/builder 19.379s
ok github.com/novox/mesh-controller/internal/catalogue 8.036s
ok github.com/novox/mesh-controller/internal/conditions 2.700s
ok github.com/novox/mesh-controller/internal/envfile 1.056s
ok github.com/novox/mesh-controller/internal/facts 1.061s
ok github.com/novox/mesh-controller/internal/identity 5.029s
ok github.com/novox/mesh-controller/internal/inventory 103.157s
ok github.com/novox/mesh-controller/internal/lease 17.732s
ok github.com/novox/mesh-controller/internal/licences 4.580s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.047s
ok github.com/novox/mesh-controller/internal/link 62.043s
ok github.com/novox/mesh-controller/internal/lint 2.503s
ok github.com/novox/mesh-controller/internal/outward 1.082s
ok github.com/novox/mesh-controller/internal/overlay 1.071s
ok github.com/novox/mesh-controller/internal/secrets 1.537s
ok github.com/novox/mesh-controller/internal/store 3.543s
ok github.com/novox/mesh-controller/internal/testbus 1.260s
ok github.com/novox/mesh-controller/internal/token 1.017s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.45s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.44s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.44s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.41s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.47s)
./cmd/mesh-controller --- PASS: TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds (0.33s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.48s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.46s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.42s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.43s)
./cmd/mesh-controller --- PASS: TestAWalkEndedByCancelOrClearIsSaidEnded (0.49s)
./cmd/mesh-controller --- PASS: TestAWalkRestartedByARebuildIsSaidStarted (0.44s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain (0.48s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.27s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.27s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.29s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.29s)
./internal/inventory --- PASS: TestAConditionsLooksAreKeptBesideTheBucket (0.25s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.28s)
**Merge check** at `f996cf7b`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-controller)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds mesh-controller → novox; no bus step
- tier 0: mesh-controller
- novox: receives mesh-controller
- read by no module's build: cmd/mesh-controller/asker_test.go, cmd/mesh-controller/delivery_order_test.go
Run by the build seat on shanks as `build-1791804306213787464` (`builds --log build-1791804306213787464`).
```
--- the gate: mesh-controller
mesh-controller: ok, 57 tool(s); WARNING: merge-window ask(s) for a setting and do(es) not say whether it is trusted, so it counts as trusted: set at the terminal alone; say "trusted" false where nothing trusts it (novox/hq issue 339); WARNING: controller stay(s) up and say(s) not how it is ready — judged by liveness alone, refused from 2026-11-18 (ADR 0240 rule 8)
long-running resources without health: 1
files asking for a setting without saying whether it is trusted: 1
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 1 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T11:22:09Z says they run
composing every machine as the mesh is
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
composing every machine with the change
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
the bus's user list leaves out 5 user(s) the mesh has minted no credential for: controller, node.f35, node.idt, node.ygcyr, node.yvbpsm. Each is a user that cannot connect until one is issued
merge gate: PASS — every machine composes with the change as it did without (4 of 4 compose)
judged by controller development build against the facts of 2026-10-12T11:22:09Z
modules the change moves:
- mesh-controller (changed)
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-controller
notes:
- cmd/mesh-controller/asker_test.go and cmd/mesh-controller/delivery_order_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.043s
ok github.com/novox/mesh-controller/cmd/mesh-controller 259.694s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.058s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.063s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.045s
ok github.com/novox/mesh-controller/examples/route-proxy 1.210s
ok github.com/novox/mesh-controller/internal/artifacts 1.100s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.454s
ok github.com/novox/mesh-controller/internal/builder 19.379s
ok github.com/novox/mesh-controller/internal/catalogue 8.036s
ok github.com/novox/mesh-controller/internal/conditions 2.700s
ok github.com/novox/mesh-controller/internal/envfile 1.056s
ok github.com/novox/mesh-controller/internal/facts 1.061s
ok github.com/novox/mesh-controller/internal/identity 5.029s
ok github.com/novox/mesh-controller/internal/inventory 103.157s
ok github.com/novox/mesh-controller/internal/lease 17.732s
ok github.com/novox/mesh-controller/internal/licences 4.580s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.047s
ok github.com/novox/mesh-controller/internal/link 62.043s
ok github.com/novox/mesh-controller/internal/lint 2.503s
ok github.com/novox/mesh-controller/internal/outward 1.082s
ok github.com/novox/mesh-controller/internal/overlay 1.071s
ok github.com/novox/mesh-controller/internal/secrets 1.537s
ok github.com/novox/mesh-controller/internal/store 3.543s
ok github.com/novox/mesh-controller/internal/testbus 1.260s
ok github.com/novox/mesh-controller/internal/token 1.017s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.45s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.44s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.43s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.44s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.41s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.47s)
./cmd/mesh-controller --- PASS: TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds (0.33s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.48s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.46s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.42s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.43s)
./cmd/mesh-controller --- PASS: TestAWalkEndedByCancelOrClearIsSaidEnded (0.49s)
./cmd/mesh-controller --- PASS: TestAWalkRestartedByARebuildIsSaidStarted (0.44s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain (0.48s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.27s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.27s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.29s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.29s)
./internal/inventory --- PASS: TestAConditionsLooksAreKeptBesideTheBucket (0.25s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.28s)
```
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Follow-up to the review of #186 (hq issue 353, hq #253). Do not merge without review.
The four review points
servingAsker.TestTheServingAskerJudgesTheGrantForRealbuilds the serving asker and calls itsgrantHeld: with nothing to read it must say not held. These mutations were each run and each fails a test: removinggrantHeld: grantHeldIn(open, conn); forcingheld = trueingrantHeldBy; makingcontrollerMayAskalways yes; reading an answer withoutin_forceas a yes; andlistBehindignoring a compose error.controllerMayAskcomposes the users from the records (broker.Users) and asksbroker.MayPublishof the controller's own permissions foraccept.ask.mesh-controller. Changes to other users no longer hold asking back. Test:TestTheControllersOwnGrantToAskIsReadFromTheComposition.nats_user_canfor usercontroller, publish, the ask subject. It must answerin_force: true, which means the user list there allows the publish and the server'sconfig_load_timeis after the list's last write. This needs mesh-catalog #166. A bus module older than that answer is not a yes. Test:TestTheBusModulesAnswerIsInForceOnlyWhenItSaysSo.judgeGranttreats every error as not held and says why (the conditions that need the operator are raised as undelivered, as in #186).brokerBehindtreats a list that cannot be composed as behind (listBehind). Before, a plan that could not be worked out read as not behind.Behaviour change to note (point 4, push): when the bus machine's plan cannot be worked out, a push now puts that machine first, and it fails there with the plan's own reason. A rollout through
sendRollout(all or nothing) now refuses instead of sending code whose grants the bus may not hold. Before, both went ahead.Checks. The branch is on main's head (
7477346), so the head is the merge result.sh merge-check.shpassed locally with mesh-host at main b0d03be and a throwaway PostgreSQL: gofmt, vet,go test -race, 26 packages ok.Order. mesh-catalog #166 first, live on the machine holding the bus. Then this. If this merges first, the controller asks the operator nothing and says why ("the bus module does not say whether a grant is in force"), raised as undelivered.
Delivery
novox/mesh-controller@f996cf7b6b9b— delivered since 2026-10-12T11:43:08ZDelivery plan — builds mesh-controller → novox; no bus step
Machines
Transitions
The commit's note under
refs/notes/mesh-plankeeps every transition:git log --notes=mesh-plan.Merge check at
00bcdccdmesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller, route-proxy)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step
Run by the build seat on novox as
build-1791568027996482179(builds --log build-1791568027996482179).00bcdccdc7to473d597fe2Merge check at
473d597fmesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-controller)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds mesh-controller → novox; no bus step
Run by the build seat on shanks as
build-1791800029376562746(builds --log build-1791800029376562746).Verdict: request changes, at head
473d597(rebased onto main 22a2f73; the old head was00bcdcc).Independent review for hq issue 353.
Rebase: asker_test.go conflicted with main's new test for issue 369 (TestARefusalIsAskedAgainAndTheUndeliveredConditionClearsOnceTaken). Both sides only added code, so I kept both. Main had also added a second caller, the settings proposer (proposals.go:594, ADR 0277), that called
grantHeldIn(open). I changed it tograntHeldIn(open, conn), so the proposer now judges the grant the same way.What holds:
if a.grantHeld != nildisabled in asker.go it fails three ways (asked while the bus lacks the grant, said 0 times, no undelivered condition).go test -race ./...passes on all 26 packages locally (the store tests skip without PostgreSQL; mesh/repo-check ran them).473d597: mesh/merge-gate pass (4 of 4 compose, no bus step), mesh/repo-check pass, mesh/delivery ready.Why I ask for changes: the order with mesh-catalog #166.
This head is mergeable now, and nothing stops it landing before #166 is live on novox. If it does, the bus module there answers nats_user_can without in_force. grantInForce reads that as not held, and it stays not held until #166 is live. In that order:
settings proposerefuses every proposal ("the operator cannot be asked yet: …") and points to the terminal.rehearsestill publishes, because it is not gated.This is not a short window between two merges. mesh-delivery.what-if says #166 needs a person's bus upgrade (
mesh-controller.bus upgrade), and that holds every delivery to novox until it runs. So the asks stay silent until the operator runs that upgrade.push novoxcarries it, and the bus reloads it in place". For an older bus module that is false. A push changes nothing; only #166 delivered by the bus upgrade does. This breaks the issue's acceptance ("says which push carries the grant") in exactly this order.A guard is needed. Merge order alone is not enough here, because the head is green and mergeable on its own. I suggest:
push. For a list not allowed on the machine, it ispush <node>. For a reload not yet done, it is to wait about 5 s.If you would rather keep the strict reading (an older module is never a yes), the alternative is a hard merge rule: #188 merges only after
nats_user_canon novox answers with in_force. Then the wrongpushtext in point 4 still has to be fixed. I recommend the fallback.Merge check at
c3371ae6mesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller, route-proxy)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step
Run by the build seat on g14 as
build-1791801278187283201(builds --log build-1791801278187283201).Verdict: request changes at
c3371aeIndependent re-review for hq issue 353, after review 22. One gap is left in one merge order; the rest holds.
What holds
in_forcefalls back to main's list check (asker_wire.go:405-426). Each case names its own remedy:push <node>for a list that does not allow it (asker_wire.go:400-403, 408-410, 418-423, 434-435), wait about 5 s for a list written but not reloaded (asker_wire.go:427-433), andbus upgradewhen the list check cannot be worked out (asker_wire.go:414-417). The two tests the review asked for exist (asker_test.go:877, 922).settings proposeuses the same judgement (proposals.go:594). The decisions coming back from Telegram (Approve/Decline) are not gated: only publishing an ask is.novox/nats.nats_user_canfor usercontroller, publish,mesh.seat.operator-channel.accept.ask.mesh-controlleranswersallowed: truewith noin_force(the module before #166). So on merge the fallback path runs. While the list sent to novox is current, the controller keeps asking, as main does.473d597and added a one-function adapter forbusGrantBy, which reproduced 473d597's judgement (grantInForcealone). Both new tests then fail: "older module, list current" is not asked, "written, not reloaded" namespush anchor, and "older module, list unknown" names push. Without the adapter the file does not compile at473d597.sh merge-check.shpassed onc3371aewith a throwaway PostgreSQL 18 (MESH_TEST_POSTGRES): gofmt,go vet,go test -race -count=1 ./...with every package ok, and all 11 listed store tests PASS.c3371ae, read at 12:49 when they had finished: mesh/merge-gate pass ("builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step; every machine composes"), mesh/repo-check pass ("its merge-check.sh passed"), mesh/delivery ready.What I ask to change: the window between #166's merge and the bus upgrade
Once mesh-catalog #166 is merged, every send to novox is refused until the operator runs the bus upgrade (push.go:1170-1181,
errBusWaits, ADR 0236). The old nats module answers on novox throughout that window, so this PR takes the fallback. In that window:brokerBehind(push.go:1086) is true after any change to the composed user list: an assign, an enrolment, a person's access. Pushes to novox are held, so it stays true until the bus upgrade. The bus module meanwhile answersallowed: truefrom a file the server reloaded long ago, so the bus does hold the controller's grant. The fallback still withholds every ask, including the Approve/Decline asks to Telegram, andsettings proposerefuses. This brings back for an older module the "any change to the user list holds asking back" behaviour that point 2 of this PR removed.push novoxcarries it" (asker_wire.go:418-423). A push of novox is refused in that window with "a new bus build waits for its planned step". Onlybus upgradesends the list.This window exists in both merge orders (#188 before #166, and #166 merged with its upgrade still pending). Main (#186) behaves the same way today, so this is not a regression. It is still the case the review was about: the asks go silent until a person's bus upgrade, and the condition names the wrong action.
The smallest fix I see is one of these:
in_forceandallowed: true, treat the grant as held with the caveat ("reload not confirmed; the bus upgrade carries a module that confirms it"), and do not consultbrokerBehind. The file allows the publish, and the entrypoint reloads within 5 s of any content change (mesh-catalog modules/nats/entrypoint.sh, POLL=5). The worst case is one ask refused in those 5 s, which issue 369's re-ask already takes back.allowed: falsekeepspush <node>.busHeldfor that node), namebus upgradeand notpush. This keeps the silence and fixes only the text.For either option, add a test case "older module, list behind, bus step pending". It asks under (a), or names
bus upgradeand not push under (b).Smaller points (not blocking)
in_force: falsewithallowed: truealso covers the cases #166 calls "not known": the write time cannot be stat'ed,/varzcannot be read, or the server gives no load time. The controller then says "written but not in force yet (); ... nothing is to be done but wait" (asker_wire.go:427-433). The why comes through, but "wait" is false for those cases. I suggest saying "wait" only when bothfile_writtenandserver_loadedare present.in_forcethen stays false until the next real change. This is rare; it belongs to #166 (compare against the digest, or reload on mtime), not to this PR.listBehindnow counts a list that cannot be composed as behind. The asker then says "push <node>carries it", but that push fails with the plan's reason. The plan's reason would be the truer remedy there.git merge-tree).Merge order I recommend: mesh-catalog #166 first, then the operator's
bus upgrade(which delivers it to novox), confirmed bynats_user_canon novox answering within_force, then this PR with the fix above. If this PR merges first, it is no worse than main, but the window described above remains until (a) or (b) lands.c3371ae6f5to218b378db6Merge check at
218b378dmesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: build-agent, mesh-controller)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step
Run by the build seat on ace as
build-1791802571702797812(builds --log build-1791802571702797812).Verdict: approve at
218b378Independent third review for hq issue 353, after reviews 22 and 31. Fresh clone; the head is three commits on main's
d881f47(main has since gained #248,fafa3fd).Review 31's ask (a) is built, and it closes the window.
in_forceand withallowed: trueis held, with the caveat, and the list check is not consulted (asker_wire.go:468-471). In the window between mesh-catalog #166's merge and the operator's bus upgrade, the old nats module keeps answeringallowed: true, so the asks keep going out even thoughbrokerBehindreads behind for good and novox cannot be pushed. The test case "older module, list behind, bus step pending" covers this (asker_test.go:946-947).allowed: false(asker_wire.go:454-466, 472-475) and for a module that cannot answer or gives an unreadable answer (asker_wire.go:438-453). That check is main's floor: held while the list sent is current.file_writtenandserver_loaded(asker_wire.go:478-481). Otherwise the reason says what is not known, with the module's why (asker_wire.go:483-488). The test checks that the word "wait" is absent in both of #166's not-known cases.Every order is safe from silenced asks and from asks published into a refusal.
allowed: true(review 31 read this live), so the controller asks, with the caveat.in_force: truemeans held. During a reload gap,in_force: falsewith both times means not held, for at most one 5 s poll.allowed: truecan be early. The file allows it, and the entrypoint reloads within one 5 s poll of a content change (mesh-catalog modules/nats/entrypoint.sh, POLL=5). Issue 369's re-ask takes back an ask refused in that time.in_forcecould stay false for good: the node-engine rewriting an unchanged file, which bumps the mtime with no reload. It does not happen. mesh-host's applyFile writes only when the content differs (internal/apply/apply.go,contentSame), so the mtime moves only with a real change.Every remedy text is true.
busListState.remedy(asker_wire.go:395-406) namesplan <node>when the list cannot be composed. Review 31's smaller point is fixed: a push would fail there on the plan's reason.busHeld(bus_step.go:130), the same function push.go:1183 uses to refuse witherrBusWaits. The remedy and the refusal cannot drift apart.push <node>.bus upgradeandplanexist as commands (bus_step.gobusCommand).Push behaviour is unchanged.
brokerBehind(push.go:1086-1089) now only projectsbrokerListState(push.go:1093-1133). The body is the same line for line except the return values. TheUncomposedfield is new and read only by the asker.busHeldis called from the asker read-only, and only when the list is behind.The tests fail without the fix. I copied 218b378's asker_test.go onto
010d7ef(the rebased c3371ae;git range-diffshows them identical) with a one-function adapter that maps the newbusGrantBysignature onto the old one. Both new tests fail:push anchor;Suite.
sh merge-check.shwith a throwaway PostgreSQL (MESH_TEST_POSTGRES) passed at218b378:go vetpassed;go test -race -count=1 ./...: 26 packages ok;Checks at
218b378, read at 13:07 once they had finished:Smaller points (not blocking)
push <node>carries it again".BusStepis looked up only when the list is behind (asker_wire.go:370), so a pending bus step is not named here. The case is narrow: the composition grants, the sent digest matches, and the file still disallows, which means the send was not applied. In that case the push named would be refused witherrBusWaits. Looking upbusHeldwhenever the answer isallowed: falsewould make the text true there too.010d7efit failed closed. This is main's judgement, as intended. Its refusal window is longer than 5 s, though: the sent digest is recorded at send time (push.go:1045), before the node-engine writes the file. It applies only while the nats module's tool is unreachable and the bus itself is up.bus upgrade) carries one that confirms it". Before #166 merges, a bus upgrade has nothing to carry. Wording such as "the nats module that answersin_force(mesh-catalog #166), delivered by a bus upgrade" would be true in every order.grantLookEveryis 30 s (asker_wire.go:512). A not-held judgement such as "wait about 5 s" is therefore kept up to 30 s, and asks arriving then go to asks-undelivered until the next look. This is unchanged from review 31's head. It delays the asks and loses none, but it is longer than the 5 s the text promises.saidGrantCaveatis reset only by a clean yes (asker.go:326-327), not by a no. A caveat, then a no, then the caveat again is said only once.Recommended order. Any order is now safe for the asks. I recommend:
bus upgrade.nats_user_canon novox answers within_force: true, and that the asker's caveat line stops.218b378db6tof996cf7b6bMerge check at
f996cf7bmesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-controller)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds mesh-controller → novox; no bus step
Run by the build seat on shanks as
build-1791804306213787464(builds --log build-1791804306213787464).