Builds the controller part of novox/hq ADR 0308 decision 1 and the bound of decision 5, for issue 532. The router part (messenger, mesh-catalog) builds against these fields and merges after this. Rebased onto main f549647.
What changes
plan-moved is extended, not doubled. Its body is still the walk whole (inventory.Plan). On the save that starts a walk and the save that ends it, the body also carries: walk, phase (started|ended), touches ([{machine, modules, carried}], never null), outcome (delivered|failed|stopped, on the end only), bound_seconds, budget_class (core|leaf|none), started_at, and ended_at on the end. Every other save carries none of these fields.
Start and end are marked once, in the store (internal/inventory/walksaid.go). SavePlansMoving marks times.started / times.ended in the same write, and takes the mark back if the write is refused. A controller replaced mid-walk (a core walk replaces it) does not say the start again and does not miss the end. A retried walk starts again. Batches, walks waiting for their word, walks that end without starting, and walks kept before times existed say nothing. Release plans are said.
Every save a verb makes is said, in one place (review 21). The controller's run installs inventory.PlanSaved for every command except serve (sayingWhatAVerbKept, delivery.go). It reaches the bus on the first save only, publishes before the save returns, and reads touches from the inventory that kept the walk: PlanSaved now receives that inventory. So cancel, clear, rebuild and any later verb say a start or end without a wrap of their own. When the bus cannot be reached, a kept start or end is said loudly on the verb's error output. sayingOnTheBus (delivery go|stop|check, plans go|stop|close|retry, upgrade release-backlog) still wraps its verbs, and now restores the hook it found instead of clearing it.
Outcome: done gives delivered. Failed with delivery.stopped, or superseded, gives stopped. Any other failure gives failed. plans stop|close keeps who stopped it and why in delivery.stopped / stopped_why, so a person's stop ends stopped, not failed.
A stop is cleared when a walk walks again (goesOnAgain): plans retry (failed builds, rollouts, a whole tier) and rebuild joining the walk clear delivery.stopped/stopped_why. A retried walk that fails again ends failed, and searchAfterFailures runs for it again. This also closes the same gap for a stop by the delivery's owner.
Touches (cmd/mesh-controller/walk_said.go):
At the start: what the walk will send. That is each module of its tiers whose policy rolls it out, on every machine running it (for a release plan, only the machines in its order).
At the end: what it reached. That is each module's first machines and rest machines, plus every module a gated send carried.
On a machine where node-tools moves, the modules whose tools it serves are added (own tools, or a bundle with loads). On a machine where mesh-host moves, every module there is added. What is on a machine is read from sent_builds plus its assignments.
Both kinds are listed in modules, and the added ones are also listed in carried.
Bound: 2 × deliveryBudgets[class] (the map the D16 over-budget probe already uses: core 10 min, leaf 3 min). That gives 1200 s for core and 360 s for leaf. The bound is 900 s, not doubled, when the walk has no budget: a release plan, a walk whose every module's policy records (held for a person, ADR 0282 §1), or a walk with no class.
What the operator sees change: a hand stop now ends a delivery as Stopped. mesh-delivery copies delivery.stopped into WalkSeen.StoppedBy (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go:531), and walkStopped() (model.go:558) and the Delivering rows (table.go:350-380) read it. After plans stop|close:
a delivery whose walk carried its own commit used to become Failed (mesh/delivery: "failure — failed: …"); it now becomes Stopped ("error — stopped: its walk … was stopped by a person (…): …");
a delivery whose walk carried a later merge used to become Published again, waiting for a walk of its own commit that never came; it now ends Stopped.
The Stop row's guard text still says "stopped through this owner"; a small mesh-catalog follow-up for it is opened separately. plans output does not change.
Limits, named
The start is said at the walk's cut or let-go, on the first save that finds it live, which is before its first tier's build and not when the first tier is sent. The router holds earlier than the contract asks (the safe side), and the bound also spends that first build's time.
The leaf bound is 6 min because the leaf budget is 3 min since mesh-controller#241 (issue 525 step 1). ADR 0308 §5 says 10 min until issue 525 makes the budget three; that step is merged, so the 6 min follows it.
A crash between the commit and the publish loses that one event: the mark is kept in the same write, and no later save says it again. A lost end fails safe (the router releases at the bound). A lost start fails open (that walk is not held). Not fixed here: an outbox would be the fix, and it is larger than this change.
started and ended may arrive out of order on a very short walk; the router orders by started_at/ended_at or revision, not by arrival.
Consumers. mesh-delivery decodes plan-moved with plain json.Unmarshal into its Walk, which has no field the new names collide with. The kept stop changes what it shows, as above. The view principal (#203) only subscribes. No schema change: the marks live in the times JSON column.
Tests (ADR 0308 How it is checked §1):
walk_said_test.go: the start names modules and machines, including what a moved tool runner carries (and not postgres, which serves no tools). A moved node-engine carries everything on its machine. A module whose policy records is not touched. The end names what was reached and the outcome. Bounds: core 1200, leaf 360, none 900. The body shape is checked, and no walk fields appear on ordinary saves.
Live on the store and a bus, each running the real verb: delivery go then delivery stop; plans stop and plans close (ended, stopped); plans retry (started); upgrade release-backlog (started). Through the controller's run, as typed: cancel <build> and clear (ended, failed, TestAWalkEndedByCancelOrClearIsSaidEnded); rebuild a (started, TestAWalkRestartedByARebuildIsSaidStarted); a walk stopped by hand, then retried by plans retry or rebuild, walks without its stop and ends failed when it fails again (TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain). Each new test fails without its fix: no event within 5 s without the hook in run; the stop kept without goesOnAgain.
walksaid_test.go: start once, end once, retry starts again, the cases that say nothing, and outcome. Store test: the mark is kept, and a refused save leaves the walk unmarked.
All eight store tests are listed in testdata/store-tests. sh merge-check.sh passed locally with a throwaway PostgreSQL 18, race on, every listed store test PASS.
Reviewers' checklist
Every process but serve has PlanSaved set by run; sayingOnTheBus restores what it found; nothing else sets it outside tests.
Start is said before anything is sent: on the first save that finds the walk live, not after the first send.
A refused save (ErrPlanMoved) leaves no mark, so the next save says it (undo in SavePlansMoving).
Every path that sets a failed walk going again calls goesOnAgain (resumed, retryRollouts, retryTierWhole).
Nothing in WalkPhases changes for release plans that now get a times holding only the marks.
A fact that cannot be read is logged and left out. The event is never held back.
This is a controller rollout, so merge only while mesh-controller.plans says no plan is mid-gate. Merge this before the messenger PR. By adr0308@g14. Not merged; waits for review.
Builds the controller part of novox/hq **ADR 0308** decision 1 and the bound of decision 5, for **issue 532**. The router part (messenger, mesh-catalog) builds against these fields and merges after this. Rebased onto main f549647.
**What changes**
- `plan-moved` is extended, not doubled. Its body is still the walk whole (`inventory.Plan`). On the save that **starts** a walk and the save that **ends** it, the body also carries:
`walk`, `phase` (`started`|`ended`), `touches` ([{machine, modules, carried}], never null), `outcome` (`delivered`|`failed`|`stopped`, on the end only), `bound_seconds`, `budget_class` (`core`|`leaf`|`none`), `started_at`, and `ended_at` on the end. Every other save carries none of these fields.
- **Start and end are marked once, in the store** (`internal/inventory/walksaid.go`). `SavePlansMoving` marks `times.started` / `times.ended` in the same write, and takes the mark back if the write is refused. A controller replaced mid-walk (a core walk replaces it) does not say the start again and does not miss the end. A retried walk starts again. Batches, walks waiting for their word, walks that end without starting, and walks kept before `times` existed say nothing. Release plans are said.
- **Every save a verb makes is said, in one place** (review 21). The controller's `run` installs `inventory.PlanSaved` for every command except `serve` (`sayingWhatAVerbKept`, delivery.go). It reaches the bus on the first save only, publishes before the save returns, and reads touches from the inventory that kept the walk: `PlanSaved` now receives that inventory. So `cancel`, `clear`, `rebuild` and any later verb say a start or end without a wrap of their own. When the bus cannot be reached, a kept start or end is said loudly on the verb's error output. `sayingOnTheBus` (`delivery go|stop|check`, `plans go|stop|close|retry`, `upgrade release-backlog`) still wraps its verbs, and now restores the hook it found instead of clearing it.
- **Outcome**: done gives delivered. Failed with `delivery.stopped`, or superseded, gives stopped. Any other failure gives failed. `plans stop|close` keeps who stopped it and why in `delivery.stopped` / `stopped_why`, so a person's stop ends `stopped`, not `failed`.
- **A stop is cleared when a walk walks again** (`goesOnAgain`): `plans retry` (failed builds, rollouts, a whole tier) and `rebuild` joining the walk clear `delivery.stopped`/`stopped_why`. A retried walk that fails again ends `failed`, and `searchAfterFailures` runs for it again. This also closes the same gap for a stop by the delivery's owner.
- **Touches** (`cmd/mesh-controller/walk_said.go`):
- At the start: what the walk will send. That is each module of its tiers whose policy rolls it out, on every machine running it (for a release plan, only the machines in its order).
- At the end: what it reached. That is each module's first machines and rest machines, plus every module a gated send carried.
- On a machine where node-tools moves, the modules whose tools it serves are added (own `tools`, or a bundle with `loads`). On a machine where mesh-host moves, every module there is added. What is on a machine is read from `sent_builds` plus its assignments.
- Both kinds are listed in `modules`, and the added ones are also listed in `carried`.
- **Bound**: 2 × `deliveryBudgets[class]` (the map the D16 over-budget probe already uses: core 10 min, leaf 3 min). That gives 1200 s for core and 360 s for leaf. The bound is 900 s, not doubled, when the walk has no budget: a release plan, a walk whose every module's policy records (held for a person, ADR 0282 §1), or a walk with no class.
**What the operator sees change: a hand stop now ends a delivery as Stopped.** mesh-delivery copies `delivery.stopped` into `WalkSeen.StoppedBy` (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go:531), and `walkStopped()` (model.go:558) and the Delivering rows (table.go:350-380) read it. After `plans stop|close`:
- a delivery whose walk carried its own commit used to become Failed (`mesh/delivery`: "failure — failed: …"); it now becomes Stopped ("error — stopped: its walk … was stopped by a person (…): …");
- a delivery whose walk carried a later merge used to become Published again, waiting for a walk of its own commit that never came; it now ends Stopped.
The Stop row's guard text still says "stopped through this owner"; a small mesh-catalog follow-up for it is opened separately. `plans` output does not change.
**Limits, named**
- **The start is said at the walk's cut or let-go**, on the first save that finds it live, which is before its first tier's build and not when the first tier is sent. The router holds earlier than the contract asks (the safe side), and the bound also spends that first build's time.
- **The leaf bound is 6 min** because the leaf budget is 3 min since mesh-controller#241 (issue 525 step 1). ADR 0308 §5 says 10 min until issue 525 makes the budget three; that step is merged, so the 6 min follows it.
- **A crash between the commit and the publish loses that one event**: the mark is kept in the same write, and no later save says it again. A lost end fails safe (the router releases at the bound). A lost start fails open (that walk is not held). Not fixed here: an outbox would be the fix, and it is larger than this change.
- `started` and `ended` may arrive out of order on a very short walk; the router orders by `started_at`/`ended_at` or `revision`, not by arrival.
**Consumers.** mesh-delivery decodes `plan-moved` with plain `json.Unmarshal` into its `Walk`, which has no field the new names collide with. The kept stop changes what it shows, as above. The view principal (#203) only subscribes. No schema change: the marks live in the `times` JSON column.
**Tests** (ADR 0308 How it is checked §1):
- `walk_said_test.go`: the start names modules and machines, including what a moved tool runner carries (and not postgres, which serves no tools). A moved node-engine carries everything on its machine. A module whose policy records is not touched. The end names what was reached and the outcome. Bounds: core 1200, leaf 360, none 900. The body shape is checked, and no walk fields appear on ordinary saves.
- Live on the store and a bus, each running the real verb: `delivery go` then `delivery stop`; `plans stop` and `plans close` (ended, stopped); `plans retry` (started); `upgrade release-backlog` (started). Through the controller's `run`, as typed: `cancel <build>` and `clear` (ended, failed, `TestAWalkEndedByCancelOrClearIsSaidEnded`); `rebuild a` (started, `TestAWalkRestartedByARebuildIsSaidStarted`); a walk stopped by hand, then retried by `plans retry` or `rebuild`, walks without its stop and ends `failed` when it fails again (`TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain`). Each new test fails without its fix: no event within 5 s without the hook in `run`; the stop kept without `goesOnAgain`.
- `walksaid_test.go`: start once, end once, retry starts again, the cases that say nothing, and outcome. Store test: the mark is kept, and a refused save leaves the walk unmarked.
- All eight store tests are listed in `testdata/store-tests`. `sh merge-check.sh` passed locally with a throwaway PostgreSQL 18, race on, every listed store test PASS.
**Reviewers' checklist**
- [ ] Every process but `serve` has `PlanSaved` set by `run`; `sayingOnTheBus` restores what it found; nothing else sets it outside tests.
- [ ] Start is said before anything is sent: on the first save that finds the walk live, not after the first send.
- [ ] A refused save (ErrPlanMoved) leaves no mark, so the next save says it (`undo` in `SavePlansMoving`).
- [ ] Every path that sets a failed walk going again calls `goesOnAgain` (`resumed`, `retryRollouts`, `retryTierWhole`).
- [ ] Nothing in `WalkPhases` changes for release plans that now get a `times` holding only the marks.
- [ ] A fact that cannot be read is logged and left out. The event is never held back.
This is a controller rollout, so merge only while `mesh-controller.plans` says no plan is mid-gate. Merge this before the messenger PR. By adr0308@g14. Not merged; waits for review.
Deliverynovox/mesh-controller@1e9dcf066750 — delivered since 2026-10-12T11:19:07Z
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
Machines
mesh-controller on novox: passed — healthy 3 times over 3m44s
mesh-controller on the rest running it: sent
build-agent on ace: passed — healthy 1 times over 1m33s
build-agent on the rest running it: sent
Transitions
2026-10-12 10:57 (new) → proposed (announced): novox/mesh-controller#247's head announced
2026-10-12 11:04 proposed → checked (checked): the verdict names this commit
2026-10-12 11:04 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
2026-10-12 11:10 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 11:10 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 11:19 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@1e9dcf066750` — **delivered** since 2026-10-12T11:19:07Z
**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
**Machines**
- mesh-controller on novox: passed — healthy 3 times over 3m44s
- mesh-controller on the rest running it: sent
- build-agent on ace: passed — healthy 1 times over 1m33s
- build-agent on the rest running it: sent
**Transitions**
- 2026-10-12 10:57 (new) → proposed (announced): novox/mesh-controller#247's head announced
- 2026-10-12 11:04 proposed → checked (checked): the verdict names this commit
- 2026-10-12 11:04 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
- 2026-10-12 11:10 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 11:10 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 11:19 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/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
read by no module's build: cmd/mesh-controller/walk_said_test.go, internal/inventory/walksaid_test.go, testdata/store-tests
Run by the build seat on ace as build-1791798923847656032 (builds --log build-1791798923847656032).
--- 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 2026-10-12T09:47:46Z 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 2026-10-12T09:47:46Z
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 2 module(s) in 2 tier(s): build-agent, mesh-controller
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.041s
ok github.com/novox/mesh-controller/cmd/mesh-controller 228.807s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.023s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.030s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.022s
ok github.com/novox/mesh-controller/examples/route-proxy 1.128s
ok github.com/novox/mesh-controller/internal/artifacts 1.065s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.049s
ok github.com/novox/mesh-controller/internal/builder 15.828s
ok github.com/novox/mesh-controller/internal/catalogue 7.630s
ok github.com/novox/mesh-controller/internal/conditions 2.632s
ok github.com/novox/mesh-controller/internal/envfile 1.026s
ok github.com/novox/mesh-controller/internal/facts 1.031s
ok github.com/novox/mesh-controller/internal/identity 4.600s
ok github.com/novox/mesh-controller/internal/inventory 89.985s
ok github.com/novox/mesh-controller/internal/lease 17.728s
ok github.com/novox/mesh-controller/internal/licences 4.205s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.024s
ok github.com/novox/mesh-controller/internal/link 60.121s
ok github.com/novox/mesh-controller/internal/lint 2.385s
ok github.com/novox/mesh-controller/internal/outward 1.079s
ok github.com/novox/mesh-controller/internal/overlay 1.058s
ok github.com/novox/mesh-controller/internal/secrets 1.480s
ok github.com/novox/mesh-controller/internal/store 3.563s
ok github.com/novox/mesh-controller/internal/testbus 1.257s
ok github.com/novox/mesh-controller/internal/token 1.018s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.39s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.40s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.40s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.37s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.38s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.41s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.44s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.25s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.24s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.26s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.25s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.24s)
**Merge check** at `bd3b23a8`
- `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
- read by no module's build: cmd/mesh-controller/walk_said_test.go, internal/inventory/walksaid_test.go, testdata/store-tests
Run by the build seat on ace as `build-1791798923847656032` (`builds --log build-1791798923847656032`).
```
--- 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 2026-10-12T09:47:46Z 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 2026-10-12T09:47:46Z
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 2 module(s) in 2 tier(s): build-agent, mesh-controller
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.041s
ok github.com/novox/mesh-controller/cmd/mesh-controller 228.807s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.023s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.030s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.022s
ok github.com/novox/mesh-controller/examples/route-proxy 1.128s
ok github.com/novox/mesh-controller/internal/artifacts 1.065s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.049s
ok github.com/novox/mesh-controller/internal/builder 15.828s
ok github.com/novox/mesh-controller/internal/catalogue 7.630s
ok github.com/novox/mesh-controller/internal/conditions 2.632s
ok github.com/novox/mesh-controller/internal/envfile 1.026s
ok github.com/novox/mesh-controller/internal/facts 1.031s
ok github.com/novox/mesh-controller/internal/identity 4.600s
ok github.com/novox/mesh-controller/internal/inventory 89.985s
ok github.com/novox/mesh-controller/internal/lease 17.728s
ok github.com/novox/mesh-controller/internal/licences 4.205s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.024s
ok github.com/novox/mesh-controller/internal/link 60.121s
ok github.com/novox/mesh-controller/internal/lint 2.385s
ok github.com/novox/mesh-controller/internal/outward 1.079s
ok github.com/novox/mesh-controller/internal/overlay 1.058s
ok github.com/novox/mesh-controller/internal/secrets 1.480s
ok github.com/novox/mesh-controller/internal/store 3.563s
ok github.com/novox/mesh-controller/internal/testbus 1.257s
ok github.com/novox/mesh-controller/internal/token 1.018s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.39s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.40s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.40s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.37s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.38s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.41s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.44s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.25s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.24s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.26s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.25s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.24s)
```
VERDICT: REQUEST_CHANGES (sent as a comment review: the forge refuses a tool's approve or request-changes, ADR 0266). Independent review at bd3b23a (ADR 0308 §1, §5; issue 532). Two blocking findings.
BLOCKING: a start or end mark is persisted but never published when a walk starts or ends through a verb that has no bus. markWalk runs inside every SavePlansMoving call (internal/inventory/plans.go:483-496), but PlanSaved is set only in serve (push.go:252) and inside sayingOnTheBus (delivery.go:1225). Three verbs save a walk's start or end outside sayingOnTheBus:
plans stop|close (cmd/mesh-controller/release_plan.go:1593-1599): open walk -> failed. Times.Ended is written, no ended event is published, and because the mark is kept, no later save says it. The router holds what the walk touched until its bound (up to 20 min) after a person stopped it.
plans retry (retryPlan -> retryRollouts, plan_retry.go:462-464): failed -> rolling. Times.Started is written anew, no started event. The retried walk is never held, and its later ended (said by serve) arrives for a walk the router never saw start.
upgrade release-backlog (release.go:821 -> releaseBacklog, release.go:664-668): creates a live release plan. Started is marked, no event.
Scenario: the operator runs mesh-controller.plans stop <walk> mid-walk; the stop is kept, the walk's ended is never on the bus, and messages about its modules stay held until the bound passes. Fix: wrap these three in sayingOnTheBus(ctx, inv, ...) as plans go already is (release_plan.go:1553), or do not mark when no one will say the save. A test that drives plans stop and plans retry through the command and expects the event would pin it.
BLOCKING: a walk stopped by a person through plans stop is reported with outcome failed, not stopped. Plan.Outcome (internal/inventory/walksaid.go:33-45) returns stopped only for PlanSuperseded or a failed plan with Delivery.Stopped set. Only stopWalkWithout (delivery.go:178-184) sets Delivery.Stopped; plans stop|close sets State=failed and a Note only (release_plan.go:1593-1594). ADR 0308 §1 names stopped for this case, and the router's end message would tell the operator their own stop as a failure. Fix: have plans stop|close record who stopped it (for example Delivery.Stopped/StoppedWhy) or let Outcome recognise the hand stop, with a test.
Non-blocking remarks:
Start moment: a walk is marked started on the first save that finds it live (walksaid.go:63-67), that is at its cut or its let-go, before its first tier's build, not when the first tier is sent as the contract says. Holding earlier is the safe side, and the comment says so; but the bound then also spends the first build's time (leaf bound 6 min). Worth one line in the PR body.
Leaf bound: deliveryBudgets has the leaf at 3 min (over_budget.go:32), so a leaf walk is bound at 6 min. ADR 0308 §5 says 10 min until issue 525's build makes the budget three; step 1 of 525 (mesh-controller#241) did make it three, so I read this as consistent, but the PR body should say so.
The event is published from a goroutine after the commit (delivery.go:1191); a crash between commit and publish loses that one event for good, since the mark is kept. The end then fails safe by the bound; the start fails open. Acceptable, worth naming.
started and ended go out from separate goroutines and may reach the router out of order on a very short walk; the router must use started_at/ended_at or revision, not arrival order.
Checked and fine: marks kept in times JSON in the same transaction and undone on a refused write (plans.go:483-496, test TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid); a restart neither repeats nor drops a mark said by serve; ended once (markWalk needs Started set and Ended nil); retry re-opens with a new start; carried = modules on the machine a moved node-engine (all) or node-tools (tool-serving ones) brings along, and modules includes them (walk_said.go:200-222); terminal states done/failed/superseded all map; no json name of walkSaid collides with Plan's, the walk fields are absent when phase is absent (nil embedded pointer), and mesh-delivery's walk model (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go) has no colliding field and decodes leniently. go vet clean; go test -race -count=1 ./... passes in full.
Required checks at bd3b23a when reviewed: mesh/merge-gate pending (building on ace), mesh/repo-check missing.
VERDICT: REQUEST_CHANGES (sent as a comment review: the forge refuses a tool's approve or request-changes, ADR 0266). Independent review at bd3b23a (ADR 0308 §1, §5; issue 532). Two blocking findings.
1. BLOCKING: a start or end mark is persisted but never published when a walk starts or ends through a verb that has no bus. markWalk runs inside every SavePlansMoving call (internal/inventory/plans.go:483-496), but PlanSaved is set only in serve (push.go:252) and inside sayingOnTheBus (delivery.go:1225). Three verbs save a walk's start or end outside sayingOnTheBus:
- `plans stop|close` (cmd/mesh-controller/release_plan.go:1593-1599): open walk -> failed. Times.Ended is written, no `ended` event is published, and because the mark is kept, no later save says it. The router holds what the walk touched until its bound (up to 20 min) after a person stopped it.
- `plans retry` (retryPlan -> retryRollouts, plan_retry.go:462-464): failed -> rolling. Times.Started is written anew, no `started` event. The retried walk is never held, and its later `ended` (said by serve) arrives for a walk the router never saw start.
- `upgrade release-backlog` (release.go:821 -> releaseBacklog, release.go:664-668): creates a live release plan. Started is marked, no event.
Scenario: the operator runs `mesh-controller.plans stop <walk>` mid-walk; the stop is kept, the walk's `ended` is never on the bus, and messages about its modules stay held until the bound passes. Fix: wrap these three in sayingOnTheBus(ctx, inv, ...) as `plans go` already is (release_plan.go:1553), or do not mark when no one will say the save. A test that drives `plans stop` and `plans retry` through the command and expects the event would pin it.
2. BLOCKING: a walk stopped by a person through `plans stop` is reported with outcome `failed`, not `stopped`. Plan.Outcome (internal/inventory/walksaid.go:33-45) returns stopped only for PlanSuperseded or a failed plan with Delivery.Stopped set. Only stopWalkWithout (delivery.go:178-184) sets Delivery.Stopped; `plans stop|close` sets State=failed and a Note only (release_plan.go:1593-1594). ADR 0308 §1 names `stopped` for this case, and the router's end message would tell the operator their own stop as a failure. Fix: have `plans stop|close` record who stopped it (for example Delivery.Stopped/StoppedWhy) or let Outcome recognise the hand stop, with a test.
Non-blocking remarks:
- Start moment: a walk is marked started on the first save that finds it live (walksaid.go:63-67), that is at its cut or its let-go, before its first tier's build, not when the first tier is sent as the contract says. Holding earlier is the safe side, and the comment says so; but the bound then also spends the first build's time (leaf bound 6 min). Worth one line in the PR body.
- Leaf bound: deliveryBudgets has the leaf at 3 min (over_budget.go:32), so a leaf walk is bound at 6 min. ADR 0308 §5 says 10 min until issue 525's build makes the budget three; step 1 of 525 (mesh-controller#241) did make it three, so I read this as consistent, but the PR body should say so.
- The event is published from a goroutine after the commit (delivery.go:1191); a crash between commit and publish loses that one event for good, since the mark is kept. The end then fails safe by the bound; the start fails open. Acceptable, worth naming.
- `started` and `ended` go out from separate goroutines and may reach the router out of order on a very short walk; the router must use started_at/ended_at or revision, not arrival order.
Checked and fine: marks kept in times JSON in the same transaction and undone on a refused write (plans.go:483-496, test TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid); a restart neither repeats nor drops a mark said by serve; ended once (markWalk needs Started set and Ended nil); retry re-opens with a new start; carried = modules on the machine a moved node-engine (all) or node-tools (tool-serving ones) brings along, and `modules` includes them (walk_said.go:200-222); terminal states done/failed/superseded all map; no json name of walkSaid collides with Plan's, the walk fields are absent when phase is absent (nil embedded pointer), and mesh-delivery's walk model (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go) has no colliding field and decodes leniently. `go vet` clean; `go test -race -count=1 ./...` passes in full.
Required checks at bd3b23a when reviewed: mesh/merge-gate pending (building on ace), mesh/repo-check missing.
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
read by no module's build: cmd/mesh-controller/walk_said_test.go, internal/inventory/walksaid_test.go, testdata/store-tests
Run by the build seat on g14 as build-1791799722338334168 (builds --log build-1791799722338334168).
--- 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 2026-10-12T10:08:43Z 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 2 module(s) in 2 tier(s): build-agent, mesh-controller
notes:
- cmd/mesh-controller/walk_said_test.go, internal/inventory/walksaid_test.go and testdata/store-tests 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 1.034s
ok github.com/novox/mesh-controller/cmd/mesh-controller 249.367s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.012s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.023s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.015s
ok github.com/novox/mesh-controller/examples/route-proxy 1.142s
ok github.com/novox/mesh-controller/internal/artifacts 1.047s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.195s
ok github.com/novox/mesh-controller/internal/builder 18.840s
ok github.com/novox/mesh-controller/internal/catalogue 9.552s
ok github.com/novox/mesh-controller/internal/conditions 2.893s
ok github.com/novox/mesh-controller/internal/envfile 1.026s
ok github.com/novox/mesh-controller/internal/facts 1.030s
ok github.com/novox/mesh-controller/internal/identity 5.650s
ok github.com/novox/mesh-controller/internal/inventory 101.710s
ok github.com/novox/mesh-controller/internal/lease 17.649s
ok github.com/novox/mesh-controller/internal/licences 5.053s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.010s
ok github.com/novox/mesh-controller/internal/link 64.763s
ok github.com/novox/mesh-controller/internal/lint 2.193s
ok github.com/novox/mesh-controller/internal/outward 1.036s
ok github.com/novox/mesh-controller/internal/overlay 1.028s
ok github.com/novox/mesh-controller/internal/secrets 1.431s
ok github.com/novox/mesh-controller/internal/store 3.348s
ok github.com/novox/mesh-controller/internal/testbus 1.187s
ok github.com/novox/mesh-controller/internal/token 1.024s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.36s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.39s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.39s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.37s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.36s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.38s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.42s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.41s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.36s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.37s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.21s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.24s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.24s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.23s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.22s)
**Merge check** at `44983d6d`
- `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
- read by no module's build: cmd/mesh-controller/walk_said_test.go, internal/inventory/walksaid_test.go, testdata/store-tests
Run by the build seat on g14 as `build-1791799722338334168` (`builds --log build-1791799722338334168`).
```
--- 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 2026-10-12T10:08:43Z 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 2 module(s) in 2 tier(s): build-agent, mesh-controller
notes:
- cmd/mesh-controller/walk_said_test.go, internal/inventory/walksaid_test.go and testdata/store-tests 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 1.034s
ok github.com/novox/mesh-controller/cmd/mesh-controller 249.367s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.012s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.023s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.015s
ok github.com/novox/mesh-controller/examples/route-proxy 1.142s
ok github.com/novox/mesh-controller/internal/artifacts 1.047s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.195s
ok github.com/novox/mesh-controller/internal/builder 18.840s
ok github.com/novox/mesh-controller/internal/catalogue 9.552s
ok github.com/novox/mesh-controller/internal/conditions 2.893s
ok github.com/novox/mesh-controller/internal/envfile 1.026s
ok github.com/novox/mesh-controller/internal/facts 1.030s
ok github.com/novox/mesh-controller/internal/identity 5.650s
ok github.com/novox/mesh-controller/internal/inventory 101.710s
ok github.com/novox/mesh-controller/internal/lease 17.649s
ok github.com/novox/mesh-controller/internal/licences 5.053s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.010s
ok github.com/novox/mesh-controller/internal/link 64.763s
ok github.com/novox/mesh-controller/internal/lint 2.193s
ok github.com/novox/mesh-controller/internal/outward 1.036s
ok github.com/novox/mesh-controller/internal/overlay 1.028s
ok github.com/novox/mesh-controller/internal/secrets 1.431s
ok github.com/novox/mesh-controller/internal/store 3.348s
ok github.com/novox/mesh-controller/internal/testbus 1.187s
ok github.com/novox/mesh-controller/internal/token 1.024s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.36s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.39s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.39s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.37s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.36s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.38s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.42s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.41s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.36s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.37s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.21s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.24s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.24s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.23s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.22s)
```
Independent re-review of 44983d6 against ADR 0308 decision 1 and review 15. Both review-15 findings are fixed and proven. Two more verbs keep a walk's start or end mark without saying it, so the claim "every verb that starts or ends a walk says it" does not hold yet. A hand stop also stays on a walk after it is retried.
Review 15, finding 1 (fixed).plans retry (cmd/mesh-controller/release_plan.go:1568), plans stop|close (release_plan.go:1588) and upgrade release-backlog (cmd/mesh-controller/release.go:818) now run inside sayingOnTheBus. Each one's HoldPlans release is deferred inside the closure, so the plans are let go before the closure returns. I ran the three new tests against bd3b23a with the 44983d6 test file. All three fail there: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped says "no walk was said ended", and TestAWalkRetriedByAPersonIsSaidStartedAgain and TestABacklogReleasedByAPersonIsSaidStarted say "no walk was said started". All three pass at 44983d6.
Review 15, finding 2 (fixed).stopByHand (release_plan.go:1903-1911) sets Delivery.Stopped and StoppedWhy, and Outcome (internal/inventory/walksaid.go:39) then gives stopped. I checked that the test fails without the kept stop. With the bus wrapping in place and only the Delivery.Stopped line removed, the end is said with outcome:failed (walk_said_test.go:339).
Blocking 1: cancel <build> and clear end a walk and its end is never said.cancelAsk (cmd/mesh-controller/queue.go:226) calls recordCancelled, which calls builds.Built, then planBuilt. planBuilt sets the walk failed and saves it (release_plan.go:427-438). This runs in the verb's own process, where inventory.PlanSaved is nil, and no sayingOnTheBus wraps it. markWalk keeps times.ended in that save, so the serving controller never says the end either. The router then holds what the walk touched until the bound. That fails safe, but the end is lost for good. clear reaches the same code through queue.go:319.
Blocking 2: rebuild <module> restarts a failed walk and its start is never said.rebuildCommand (queue.go:351, dispatched at main.go:93) calls joinAPlan (cmd/mesh-controller/plan_retry.go:346-377). joinAPlan uses resumed to set a failed walk building again and saves it. markWalk sees the walk live with times.ended set, so it marks a new start (walksaid.go:70-72). That start is kept in a process with no bus, and no later save says it. This is the fail-open case the PR body names: the retried walk is not held. It is the same restart as plans retry, reached through another verb. Wrap rebuildCommand's joinAPlan, and cancelCommand/clearCommand, in sayingOnTheBus, and add a live test for each the way review 15's tests do.
I went through every other caller of SavePlan/SavePlansMoving. All of them are on the serving side (planBuilt from Built/planFailedBuild, advancePlans/advanceHeld from planTicker, gate and putBack, batches, healers) or already inside sayingOnTheBus (delivery go|stop|check, plans go|stop|close|retry, upgrade release-backlog). replay, kill, pause/resume, pin and node save no plan.
Blocking 3: a hand stop stays on a walk after it is retried.retryPlan, retryRollouts, retryTierWhole and resumed never clear Delivery.Stopped. Nothing in the repository sets it back to empty. A hand-stopped walk can still be retried: retryPlan refuses only walked_again_by/left_out, and a build asked before the stop can fail after it. Once retried, the walk is live again but still carries Stopped, and that has two effects:
If the retried walk then fails, Outcome says stopped, not failed. That is wrong under decision 1.
searchAfterFailures (cmd/mesh-controller/batches.go:1072) skips the walk because Delivery.Stopped != "". Before this change, the retry replaced the "stopped by hand" note, so a retried walk that failed again did start the search for the merge that brought the failure.
Clear Stopped/StoppedWhy when a walk is resumed (in resumed, and where retryRollouts/retryTierWhole set it rolling or building). The delivery owner's stop already had this gap before this change, and the same clear fixes it too.
Not blocking, but name it in the PR body: a hand stop now looks different to mesh-delivery. The PR body says nothing changes for mesh-delivery. That is true of the new walk fields, but not of the kept stop. mesh-delivery copies delivery.stopped into WalkSeen.StoppedBy (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go:531). walkStopped() (model.go:558) and the Delivering rows in table.go:350-380 read that field. After plans stop|close:
A delivery whose walk carried its own commit used to become Failed, with mesh/delivery saying "failure — failed: …". It now becomes Stopped, saying "error — stopped: its walk … was stopped by a person (…): …".
A delivery whose walk carried a later merge used to become Published again, waiting for a walk of its own commit. That walk never came, because searchAfterFailures skips hand stops. It now ends Stopped.
I think this is the right result, and it matches the ADR's outcome. It does change what the operator sees on the pull request. The Stop row's guard still says "stopped through this owner", which is now inaccurate, so that needs a small mesh-catalog follow-up. plans output does not change: the note is the same text, and nothing in the listing reads Delivery.Stopped. A walk that had no Delivery now gets {"awaits":""} from stopByHand. I checked every p.Delivery != nil reader (Waiting, Batch, letGo, searchAfterFailures, release_plan.go:476, batches.go) and none changes behaviour on it.
Ran.sh merge-check.sh at 44983d6 against a throwaway PostgreSQL 18.6 exits 0. That covers gofmt, go vet, go test -race ./... (gcc present) and all 16 listed store tests, including the three new ones (PASS).
Checks at 44983d6:mesh/merge-gate passed at 10:16:52 UTC. It builds build-agent and mesh-controller for ace, g14, novox and shanks, and every machine composes with the change as it did without it. mesh/repo-check passed (its merge-check.sh passed). mesh/delivery says "ready". The branch is based on da1473e, and main is now at 22a2f73 (#245). Both sides change delivery.go and release_plan.go, and git merge-tree merges them without a conflict.
Verdict: request changes at 44983d6
Independent re-review of 44983d6 against ADR 0308 decision 1 and review 15. Both review-15 findings are fixed and proven. Two more verbs keep a walk's start or end mark without saying it, so the claim "every verb that starts or ends a walk says it" does not hold yet. A hand stop also stays on a walk after it is retried.
**Review 15, finding 1 (fixed).** `plans retry` (cmd/mesh-controller/release_plan.go:1568), `plans stop|close` (release_plan.go:1588) and `upgrade release-backlog` (cmd/mesh-controller/release.go:818) now run inside `sayingOnTheBus`. Each one's `HoldPlans` release is deferred inside the closure, so the plans are let go before the closure returns. I ran the three new tests against bd3b23a with the 44983d6 test file. All three fail there: `TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped` says "no walk was said ended", and `TestAWalkRetriedByAPersonIsSaidStartedAgain` and `TestABacklogReleasedByAPersonIsSaidStarted` say "no walk was said started". All three pass at 44983d6.
**Review 15, finding 2 (fixed).** `stopByHand` (release_plan.go:1903-1911) sets `Delivery.Stopped` and `StoppedWhy`, and `Outcome` (internal/inventory/walksaid.go:39) then gives `stopped`. I checked that the test fails without the kept stop. With the bus wrapping in place and only the `Delivery.Stopped` line removed, the end is said with `outcome:failed` (walk_said_test.go:339).
**Blocking 1: `cancel <build>` and `clear` end a walk and its end is never said.** `cancelAsk` (cmd/mesh-controller/queue.go:226) calls `recordCancelled`, which calls `builds.Built`, then `planBuilt`. `planBuilt` sets the walk failed and saves it (release_plan.go:427-438). This runs in the verb's own process, where `inventory.PlanSaved` is nil, and no `sayingOnTheBus` wraps it. `markWalk` keeps `times.ended` in that save, so the serving controller never says the end either. The router then holds what the walk touched until the bound. That fails safe, but the end is lost for good. `clear` reaches the same code through queue.go:319.
**Blocking 2: `rebuild <module>` restarts a failed walk and its start is never said.** `rebuildCommand` (queue.go:351, dispatched at main.go:93) calls `joinAPlan` (cmd/mesh-controller/plan_retry.go:346-377). `joinAPlan` uses `resumed` to set a failed walk building again and saves it. `markWalk` sees the walk live with `times.ended` set, so it marks a new start (walksaid.go:70-72). That start is kept in a process with no bus, and no later save says it. This is the fail-open case the PR body names: the retried walk is not held. It is the same restart as `plans retry`, reached through another verb. Wrap `rebuildCommand`'s `joinAPlan`, and `cancelCommand`/`clearCommand`, in `sayingOnTheBus`, and add a live test for each the way review 15's tests do.
I went through every other caller of `SavePlan`/`SavePlansMoving`. All of them are on the serving side (planBuilt from `Built`/`planFailedBuild`, advancePlans/advanceHeld from planTicker, gate and putBack, batches, healers) or already inside `sayingOnTheBus` (`delivery go|stop|check`, `plans go|stop|close|retry`, `upgrade release-backlog`). `replay`, `kill`, `pause`/`resume`, `pin` and `node` save no plan.
**Blocking 3: a hand stop stays on a walk after it is retried.** `retryPlan`, `retryRollouts`, `retryTierWhole` and `resumed` never clear `Delivery.Stopped`. Nothing in the repository sets it back to empty. A hand-stopped walk can still be retried: `retryPlan` refuses only `walked_again_by`/`left_out`, and a build asked before the stop can fail after it. Once retried, the walk is live again but still carries `Stopped`, and that has two effects:
- If the retried walk then fails, `Outcome` says `stopped`, not `failed`. That is wrong under decision 1.
- `searchAfterFailures` (cmd/mesh-controller/batches.go:1072) skips the walk because `Delivery.Stopped != ""`. Before this change, the retry replaced the "stopped by hand" note, so a retried walk that failed again did start the search for the merge that brought the failure.
Clear `Stopped`/`StoppedWhy` when a walk is resumed (in `resumed`, and where `retryRollouts`/`retryTierWhole` set it rolling or building). The delivery owner's stop already had this gap before this change, and the same clear fixes it too.
**Not blocking, but name it in the PR body: a hand stop now looks different to mesh-delivery.** The PR body says nothing changes for mesh-delivery. That is true of the new walk fields, but not of the kept stop. mesh-delivery copies `delivery.stopped` into `WalkSeen.StoppedBy` (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go:531). `walkStopped()` (model.go:558) and the Delivering rows in table.go:350-380 read that field. After `plans stop|close`:
- A delivery whose walk carried its own commit used to become Failed, with `mesh/delivery` saying "failure — failed: …". It now becomes Stopped, saying "error — stopped: its walk … was stopped by a person (…): …".
- A delivery whose walk carried a later merge used to become Published again, waiting for a walk of its own commit. That walk never came, because `searchAfterFailures` skips hand stops. It now ends Stopped.
I think this is the right result, and it matches the ADR's outcome. It does change what the operator sees on the pull request. The Stop row's guard still says "stopped through this owner", which is now inaccurate, so that needs a small mesh-catalog follow-up. `plans` output does not change: the note is the same text, and nothing in the listing reads `Delivery.Stopped`. A walk that had no `Delivery` now gets `{"awaits":""}` from `stopByHand`. I checked every `p.Delivery != nil` reader (Waiting, Batch, letGo, searchAfterFailures, release_plan.go:476, batches.go) and none changes behaviour on it.
**Ran.** `sh merge-check.sh` at 44983d6 against a throwaway PostgreSQL 18.6 exits 0. That covers gofmt, `go vet`, `go test -race ./...` (gcc present) and all 16 listed store tests, including the three new ones (PASS).
**Checks at 44983d6:** `mesh/merge-gate` passed at 10:16:52 UTC. It builds build-agent and mesh-controller for ace, g14, novox and shanks, and every machine composes with the change as it did without it. `mesh/repo-check` passed (its merge-check.sh passed). `mesh/delivery` says "ready". The branch is based on da1473e, and main is now at 22a2f73 (#245). Both sides change delivery.go and release_plan.go, and `git merge-tree` merges them without a conflict.
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 ace as build-1791800966066838456 (builds --log build-1791800966066838456).
--- 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 1.043s
ok github.com/novox/mesh-controller/cmd/mesh-controller 235.643s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.017s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.026s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.015s
ok github.com/novox/mesh-controller/examples/route-proxy 1.163s
ok github.com/novox/mesh-controller/internal/artifacts 1.110s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.130s
ok github.com/novox/mesh-controller/internal/builder 16.056s
ok github.com/novox/mesh-controller/internal/catalogue 7.732s
ok github.com/novox/mesh-controller/internal/conditions 2.644s
ok github.com/novox/mesh-controller/internal/envfile 1.059s
ok github.com/novox/mesh-controller/internal/facts 1.052s
ok github.com/novox/mesh-controller/internal/identity 4.648s
ok github.com/novox/mesh-controller/internal/inventory 91.435s
ok github.com/novox/mesh-controller/internal/lease 17.762s
ok github.com/novox/mesh-controller/internal/licences 4.233s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.063s
ok github.com/novox/mesh-controller/internal/link 60.115s
ok github.com/novox/mesh-controller/internal/lint 2.412s
ok github.com/novox/mesh-controller/internal/outward 1.094s
ok github.com/novox/mesh-controller/internal/overlay 1.084s
ok github.com/novox/mesh-controller/internal/secrets 1.493s
ok github.com/novox/mesh-controller/internal/store 3.544s
ok github.com/novox/mesh-controller/internal/testbus 1.198s
ok github.com/novox/mesh-controller/internal/token 1.020s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.41s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.42s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.39s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.38s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.40s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.43s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.45s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.44s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.40s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.40s)
./cmd/mesh-controller --- PASS: TestAWalkEndedByCancelOrClearIsSaidEnded (0.47s)
./cmd/mesh-controller --- PASS: TestAWalkRestartedByARebuildIsSaidStarted (0.43s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain (0.44s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.28s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.27s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.26s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.24s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.24s)
**Merge check** at `f9bd41d4`
- `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 ace as `build-1791800966066838456` (`builds --log build-1791800966066838456`).
```
--- 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 1.043s
ok github.com/novox/mesh-controller/cmd/mesh-controller 235.643s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.017s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.026s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.015s
ok github.com/novox/mesh-controller/examples/route-proxy 1.163s
ok github.com/novox/mesh-controller/internal/artifacts 1.110s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.130s
ok github.com/novox/mesh-controller/internal/builder 16.056s
ok github.com/novox/mesh-controller/internal/catalogue 7.732s
ok github.com/novox/mesh-controller/internal/conditions 2.644s
ok github.com/novox/mesh-controller/internal/envfile 1.059s
ok github.com/novox/mesh-controller/internal/facts 1.052s
ok github.com/novox/mesh-controller/internal/identity 4.648s
ok github.com/novox/mesh-controller/internal/inventory 91.435s
ok github.com/novox/mesh-controller/internal/lease 17.762s
ok github.com/novox/mesh-controller/internal/licences 4.233s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.063s
ok github.com/novox/mesh-controller/internal/link 60.115s
ok github.com/novox/mesh-controller/internal/lint 2.412s
ok github.com/novox/mesh-controller/internal/outward 1.094s
ok github.com/novox/mesh-controller/internal/overlay 1.084s
ok github.com/novox/mesh-controller/internal/secrets 1.493s
ok github.com/novox/mesh-controller/internal/store 3.544s
ok github.com/novox/mesh-controller/internal/testbus 1.198s
ok github.com/novox/mesh-controller/internal/token 1.020s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.41s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.42s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.39s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.38s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.40s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.33s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.43s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.45s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.44s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.40s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.40s)
./cmd/mesh-controller --- PASS: TestAWalkEndedByCancelOrClearIsSaidEnded (0.47s)
./cmd/mesh-controller --- PASS: TestAWalkRestartedByARebuildIsSaidStarted (0.43s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain (0.44s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.28s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.27s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.26s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.24s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.24s)
```
Independent third review of f9bd41d against ADR 0308 decision 1 and reviews 15 and 21. The three blocking findings of review 21 are fixed, and each fix is pinned by a test that fails without it. I found nothing blocking. (This is a COMMENT review because the forge refuses a tool's approve, ADR 0266.)
The branch was rebased onto f549647 (#246). 6a898e3 is the rebased 44983d6, and 6a898e3..f9bd41d is the new work.
Each saved walk is published exactly once.
run() installs sayingWhatAVerbKept for every command except serve (cmd/mesh-controller/main.go:66-73). serve keeps its own hook (push.go:252).
sayingOnTheBus replaces the hook for its closure and puts back the hook it found (delivery.go:1257-1260). Inside plans go|stop|close|retry, delivery … and upgrade release-backlog, only its hook runs, so these do not publish twice.
No other code calls publishPlanMoved or sayPlanMoved. The serving controller never publishes a save that another process made.
I proved this with a scratch test I did not commit. It drove plans stop (both hooks in play) and rebuild a (the run() hook alone) through run(), and counted every plan-moved for the walk over 2 s. Each said its end or start exactly once (ended:1 msg:1, started:1 msg:1).
The bus can be unreachable, and the save is still kept. With no bus, SavePlan is refused before anything is kept, because the lease guard reads the bus. I ran rebuild with MESH_BUS_NATS pointed at a dead port and saw: "the plan … is not written: the bus cannot be reached, so whether a controller holds the lease cannot be read and nothing is done". The loud stderr branch in delivery.go:1297-1306 is therefore reached only when the bus goes away between the lease read and the hook's own dial. That case is rare, the branch handles it correctly, and the save is kept.
Verbs are not slowed. The hook dials only on the first save (the reached flag under its mutex), and never for a verb that saves no plan. I checked every caller of SavePlan/SavePlansMoving: no read-only verb (plans listing, queue, status, conditions, node) saves one. A verb that does save opens one more connection, and on a bus that just vanished it waits at most the 10 s dial timeout once (broker/jetstream.go:55). Publishing happens in-line with a 10 s timeout, so the event is on the bus before the verb exits. For the serving controller it stays a goroutine.
goesOnAgain covers every way a walk walks again. Apart from advancing open walks, only three places set a plan building or rolling: resumed (plan_retry.go:250, reached from retryPlan and joinAPlan/rebuild), retryRollouts (:472) and retryTierWhole (:538). All three clear the stop. The other writes of state only work on open plans and never revive a failed one: advancePlans and advanceHeld; planBuilt, which reads OpenPlans; and healer H2, which needs p.Open(). delivery.go:215, release_plan.go:1945 and gate.go:938 only set a walk failed or stopped.
The tests fail without each fix.
At 44983d6 with f9bd41d's walk_said_test.go, all three new tests fail: "no walk was said ended" (:440), "no walk was said started" (:472), and "the walk is building and keeps its stop … Stopped:a person (jochen)" (:506).
At f9bd41d with only the run() hook switched off, the cancel/clear and rebuild tests fail and the third test passes.
At f9bd41d with only the clear in resumed removed, TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain fails.
Ran.sh merge-check.sh at f9bd41d against a throwaway PostgreSQL exits 0. That covers gofmt, go vet, go test -race -count=1 ./... (gcc present) and all 19 listed store tests, including the three new ones (PASS).
Checks at f9bd41d:mesh/merge-gate passed at 10:42:52 UTC. It built build-agent, mesh-controller and route-proxy for ace, g14, novox and shanks, and every machine composes with the change. mesh/repo-check passed (its merge-check.sh passed). mesh/delivery says "ready: it delivers once merged".
Not blocking:
No test covers the clears in retryRollouts and retryTierWhole. With both removed, every test still passes. A walk stopped while it rolls, with nothing failed, and then plans retry'd goes through retryRollouts, so one case there would pin it.
The tests reach the bus through handActConn, so the hook's aBus() dial (the path a real verb process takes) runs in no test. The code is simple, but nothing exercises it.
done sets PlanSaved to nil and does not put back what it found, unlike sayingOnTheBus. This is harmless at process end. In tests that call run() through asTyped, it overwrites a hook the test installed until the test's own cleanup runs.
Review 21's mesh-delivery remark still applies: after plans stop|close a delivery now ends Stopped, and the Stop row's guard text in mesh-catalog needs a small follow-up.
Verdict: approve at f9bd41d
Independent third review of f9bd41d against ADR 0308 decision 1 and reviews 15 and 21. The three blocking findings of review 21 are fixed, and each fix is pinned by a test that fails without it. I found nothing blocking. (This is a COMMENT review because the forge refuses a tool's approve, ADR 0266.)
The branch was rebased onto f549647 (#246). 6a898e3 is the rebased 44983d6, and 6a898e3..f9bd41d is the new work.
**Each saved walk is published exactly once.**
- `run()` installs `sayingWhatAVerbKept` for every command except `serve` (cmd/mesh-controller/main.go:66-73). `serve` keeps its own hook (push.go:252).
- `sayingOnTheBus` replaces the hook for its closure and puts back the hook it found (delivery.go:1257-1260). Inside `plans go|stop|close|retry`, `delivery …` and `upgrade release-backlog`, only its hook runs, so these do not publish twice.
- No other code calls `publishPlanMoved` or `sayPlanMoved`. The serving controller never publishes a save that another process made.
- I proved this with a scratch test I did not commit. It drove `plans stop` (both hooks in play) and `rebuild a` (the run() hook alone) through `run()`, and counted every `plan-moved` for the walk over 2 s. Each said its end or start exactly once (`ended:1 msg:1`, `started:1 msg:1`).
**The bus can be unreachable, and the save is still kept.** With no bus, `SavePlan` is refused before anything is kept, because the lease guard reads the bus. I ran `rebuild` with MESH_BUS_NATS pointed at a dead port and saw: "the plan … is not written: the bus cannot be reached, so whether a controller holds the lease cannot be read and nothing is done". The loud stderr branch in delivery.go:1297-1306 is therefore reached only when the bus goes away between the lease read and the hook's own dial. That case is rare, the branch handles it correctly, and the save is kept.
**Verbs are not slowed.** The hook dials only on the first save (the `reached` flag under its mutex), and never for a verb that saves no plan. I checked every caller of `SavePlan`/`SavePlansMoving`: no read-only verb (`plans` listing, `queue`, `status`, `conditions`, `node`) saves one. A verb that does save opens one more connection, and on a bus that just vanished it waits at most the 10 s dial timeout once (broker/jetstream.go:55). Publishing happens in-line with a 10 s timeout, so the event is on the bus before the verb exits. For the serving controller it stays a goroutine.
**`goesOnAgain` covers every way a walk walks again.** Apart from advancing open walks, only three places set a plan building or rolling: `resumed` (plan_retry.go:250, reached from `retryPlan` and `joinAPlan`/`rebuild`), `retryRollouts` (:472) and `retryTierWhole` (:538). All three clear the stop. The other writes of state only work on open plans and never revive a failed one: advancePlans and advanceHeld; planBuilt, which reads `OpenPlans`; and healer H2, which needs `p.Open()`. `delivery.go:215`, `release_plan.go:1945` and gate.go:938 only set a walk failed or stopped.
**The tests fail without each fix.**
- At 44983d6 with f9bd41d's walk_said_test.go, all three new tests fail: "no walk was said ended" (:440), "no walk was said started" (:472), and "the walk is building and keeps its stop … Stopped:a person (jochen)" (:506).
- At f9bd41d with only the run() hook switched off, the cancel/clear and rebuild tests fail and the third test passes.
- At f9bd41d with only the clear in `resumed` removed, `TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain` fails.
**Ran.** `sh merge-check.sh` at f9bd41d against a throwaway PostgreSQL exits 0. That covers gofmt, `go vet`, `go test -race -count=1 ./...` (gcc present) and all 19 listed store tests, including the three new ones (PASS).
**Checks at f9bd41d:** `mesh/merge-gate` passed at 10:42:52 UTC. It built build-agent, mesh-controller and route-proxy for ace, g14, novox and shanks, and every machine composes with the change. `mesh/repo-check` passed (its merge-check.sh passed). `mesh/delivery` says "ready: it delivers once merged".
Not blocking:
1. No test covers the clears in `retryRollouts` and `retryTierWhole`. With both removed, every test still passes. A walk stopped while it rolls, with nothing failed, and then `plans retry`'d goes through `retryRollouts`, so one case there would pin it.
2. The tests reach the bus through `handActConn`, so the hook's `aBus()` dial (the path a real verb process takes) runs in no test. The code is simple, but nothing exercises it.
3. `done` sets `PlanSaved` to nil and does not put back what it found, unlike `sayingOnTheBus`. This is harmless at process end. In tests that call `run()` through `asTyped`, it overwrites a hook the test installed until the test's own cleanup runs.
4. Review 21's mesh-delivery remark still applies: after `plans stop|close` a delivery now ends Stopped, and the Stop row's guard text in mesh-catalog needs a small follow-up.
The router holds what an open walk touches and tells its result once; it can do so
only if plan-moved names the walk's start and end, the modules and machines it moves,
what a moved tool runner or node-engine carries there, and how long to hold.
plans stop|close, plans retry and upgrade release-backlog kept a walk's start or end mark without saying it,
so the router would hold what the walk touched until its bound; each now runs on the bus as plans go does.
A hand stop keeps who stopped it, so its outcome is stopped, not failed (review 15, ADR 0308, issue 532).
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 …
cancel, clear and rebuild kept a walk's end or restart without saying it, as plans stop, retry and
release-backlog did before them. The controller's run now says every save a verb makes on the bus, so the
next verb cannot lose a mark the same way. A stop is cleared when a walk is resumed, so one that fails again
ends failed and the search for its merge runs (review 21, ADR 0308, issue 532).
Rebased onto main fafa3fd (#249 issue 420, #248 issue 332); new head 1e9dcf0. The body's "Rebased onto main f549647" is superseded by this.
One conflict, in testdata/store-tests only, in the first commit: #249 appended ./internal/inventory TestAConditionsLooksAreKeptBesideTheBucket and ./cmd/mesh-controller TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds, this branch appended TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid and TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded at the same place. Resolved by keeping both: #249's two lines first, then this branch's; the later commits' six lines follow. No other file is changed by both sides (#249 and #248 touch conditions, looks, migration 0095, catalogue seats; neither touches PlanSaved, plans.go, delivery.go, main.go or release_plan.go). The code is otherwise unchanged from the approved f9bd41d.
sh merge-check.sh passed at 1e9dcf0 against a throwaway PostgreSQL 18.6, race on: all 21 listed store tests PASS, including #249's two.
Rebased onto main fafa3fd (#249 issue 420, #248 issue 332); new head 1e9dcf0. The body's "Rebased onto main f549647" is superseded by this.
One conflict, in `testdata/store-tests` only, in the first commit: #249 appended `./internal/inventory TestAConditionsLooksAreKeptBesideTheBucket` and `./cmd/mesh-controller TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds`, this branch appended `TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid` and `TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded` at the same place. Resolved by keeping both: #249's two lines first, then this branch's; the later commits' six lines follow. No other file is changed by both sides (#249 and #248 touch conditions, looks, migration 0095, catalogue seats; neither touches `PlanSaved`, plans.go, delivery.go, main.go or release_plan.go). The code is otherwise unchanged from the approved f9bd41d.
`sh merge-check.sh` passed at 1e9dcf0 against a throwaway PostgreSQL 18.6, race on: all 21 listed store tests PASS, including #249's two.
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 g14 as build-1791802663905073882 (builds --log build-1791802663905073882).
--- 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 2026-10-12T10:57:02Z 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 2026-10-12T10:57:02Z
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 2 module(s) in 2 tier(s): build-agent, mesh-controller
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.044s
ok github.com/novox/mesh-controller/cmd/mesh-controller 248.791s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.008s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.015s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.008s
ok github.com/novox/mesh-controller/examples/route-proxy 1.195s
ok github.com/novox/mesh-controller/internal/artifacts 1.074s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.496s
ok github.com/novox/mesh-controller/internal/builder 15.808s
ok github.com/novox/mesh-controller/internal/catalogue 7.885s
ok github.com/novox/mesh-controller/internal/conditions 2.664s
ok github.com/novox/mesh-controller/internal/envfile 1.034s
ok github.com/novox/mesh-controller/internal/facts 1.042s
ok github.com/novox/mesh-controller/internal/identity 4.951s
ok github.com/novox/mesh-controller/internal/inventory 98.064s
ok github.com/novox/mesh-controller/internal/lease 17.613s
ok github.com/novox/mesh-controller/internal/licences 4.514s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.009s
ok github.com/novox/mesh-controller/internal/link 60.986s
ok github.com/novox/mesh-controller/internal/lint 2.440s
ok github.com/novox/mesh-controller/internal/outward 1.059s
ok github.com/novox/mesh-controller/internal/overlay 1.038s
ok github.com/novox/mesh-controller/internal/secrets 1.552s
ok github.com/novox/mesh-controller/internal/store 3.398s
ok github.com/novox/mesh-controller/internal/testbus 1.201s
ok github.com/novox/mesh-controller/internal/token 1.024s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.36s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.36s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.36s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.34s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.33s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.31s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.36s)
./cmd/mesh-controller --- PASS: TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds (0.25s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.40s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.41s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.34s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.36s)
./cmd/mesh-controller --- PASS: TestAWalkEndedByCancelOrClearIsSaidEnded (0.42s)
./cmd/mesh-controller --- PASS: TestAWalkRestartedByARebuildIsSaidStarted (0.36s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain (0.42s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.23s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.20s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.22s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.21s)
./internal/inventory --- PASS: TestAConditionsLooksAreKeptBesideTheBucket (0.19s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.22s)
**Merge check** at `1e9dcf06`
- `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 g14 as `build-1791802663905073882` (`builds --log build-1791802663905073882`).
```
--- 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 2026-10-12T10:57:02Z 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 2026-10-12T10:57:02Z
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 2 module(s) in 2 tier(s): build-agent, mesh-controller
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-controller/cmd/mesh-builder 1.044s
ok github.com/novox/mesh-controller/cmd/mesh-controller 248.791s
ok github.com/novox/mesh-controller/examples/objectstore-provisioner 1.008s
ok github.com/novox/mesh-controller/examples/postgres-provisioner 1.015s
ok github.com/novox/mesh-controller/examples/redis-provisioner 1.008s
ok github.com/novox/mesh-controller/examples/route-proxy 1.195s
ok github.com/novox/mesh-controller/internal/artifacts 1.074s
? github.com/novox/mesh-controller/internal/beside [no test files]
ok github.com/novox/mesh-controller/internal/broker 4.496s
ok github.com/novox/mesh-controller/internal/builder 15.808s
ok github.com/novox/mesh-controller/internal/catalogue 7.885s
ok github.com/novox/mesh-controller/internal/conditions 2.664s
ok github.com/novox/mesh-controller/internal/envfile 1.034s
ok github.com/novox/mesh-controller/internal/facts 1.042s
ok github.com/novox/mesh-controller/internal/identity 4.951s
ok github.com/novox/mesh-controller/internal/inventory 98.064s
ok github.com/novox/mesh-controller/internal/lease 17.613s
ok github.com/novox/mesh-controller/internal/licences 4.514s
ok github.com/novox/mesh-controller/internal/licences/adapters 1.009s
ok github.com/novox/mesh-controller/internal/link 60.986s
ok github.com/novox/mesh-controller/internal/lint 2.440s
ok github.com/novox/mesh-controller/internal/outward 1.059s
ok github.com/novox/mesh-controller/internal/overlay 1.038s
ok github.com/novox/mesh-controller/internal/secrets 1.552s
ok github.com/novox/mesh-controller/internal/store 3.398s
ok github.com/novox/mesh-controller/internal/testbus 1.201s
ok github.com/novox/mesh-controller/internal/token 1.024s
store tests, each run against the store:
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeWalksTheOthersAgain (0.36s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeEndsTheMergesThatContainIt (0.36s)
./cmd/mesh-controller --- PASS: TestAStopWalksTheOthersAgainInTheOpenBatch (0.36s)
./cmd/mesh-controller --- PASS: TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt (0.34s)
./cmd/mesh-controller --- PASS: TestAStopWithoutAMergeLeavesItOutOfABatch (0.33s)
./cmd/mesh-controller --- PASS: TestSavingPlansAndMovingMergesIsOneAct (0.31s)
./cmd/mesh-controller --- PASS: TestNodeShowAndStatusReadTheLastSendFromTheStore (0.36s)
./cmd/mesh-controller --- PASS: TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds (0.25s)
./cmd/mesh-controller --- PASS: TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded (0.40s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedOrClosedByAPersonIsSaidEndedStopped (0.41s)
./cmd/mesh-controller --- PASS: TestAWalkRetriedByAPersonIsSaidStartedAgain (0.34s)
./cmd/mesh-controller --- PASS: TestABacklogReleasedByAPersonIsSaidStarted (0.36s)
./cmd/mesh-controller --- PASS: TestAWalkEndedByCancelOrClearIsSaidEnded (0.42s)
./cmd/mesh-controller --- PASS: TestAWalkRestartedByARebuildIsSaidStarted (0.36s)
./cmd/mesh-controller --- PASS: TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain (0.42s)
./internal/inventory --- PASS: TestTheLastSendSaysWhatTheMachineAnswered (0.23s)
./internal/inventory --- PASS: TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend (0.20s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain (0.22s)
./internal/inventory --- PASS: TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance (0.21s)
./internal/inventory --- PASS: TestAConditionsLooksAreKeptBesideTheBucket (0.19s)
./internal/inventory --- PASS: TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid (0.22s)
```
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.
Builds the controller part of novox/hq ADR 0308 decision 1 and the bound of decision 5, for issue 532. The router part (messenger, mesh-catalog) builds against these fields and merges after this. Rebased onto main
f549647.What changes
plan-movedis extended, not doubled. Its body is still the walk whole (inventory.Plan). On the save that starts a walk and the save that ends it, the body also carries:walk,phase(started|ended),touches([{machine, modules, carried}], never null),outcome(delivered|failed|stopped, on the end only),bound_seconds,budget_class(core|leaf|none),started_at, andended_aton the end. Every other save carries none of these fields.internal/inventory/walksaid.go).SavePlansMovingmarkstimes.started/times.endedin the same write, and takes the mark back if the write is refused. A controller replaced mid-walk (a core walk replaces it) does not say the start again and does not miss the end. A retried walk starts again. Batches, walks waiting for their word, walks that end without starting, and walks kept beforetimesexisted say nothing. Release plans are said.runinstallsinventory.PlanSavedfor every command exceptserve(sayingWhatAVerbKept, delivery.go). It reaches the bus on the first save only, publishes before the save returns, and reads touches from the inventory that kept the walk:PlanSavednow receives that inventory. Socancel,clear,rebuildand any later verb say a start or end without a wrap of their own. When the bus cannot be reached, a kept start or end is said loudly on the verb's error output.sayingOnTheBus(delivery go|stop|check,plans go|stop|close|retry,upgrade release-backlog) still wraps its verbs, and now restores the hook it found instead of clearing it.delivery.stopped, or superseded, gives stopped. Any other failure gives failed.plans stop|closekeeps who stopped it and why indelivery.stopped/stopped_why, so a person's stop endsstopped, notfailed.goesOnAgain):plans retry(failed builds, rollouts, a whole tier) andrebuildjoining the walk cleardelivery.stopped/stopped_why. A retried walk that fails again endsfailed, andsearchAfterFailuresruns for it again. This also closes the same gap for a stop by the delivery's owner.cmd/mesh-controller/walk_said.go):tools, or a bundle withloads). On a machine where mesh-host moves, every module there is added. What is on a machine is read fromsent_buildsplus its assignments.modules, and the added ones are also listed incarried.deliveryBudgets[class](the map the D16 over-budget probe already uses: core 10 min, leaf 3 min). That gives 1200 s for core and 360 s for leaf. The bound is 900 s, not doubled, when the walk has no budget: a release plan, a walk whose every module's policy records (held for a person, ADR 0282 §1), or a walk with no class.What the operator sees change: a hand stop now ends a delivery as Stopped. mesh-delivery copies
delivery.stoppedintoWalkSeen.StoppedBy(mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go:531), andwalkStopped()(model.go:558) and the Delivering rows (table.go:350-380) read it. Afterplans stop|close:mesh/delivery: "failure — failed: …"); it now becomes Stopped ("error — stopped: its walk … was stopped by a person (…): …");The Stop row's guard text still says "stopped through this owner"; a small mesh-catalog follow-up for it is opened separately.
plansoutput does not change.Limits, named
startedandendedmay arrive out of order on a very short walk; the router orders bystarted_at/ended_atorrevision, not by arrival.Consumers. mesh-delivery decodes
plan-movedwith plainjson.Unmarshalinto itsWalk, which has no field the new names collide with. The kept stop changes what it shows, as above. The view principal (#203) only subscribes. No schema change: the marks live in thetimesJSON column.Tests (ADR 0308 How it is checked §1):
walk_said_test.go: the start names modules and machines, including what a moved tool runner carries (and not postgres, which serves no tools). A moved node-engine carries everything on its machine. A module whose policy records is not touched. The end names what was reached and the outcome. Bounds: core 1200, leaf 360, none 900. The body shape is checked, and no walk fields appear on ordinary saves.delivery gothendelivery stop;plans stopandplans close(ended, stopped);plans retry(started);upgrade release-backlog(started). Through the controller'srun, as typed:cancel <build>andclear(ended, failed,TestAWalkEndedByCancelOrClearIsSaidEnded);rebuild a(started,TestAWalkRestartedByARebuildIsSaidStarted); a walk stopped by hand, then retried byplans retryorrebuild, walks without its stop and endsfailedwhen it fails again (TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgain). Each new test fails without its fix: no event within 5 s without the hook inrun; the stop kept withoutgoesOnAgain.walksaid_test.go: start once, end once, retry starts again, the cases that say nothing, and outcome. Store test: the mark is kept, and a refused save leaves the walk unmarked.testdata/store-tests.sh merge-check.shpassed locally with a throwaway PostgreSQL 18, race on, every listed store test PASS.Reviewers' checklist
servehasPlanSavedset byrun;sayingOnTheBusrestores what it found; nothing else sets it outside tests.undoinSavePlansMoving).goesOnAgain(resumed,retryRollouts,retryTierWhole).WalkPhaseschanges for release plans that now get atimesholding only the marks.This is a controller rollout, so merge only while
mesh-controller.planssays no plan is mid-gate. Merge this before the messenger PR. By adr0308@g14. Not merged; waits for review.Delivery
novox/mesh-controller@1e9dcf066750— delivered since 2026-10-12T11:19:07ZDelivery plan — builds build-agent, mesh-controller → ace, g14, novox, shanks; no bus step
Machines
Transitions
The commit's note under
refs/notes/mesh-plankeeps every transition:git log --notes=mesh-plan.Merge check at
bd3b23a8mesh/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-1791798923847656032(builds --log build-1791798923847656032).VERDICT: REQUEST_CHANGES (sent as a comment review: the forge refuses a tool's approve or request-changes, ADR 0266). Independent review at
bd3b23a(ADR 0308 §1, §5; issue 532). Two blocking findings.BLOCKING: a start or end mark is persisted but never published when a walk starts or ends through a verb that has no bus. markWalk runs inside every SavePlansMoving call (internal/inventory/plans.go:483-496), but PlanSaved is set only in serve (push.go:252) and inside sayingOnTheBus (delivery.go:1225). Three verbs save a walk's start or end outside sayingOnTheBus:
plans stop|close(cmd/mesh-controller/release_plan.go:1593-1599): open walk -> failed. Times.Ended is written, noendedevent is published, and because the mark is kept, no later save says it. The router holds what the walk touched until its bound (up to 20 min) after a person stopped it.plans retry(retryPlan -> retryRollouts, plan_retry.go:462-464): failed -> rolling. Times.Started is written anew, nostartedevent. The retried walk is never held, and its laterended(said by serve) arrives for a walk the router never saw start.upgrade release-backlog(release.go:821 -> releaseBacklog, release.go:664-668): creates a live release plan. Started is marked, no event.Scenario: the operator runs
mesh-controller.plans stop <walk>mid-walk; the stop is kept, the walk'sendedis never on the bus, and messages about its modules stay held until the bound passes. Fix: wrap these three in sayingOnTheBus(ctx, inv, ...) asplans goalready is (release_plan.go:1553), or do not mark when no one will say the save. A test that drivesplans stopandplans retrythrough the command and expects the event would pin it.BLOCKING: a walk stopped by a person through
plans stopis reported with outcomefailed, notstopped. Plan.Outcome (internal/inventory/walksaid.go:33-45) returns stopped only for PlanSuperseded or a failed plan with Delivery.Stopped set. Only stopWalkWithout (delivery.go:178-184) sets Delivery.Stopped;plans stop|closesets State=failed and a Note only (release_plan.go:1593-1594). ADR 0308 §1 namesstoppedfor this case, and the router's end message would tell the operator their own stop as a failure. Fix: haveplans stop|closerecord who stopped it (for example Delivery.Stopped/StoppedWhy) or let Outcome recognise the hand stop, with a test.Non-blocking remarks:
startedandendedgo out from separate goroutines and may reach the router out of order on a very short walk; the router must use started_at/ended_at or revision, not arrival order.Checked and fine: marks kept in times JSON in the same transaction and undone on a refused write (plans.go:483-496, test TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid); a restart neither repeats nor drops a mark said by serve; ended once (markWalk needs Started set and Ended nil); retry re-opens with a new start; carried = modules on the machine a moved node-engine (all) or node-tools (tool-serving ones) brings along, and
modulesincludes them (walk_said.go:200-222); terminal states done/failed/superseded all map; no json name of walkSaid collides with Plan's, the walk fields are absent when phase is absent (nil embedded pointer), and mesh-delivery's walk model (mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go) has no colliding field and decodes leniently.go vetclean;go test -race -count=1 ./...passes in full.Required checks at
bd3b23awhen reviewed: mesh/merge-gate pending (building on ace), mesh/repo-check missing.Merge check at
44983d6dmesh/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 g14 as
build-1791799722338334168(builds --log build-1791799722338334168).Verdict: request changes at
44983d6Independent re-review of
44983d6against ADR 0308 decision 1 and review 15. Both review-15 findings are fixed and proven. Two more verbs keep a walk's start or end mark without saying it, so the claim "every verb that starts or ends a walk says it" does not hold yet. A hand stop also stays on a walk after it is retried.Review 15, finding 1 (fixed).
plans retry(cmd/mesh-controller/release_plan.go:1568),plans stop|close(release_plan.go:1588) andupgrade release-backlog(cmd/mesh-controller/release.go:818) now run insidesayingOnTheBus. Each one'sHoldPlansrelease is deferred inside the closure, so the plans are let go before the closure returns. I ran the three new tests againstbd3b23awith the44983d6test file. All three fail there:TestAWalkStoppedOrClosedByAPersonIsSaidEndedStoppedsays "no walk was said ended", andTestAWalkRetriedByAPersonIsSaidStartedAgainandTestABacklogReleasedByAPersonIsSaidStartedsay "no walk was said started". All three pass at44983d6.Review 15, finding 2 (fixed).
stopByHand(release_plan.go:1903-1911) setsDelivery.StoppedandStoppedWhy, andOutcome(internal/inventory/walksaid.go:39) then givesstopped. I checked that the test fails without the kept stop. With the bus wrapping in place and only theDelivery.Stoppedline removed, the end is said withoutcome:failed(walk_said_test.go:339).Blocking 1:
cancel <build>andclearend a walk and its end is never said.cancelAsk(cmd/mesh-controller/queue.go:226) callsrecordCancelled, which callsbuilds.Built, thenplanBuilt.planBuiltsets the walk failed and saves it (release_plan.go:427-438). This runs in the verb's own process, whereinventory.PlanSavedis nil, and nosayingOnTheBuswraps it.markWalkkeepstimes.endedin that save, so the serving controller never says the end either. The router then holds what the walk touched until the bound. That fails safe, but the end is lost for good.clearreaches the same code through queue.go:319.Blocking 2:
rebuild <module>restarts a failed walk and its start is never said.rebuildCommand(queue.go:351, dispatched at main.go:93) callsjoinAPlan(cmd/mesh-controller/plan_retry.go:346-377).joinAPlanusesresumedto set a failed walk building again and saves it.markWalksees the walk live withtimes.endedset, so it marks a new start (walksaid.go:70-72). That start is kept in a process with no bus, and no later save says it. This is the fail-open case the PR body names: the retried walk is not held. It is the same restart asplans retry, reached through another verb. WraprebuildCommand'sjoinAPlan, andcancelCommand/clearCommand, insayingOnTheBus, and add a live test for each the way review 15's tests do.I went through every other caller of
SavePlan/SavePlansMoving. All of them are on the serving side (planBuilt fromBuilt/planFailedBuild, advancePlans/advanceHeld from planTicker, gate and putBack, batches, healers) or already insidesayingOnTheBus(delivery go|stop|check,plans go|stop|close|retry,upgrade release-backlog).replay,kill,pause/resume,pinandnodesave no plan.Blocking 3: a hand stop stays on a walk after it is retried.
retryPlan,retryRollouts,retryTierWholeandresumednever clearDelivery.Stopped. Nothing in the repository sets it back to empty. A hand-stopped walk can still be retried:retryPlanrefuses onlywalked_again_by/left_out, and a build asked before the stop can fail after it. Once retried, the walk is live again but still carriesStopped, and that has two effects:Outcomesaysstopped, notfailed. That is wrong under decision 1.searchAfterFailures(cmd/mesh-controller/batches.go:1072) skips the walk becauseDelivery.Stopped != "". Before this change, the retry replaced the "stopped by hand" note, so a retried walk that failed again did start the search for the merge that brought the failure.Clear
Stopped/StoppedWhywhen a walk is resumed (inresumed, and whereretryRollouts/retryTierWholeset it rolling or building). The delivery owner's stop already had this gap before this change, and the same clear fixes it too.Not blocking, but name it in the PR body: a hand stop now looks different to mesh-delivery. The PR body says nothing changes for mesh-delivery. That is true of the new walk fields, but not of the kept stop. mesh-delivery copies
delivery.stoppedintoWalkSeen.StoppedBy(mesh-catalog modules/mesh-delivery/cmd/mesh-delivery/model.go:531).walkStopped()(model.go:558) and the Delivering rows in table.go:350-380 read that field. Afterplans stop|close:mesh/deliverysaying "failure — failed: …". It now becomes Stopped, saying "error — stopped: its walk … was stopped by a person (…): …".searchAfterFailuresskips hand stops. It now ends Stopped.I think this is the right result, and it matches the ADR's outcome. It does change what the operator sees on the pull request. The Stop row's guard still says "stopped through this owner", which is now inaccurate, so that needs a small mesh-catalog follow-up.
plansoutput does not change: the note is the same text, and nothing in the listing readsDelivery.Stopped. A walk that had noDeliverynow gets{"awaits":""}fromstopByHand. I checked everyp.Delivery != nilreader (Waiting, Batch, letGo, searchAfterFailures, release_plan.go:476, batches.go) and none changes behaviour on it.Ran.
sh merge-check.shat44983d6against a throwaway PostgreSQL 18.6 exits 0. That covers gofmt,go vet,go test -race ./...(gcc present) and all 16 listed store tests, including the three new ones (PASS).Checks at
44983d6:mesh/merge-gatepassed at 10:16:52 UTC. It builds build-agent and mesh-controller for ace, g14, novox and shanks, and every machine composes with the change as it did without it.mesh/repo-checkpassed (its merge-check.sh passed).mesh/deliverysays "ready". The branch is based onda1473e, and main is now at22a2f73(#245). Both sides change delivery.go and release_plan.go, andgit merge-treemerges them without a conflict.44983d6d29tof9bd41d436Merge check at
f9bd41d4mesh/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 ace as
build-1791800966066838456(builds --log build-1791800966066838456).Verdict: approve at
f9bd41dIndependent third review of
f9bd41dagainst ADR 0308 decision 1 and reviews 15 and 21. The three blocking findings of review 21 are fixed, and each fix is pinned by a test that fails without it. I found nothing blocking. (This is a COMMENT review because the forge refuses a tool's approve, ADR 0266.)The branch was rebased onto
f549647(#246).6a898e3is the rebased44983d6, and 6a898e3..f9bd41d is the new work.Each saved walk is published exactly once.
run()installssayingWhatAVerbKeptfor every command exceptserve(cmd/mesh-controller/main.go:66-73).servekeeps its own hook (push.go:252).sayingOnTheBusreplaces the hook for its closure and puts back the hook it found (delivery.go:1257-1260). Insideplans go|stop|close|retry,delivery …andupgrade release-backlog, only its hook runs, so these do not publish twice.publishPlanMovedorsayPlanMoved. The serving controller never publishes a save that another process made.plans stop(both hooks in play) andrebuild a(the run() hook alone) throughrun(), and counted everyplan-movedfor the walk over 2 s. Each said its end or start exactly once (ended:1 msg:1,started:1 msg:1).The bus can be unreachable, and the save is still kept. With no bus,
SavePlanis refused before anything is kept, because the lease guard reads the bus. I ranrebuildwith MESH_BUS_NATS pointed at a dead port and saw: "the plan … is not written: the bus cannot be reached, so whether a controller holds the lease cannot be read and nothing is done". The loud stderr branch in delivery.go:1297-1306 is therefore reached only when the bus goes away between the lease read and the hook's own dial. That case is rare, the branch handles it correctly, and the save is kept.Verbs are not slowed. The hook dials only on the first save (the
reachedflag under its mutex), and never for a verb that saves no plan. I checked every caller ofSavePlan/SavePlansMoving: no read-only verb (planslisting,queue,status,conditions,node) saves one. A verb that does save opens one more connection, and on a bus that just vanished it waits at most the 10 s dial timeout once (broker/jetstream.go:55). Publishing happens in-line with a 10 s timeout, so the event is on the bus before the verb exits. For the serving controller it stays a goroutine.goesOnAgaincovers every way a walk walks again. Apart from advancing open walks, only three places set a plan building or rolling:resumed(plan_retry.go:250, reached fromretryPlanandjoinAPlan/rebuild),retryRollouts(:472) andretryTierWhole(:538). All three clear the stop. The other writes of state only work on open plans and never revive a failed one: advancePlans and advanceHeld; planBuilt, which readsOpenPlans; and healer H2, which needsp.Open().delivery.go:215,release_plan.go:1945and gate.go:938 only set a walk failed or stopped.The tests fail without each fix.
44983d6with f9bd41d's walk_said_test.go, all three new tests fail: "no walk was said ended" (:440), "no walk was said started" (:472), and "the walk is building and keeps its stop … Stopped:a person (jochen)" (:506).f9bd41dwith only the run() hook switched off, the cancel/clear and rebuild tests fail and the third test passes.f9bd41dwith only the clear inresumedremoved,TestAWalkStoppedByHandAndRetriedEndsFailedWhenItFailsAgainfails.Ran.
sh merge-check.shatf9bd41dagainst a throwaway PostgreSQL exits 0. That covers gofmt,go vet,go test -race -count=1 ./...(gcc present) and all 19 listed store tests, including the three new ones (PASS).Checks at
f9bd41d:mesh/merge-gatepassed at 10:42:52 UTC. It built build-agent, mesh-controller and route-proxy for ace, g14, novox and shanks, and every machine composes with the change.mesh/repo-checkpassed (its merge-check.sh passed).mesh/deliverysays "ready: it delivers once merged".Not blocking:
retryRolloutsandretryTierWhole. With both removed, every test still passes. A walk stopped while it rolls, with nothing failed, and thenplans retry'd goes throughretryRollouts, so one case there would pin it.handActConn, so the hook'saBus()dial (the path a real verb process takes) runs in no test. The code is simple, but nothing exercises it.donesetsPlanSavedto nil and does not put back what it found, unlikesayingOnTheBus. This is harmless at process end. In tests that callrun()throughasTyped, it overwrites a hook the test installed until the test's own cleanup runs.plans stop|closea delivery now ends Stopped, and the Stop row's guard text in mesh-catalog needs a small follow-up.f9bd41d436to1e9dcf0667Rebased onto main
fafa3fd(#249 issue 420, #248 issue 332); new head1e9dcf0. The body's "Rebased onto main f549647" is superseded by this.One conflict, in
testdata/store-testsonly, in the first commit: #249 appended./internal/inventory TestAConditionsLooksAreKeptBesideTheBucketand./cmd/mesh-controller TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds, this branch appendedTestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaidandTestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEndedat the same place. Resolved by keeping both: #249's two lines first, then this branch's; the later commits' six lines follow. No other file is changed by both sides (#249 and #248 touch conditions, looks, migration 0095, catalogue seats; neither touchesPlanSaved, plans.go, delivery.go, main.go or release_plan.go). The code is otherwise unchanged from the approvedf9bd41d.sh merge-check.shpassed at1e9dcf0against a throwaway PostgreSQL 18.6, race on: all 21 listed store tests PASS, including #249's two.Merge check at
1e9dcf06mesh/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 g14 as
build-1791802663905073882(builds --log build-1791802663905073882).