Replaces #89/#261; same change, history without a built binary. This PR carries #89's commit unchanged (cb0a07b).
What
The pre-apply probe of the package manager (sys.Confirm) no longer refuses the whole declaration on arch and alpine. When pacman or apk does not answer, the node-engine says so before applying, applies everything else (files, units, accounts, its own upgrade), and every package resource fails with the probe's words. The system's PackageInstalled already asks the same probe first. The report names the probe under one stable key, the package manager (pacman), beside apply. That lets the controller count it as one failure that goes on (the condition is in mesh-controller, same branch name). mesh-host apply by hand does the same and exits non-zero with "applied only in part" when no package was declared, and it never records known-good.
The android host still refuses whole. Its probe asks the system itself, which nothing the mesh declares can remove, and nothing else guards against the wrong machine.
Issue 234's generation check still runs first. A declaration composed from an older view of the assignments is refused before anything is applied, as the decision of 2026-10-11 requires.
Why
On 2026-10-04 a declaration that left the package manager's module out removed pacman.conf. From then on, every declaration was refused whole for 90 minutes, including the one that would have written the file back and the node-engine's own upgrade. A person on the machine was the only way back. The decision recorded on the issue (2026-10-11): a failed probe fails only what needs the tool.
How it was tested
TestReplay233 (cmd/mesh-host/replay233_test.go) is the replay. A pacman on PATH answers as the workstation's did. The test sends a declaration with the configuration file plus a package through applyAndKeep, the link's path. It asserts that nothing was refused whole, the file is written and recorded, and the report carries the probe's words and the package. It uses only what the node-engine had before the fix: it fails on main and passes here.
The same test covers mesh-host apply by hand. A further test checks that with no package declared the apply is still said to be partial. Another checks that android still refuses whole.
sh merge-check.sh: green.
Trial merge with current main (dce0eef, #88 merged): no conflict, and the merged tree builds, vets, and passes cmd/mesh-host, internal/apply and internal/link. Trial merge of that result with #87's branch: no conflict.
Replay: R233 (cmd/mesh-host TestReplay233), to be registered in mesh-lab once this merges.
Fixes issue 233
Replaces #89/#261; same change, history without a built binary. This PR carries #89's commit unchanged (cb0a07b).
## What
The pre-apply probe of the package manager (`sys.Confirm`) no longer refuses the whole declaration on arch and alpine. When pacman or apk does not answer, the node-engine says so before applying, applies everything else (files, units, accounts, its own upgrade), and every package resource fails with the probe's words. The system's `PackageInstalled` already asks the same probe first. The report names the probe under one stable key, `the package manager (pacman)`, beside `apply`. That lets the controller count it as one failure that goes on (the condition is in mesh-controller, same branch name). `mesh-host apply` by hand does the same and exits non-zero with "applied only in part" when no package was declared, and it never records known-good.
The android host still refuses whole. Its probe asks the system itself, which nothing the mesh declares can remove, and nothing else guards against the wrong machine.
Issue 234's generation check still runs first. A declaration composed from an older view of the assignments is refused before anything is applied, as the decision of 2026-10-11 requires.
## Why
On 2026-10-04 a declaration that left the package manager's module out removed `pacman.conf`. From then on, every declaration was refused whole for 90 minutes, including the one that would have written the file back and the node-engine's own upgrade. A person on the machine was the only way back. The decision recorded on the issue (2026-10-11): a failed probe fails only what needs the tool.
## How it was tested
- `TestReplay233` (cmd/mesh-host/replay233_test.go) is the replay. A pacman on PATH answers as the workstation's did. The test sends a declaration with the configuration file plus a package through `applyAndKeep`, the link's path. It asserts that nothing was refused whole, the file is written and recorded, and the report carries the probe's words and the package. It uses only what the node-engine had before the fix: it fails on main and passes here.
- The same test covers `mesh-host apply` by hand. A further test checks that with no package declared the apply is still said to be partial. Another checks that android still refuses whole.
- `sh merge-check.sh`: green.
- Trial merge with current main (dce0eef, #88 merged): no conflict, and the merged tree builds, vets, and passes `cmd/mesh-host`, `internal/apply` and `internal/link`. Trial merge of that result with #87's branch: no conflict.
Replay: R233 (cmd/mesh-host `TestReplay233`), to be registered in mesh-lab once this merges.
Fixes issue 233
mesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)
mesh/repo-check: PASS — its merge-check.sh passed
Delivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
tier 0: mesh-host
ace: receives mesh-host
g14: receives mesh-host
novox: receives mesh-host
shanks: receives mesh-host
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 mesh-host is judged on its first machine
read by no module's build: cmd/mesh-host/probe_test.go, cmd/mesh-host/replay233_test.go
Run by the build seat on ace as build-1791809446739400983 (builds --log build-1791809446739400983).
--- the gate: mesh-host
mesh-host: ok
long-running resources without health: 0
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 0 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T12:51:20Z 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 2026-10-12T12:51:20Z
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-host
notes:
- cmd/mesh-host/probe_test.go and cmd/mesh-host/replay233_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-host/cmd/mesh-bootstrap 1.014s
ok github.com/novox/mesh-host/cmd/mesh-host 5.784s
ok github.com/novox/mesh-host/examples 1.021s
ok github.com/novox/mesh-host/internal/accounts 1.570s
ok github.com/novox/mesh-host/internal/apply 13.240s
ok github.com/novox/mesh-host/internal/bootstrap 1.592s
ok github.com/novox/mesh-host/internal/bundle 1.016s
ok github.com/novox/mesh-host/internal/declaration 1.085s
ok github.com/novox/mesh-host/internal/firewall 1.770s
ok github.com/novox/mesh-host/internal/identity 1.164s
ok github.com/novox/mesh-host/internal/image 1.068s
ok github.com/novox/mesh-host/internal/inventory 1.095s
ok github.com/novox/mesh-host/internal/link 10.799s
ok github.com/novox/mesh-host/internal/liveness 2.175s
ok github.com/novox/mesh-host/internal/meshcli 1.626s
ok github.com/novox/mesh-host/internal/network 1.113s
ok github.com/novox/mesh-host/internal/outward 1.077s
ok github.com/novox/mesh-host/internal/profile 1.535s
ok github.com/novox/mesh-host/internal/reachable 1.060s
ok github.com/novox/mesh-host/internal/store 1.237s
ok github.com/novox/mesh-host/internal/system 1.065s
ok github.com/novox/mesh-host/internal/tunnel 1.066s
ok github.com/novox/mesh-host/internal/units 1.062s
ok github.com/novox/mesh-host/internal/upgrade 1.180s
ok github.com/novox/mesh-host/internal/witness 15.137s
ok github.com/novox/mesh-host/packaging 1.037s
? github.com/novox/mesh-host/rootsearch [no test files]
ok github.com/novox/mesh-host/validate 1.049s
**Merge check** at `cb0a07ba`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds mesh-host → ace, g14, novox, shanks; no bus step
- tier 0: mesh-host
- ace: receives mesh-host
- g14: receives mesh-host
- novox: receives mesh-host
- shanks: receives mesh-host
- 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 mesh-host is judged on its first machine
- read by no module's build: cmd/mesh-host/probe_test.go, cmd/mesh-host/replay233_test.go
Run by the build seat on ace as `build-1791809446739400983` (`builds --log build-1791809446739400983`).
```
--- the gate: mesh-host
mesh-host: ok
long-running resources without health: 0
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 0 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T12:51:20Z 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 2026-10-12T12:51:20Z
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-host
notes:
- cmd/mesh-host/probe_test.go and cmd/mesh-host/replay233_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-host/cmd/mesh-bootstrap 1.014s
ok github.com/novox/mesh-host/cmd/mesh-host 5.784s
ok github.com/novox/mesh-host/examples 1.021s
ok github.com/novox/mesh-host/internal/accounts 1.570s
ok github.com/novox/mesh-host/internal/apply 13.240s
ok github.com/novox/mesh-host/internal/bootstrap 1.592s
ok github.com/novox/mesh-host/internal/bundle 1.016s
ok github.com/novox/mesh-host/internal/declaration 1.085s
ok github.com/novox/mesh-host/internal/firewall 1.770s
ok github.com/novox/mesh-host/internal/identity 1.164s
ok github.com/novox/mesh-host/internal/image 1.068s
ok github.com/novox/mesh-host/internal/inventory 1.095s
ok github.com/novox/mesh-host/internal/link 10.799s
ok github.com/novox/mesh-host/internal/liveness 2.175s
ok github.com/novox/mesh-host/internal/meshcli 1.626s
ok github.com/novox/mesh-host/internal/network 1.113s
ok github.com/novox/mesh-host/internal/outward 1.077s
ok github.com/novox/mesh-host/internal/profile 1.535s
ok github.com/novox/mesh-host/internal/reachable 1.060s
ok github.com/novox/mesh-host/internal/store 1.237s
ok github.com/novox/mesh-host/internal/system 1.065s
ok github.com/novox/mesh-host/internal/tunnel 1.066s
ok github.com/novox/mesh-host/internal/units 1.062s
ok github.com/novox/mesh-host/internal/upgrade 1.180s
ok github.com/novox/mesh-host/internal/witness 15.137s
ok github.com/novox/mesh-host/packaging 1.037s
? github.com/novox/mesh-host/rootsearch [no test files]
ok github.com/novox/mesh-host/validate 1.049s
```
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
Deliverynovox/mesh-host@74ed7cbe0ab4 — delivered since 2026-10-12T14:20:23Z
Delivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
tier 0: mesh-host
ace: receives mesh-host
g14: receives mesh-host
novox: receives mesh-host
shanks: receives mesh-host
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 mesh-host is judged on its first machine
composed together: pass — every machine composes with the change as it did without (4 of 4 compose)
Machines
mesh-host on ace: passed — healthy 3 times over 2m50s
mesh-host on the rest running it: sent
Transitions
2026-10-12 13:39 (new) → proposed (announced): novox/mesh-host#90's head announced
2026-10-12 13:42 proposed → checked (checked): the verdict names this commit
2026-10-12 13:42 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
2026-10-12 14:16 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 14:16 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 14:20 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-host@74ed7cbe0ab4` — **delivered** since 2026-10-12T14:20:23Z
**Delivery plan** — builds mesh-host → ace, g14, novox, shanks; no bus step
- tier 0: mesh-host
- ace: receives mesh-host
- g14: receives mesh-host
- novox: receives mesh-host
- shanks: receives mesh-host
- 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 mesh-host is judged on its first machine
**Group** `fix/233-a-failed-probe-fails-only-what-needs-the-tool-2`, in order: novox/mesh-host@74ed7cbe0ab4 → novox/mesh-controller@d4de4543ccbf
- novox/mesh-host@74ed7cbe0ab4 before novox/mesh-controller@d4de4543ccbf: engine before controller
- composed together: pass — every machine composes with the change as it did without (4 of 4 compose)
**Machines**
- mesh-host on ace: passed — healthy 3 times over 2m50s
- mesh-host on the rest running it: sent
**Transitions**
- 2026-10-12 13:39 (new) → proposed (announced): novox/mesh-host#90's head announced
- 2026-10-12 13:42 proposed → checked (checked): the verdict names this commit
- 2026-10-12 13:42 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
- 2026-10-12 14:16 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 14:16 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 14:20 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: mesh-host)
mesh/repo-check: PASS — its merge-check.sh passed
Delivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
tier 0: mesh-host
ace: receives mesh-host
g14: receives mesh-host
novox: receives mesh-host
shanks: receives mesh-host
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 mesh-host is judged on its first machine
read by no module's build: cmd/mesh-host/probe_test.go, cmd/mesh-host/replay233_test.go
Run by the build seat on shanks as build-1791810799833572095 (builds --log build-1791810799833572095).
--- the gate: mesh-host
mesh-host: ok
long-running resources without health: 0
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 0 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T13:06:42Z 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 2026-10-12T13:06:42Z
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-host
notes:
- cmd/mesh-host/probe_test.go and cmd/mesh-host/replay233_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-host/cmd/mesh-bootstrap 1.015s
ok github.com/novox/mesh-host/cmd/mesh-host 5.545s
ok github.com/novox/mesh-host/examples 1.058s
ok github.com/novox/mesh-host/internal/accounts 1.480s
ok github.com/novox/mesh-host/internal/apply 3.310s
ok github.com/novox/mesh-host/internal/bootstrap 1.524s
ok github.com/novox/mesh-host/internal/bundle 1.052s
ok github.com/novox/mesh-host/internal/declaration 1.093s
ok github.com/novox/mesh-host/internal/firewall 1.859s
ok github.com/novox/mesh-host/internal/identity 1.094s
ok github.com/novox/mesh-host/internal/image 1.051s
ok github.com/novox/mesh-host/internal/inventory 1.052s
ok github.com/novox/mesh-host/internal/link 10.485s
ok github.com/novox/mesh-host/internal/liveness 2.115s
ok github.com/novox/mesh-host/internal/meshcli 1.582s
ok github.com/novox/mesh-host/internal/network 1.071s
ok github.com/novox/mesh-host/internal/outward 1.059s
ok github.com/novox/mesh-host/internal/profile 1.539s
ok github.com/novox/mesh-host/internal/reachable 1.049s
ok github.com/novox/mesh-host/internal/store 1.163s
ok github.com/novox/mesh-host/internal/system 1.054s
ok github.com/novox/mesh-host/internal/tunnel 1.065s
ok github.com/novox/mesh-host/internal/units 1.057s
ok github.com/novox/mesh-host/internal/upgrade 1.122s
ok github.com/novox/mesh-host/internal/witness 2.675s
ok github.com/novox/mesh-host/packaging 1.009s
? github.com/novox/mesh-host/rootsearch [no test files]
ok github.com/novox/mesh-host/validate 1.009s
**Merge check** at `6e44a27a`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds mesh-host → ace, g14, novox, shanks; no bus step
- tier 0: mesh-host
- ace: receives mesh-host
- g14: receives mesh-host
- novox: receives mesh-host
- shanks: receives mesh-host
- 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 mesh-host is judged on its first machine
- read by no module's build: cmd/mesh-host/probe_test.go, cmd/mesh-host/replay233_test.go
Run by the build seat on shanks as `build-1791810799833572095` (`builds --log build-1791810799833572095`).
```
--- the gate: mesh-host
mesh-host: ok
long-running resources without health: 0
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 0 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T13:06:42Z 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 2026-10-12T13:06:42Z
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-host
notes:
- cmd/mesh-host/probe_test.go and cmd/mesh-host/replay233_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-host/cmd/mesh-bootstrap 1.015s
ok github.com/novox/mesh-host/cmd/mesh-host 5.545s
ok github.com/novox/mesh-host/examples 1.058s
ok github.com/novox/mesh-host/internal/accounts 1.480s
ok github.com/novox/mesh-host/internal/apply 3.310s
ok github.com/novox/mesh-host/internal/bootstrap 1.524s
ok github.com/novox/mesh-host/internal/bundle 1.052s
ok github.com/novox/mesh-host/internal/declaration 1.093s
ok github.com/novox/mesh-host/internal/firewall 1.859s
ok github.com/novox/mesh-host/internal/identity 1.094s
ok github.com/novox/mesh-host/internal/image 1.051s
ok github.com/novox/mesh-host/internal/inventory 1.052s
ok github.com/novox/mesh-host/internal/link 10.485s
ok github.com/novox/mesh-host/internal/liveness 2.115s
ok github.com/novox/mesh-host/internal/meshcli 1.582s
ok github.com/novox/mesh-host/internal/network 1.071s
ok github.com/novox/mesh-host/internal/outward 1.059s
ok github.com/novox/mesh-host/internal/profile 1.539s
ok github.com/novox/mesh-host/internal/reachable 1.049s
ok github.com/novox/mesh-host/internal/store 1.163s
ok github.com/novox/mesh-host/internal/system 1.054s
ok github.com/novox/mesh-host/internal/tunnel 1.065s
ok github.com/novox/mesh-host/internal/units 1.057s
ok github.com/novox/mesh-host/internal/upgrade 1.122s
ok github.com/novox/mesh-host/internal/witness 2.675s
ok github.com/novox/mesh-host/packaging 1.009s
? github.com/novox/mesh-host/rootsearch [no test files]
ok github.com/novox/mesh-host/validate 1.009s
```
Independent review of the issue 233 delivery group (this and mesh-controller #263), at 6e44a27.
What I ran.go vet ./... is clean. go test -race ./... passes in every package. TestReplay233, laid over origin/main (dce0eef), fails there with "the declaration that restores the package manager's configuration was refused whole: … pacman exited 1: error: config file /etc/pacman.conf could not be read". All four checks are green at this head: merge-gate (4 of 4 compose), repo-check, delivery and delivery-group.
What holds.
A probe that fails on arch or alpine now gives a partial apply. Every package fails through PackageInstalled, which runs the same probe first, and does so with the probe's words. The report's Failed gains the package manager (pacman) under one stable key, and the controller's sameFailure counts by key, so that key works.
The issue 234 generation refusal still comes before the probe on both paths. Android still refuses whole.
A manual mesh-host apply returns applyErr or probe.partial() before RecordKnownGood, ClearAttempts and Retire, so it exits non-zero and records nothing.
Nothing that depends on a package is reported as fine. A failed package gates nothing, but that is the existing policy (issue 011, ADR 0136) and not new here. A unit whose binary was never installed is still started, fails the service read-back (the stayed-running check) and is reported failed. The retries are bounded by systemd's start limit and run once per reconcile, as for a package that fails for any other reason. The new harm is limited to packages that were not installed before. A package already installed keeps its files.
The node-engine's own upgrade is an archive resource, not a package, so it applies. Standing aside for the successor runs whatever the report says.
On the daemon path, known-good comes from the witness (proving.heard): any taken report about a declaration that was not refused proves the build, including a partial one. That matches ADR 0005, and it is exactly what keeps an upgrade landing on a broken-pacman node from being rolled back. I am noting it so nobody reads "never on a partial apply" as covering the daemon.
Blocking: the controller never hears a reconcile's outcome change, so S21 (#263) cannot count or clear.reconcileKept's report reaches the bus only when worthSaying(r) && watch.differs(r). adoptionFingerprint carries no Refused and no Failed, and worthSaying is false for a refusal. A reconcile that is still failing is therefore never said, and neither is a reconcile that turned clean. Two consequences follow:
The issue's own path. The restoring delivery runs the probe before the file is written, so its report is partial (every package failed). The next reconcile finds pacman answering and applies clean, but says nothing. The controller keeps "failed: the package manager (pacman)" until some later delivery. If S21 was raised, it stays raised over a healthy machine.
A failure that persists with no new delivery stays at Times = 1, so S21's "≥ 2 reports" is never reached. Only further pushes advance it.
The fix belongs here. A reconcile should be news when its outcome differs from the last one said: refused or not, the failed ids, clean or not. Either it is also said once per interval while it fails, so Times counts as issue 065 assumed, or #263 judges by time since the first report while the machine is heard. A test should defend "a failed reconcile followed by a clean one is said".
Should fix (not blocking): the wrong-machine guard is weaker. Arch's host on a machine with no pacman at all now applies files, units and accounts, and fails only the packages. The comment above the probe in runApply ("must say so once, at the start") no longer describes this. You could still refuse whole when the tool is absent (exec not found) and go partial only when it is present and errors, which was the 233 case. The decision's words allow either reading, so this needs a decision rather than me ruling on it.
Suggestion. After a partial apply whose rest wrote files, probe again. If the package manager now answers, apply the packages in the same apply. The 233 replay would then end clean in one apply rather than one partial apply and a silent clean reconcile.
Verdict: request changes at 6e44a27
Independent review of the issue 233 delivery group (this and mesh-controller #263), at 6e44a27.
**What I ran.** `go vet ./...` is clean. `go test -race ./...` passes in every package. TestReplay233, laid over origin/main (dce0eef), fails there with "the declaration that restores the package manager's configuration was refused whole: … pacman exited 1: error: config file /etc/pacman.conf could not be read". All four checks are green at this head: merge-gate (4 of 4 compose), repo-check, delivery and delivery-group.
**What holds.**
- A probe that fails on arch or alpine now gives a partial apply. Every package fails through `PackageInstalled`, which runs the same probe first, and does so with the probe's words. The report's Failed gains `the package manager (pacman)` under one stable key, and the controller's sameFailure counts by key, so that key works.
- The issue 234 generation refusal still comes before the probe on both paths. Android still refuses whole.
- A manual `mesh-host apply` returns `applyErr` or `probe.partial()` before RecordKnownGood, ClearAttempts and Retire, so it exits non-zero and records nothing.
- Nothing that depends on a package is reported as fine. A failed package gates nothing, but that is the existing policy (issue 011, ADR 0136) and not new here. A unit whose binary was never installed is still started, fails the service read-back (the stayed-running check) and is reported failed. The retries are bounded by systemd's start limit and run once per reconcile, as for a package that fails for any other reason. The new harm is limited to packages that were not installed before. A package already installed keeps its files.
- The node-engine's own upgrade is an archive resource, not a package, so it applies. Standing aside for the successor runs whatever the report says.
- On the daemon path, known-good comes from the witness (`proving.heard`): any taken report about a declaration that was not refused proves the build, including a partial one. That matches ADR 0005, and it is exactly what keeps an upgrade landing on a broken-pacman node from being rolled back. I am noting it so nobody reads "never on a partial apply" as covering the daemon.
**Blocking: the controller never hears a reconcile's outcome change, so S21 (#263) cannot count or clear.** `reconcileKept`'s report reaches the bus only when `worthSaying(r) && watch.differs(r)`. `adoptionFingerprint` carries no Refused and no Failed, and worthSaying is false for a refusal. A reconcile that is still failing is therefore never said, and neither is a reconcile that turned clean. Two consequences follow:
1. The issue's own path. The restoring delivery runs the probe before the file is written, so its report is partial (every package failed). The next reconcile finds pacman answering and applies clean, but says nothing. The controller keeps "failed: the package manager (pacman)" until some later delivery. If S21 was raised, it stays raised over a healthy machine.
2. A failure that persists with no new delivery stays at Times = 1, so S21's "≥ 2 reports" is never reached. Only further pushes advance it.
The fix belongs here. A reconcile should be news when its outcome differs from the last one said: refused or not, the failed ids, clean or not. Either it is also said once per interval while it fails, so Times counts as issue 065 assumed, or #263 judges by time since the first report while the machine is heard. A test should defend "a failed reconcile followed by a clean one is said".
**Should fix (not blocking): the wrong-machine guard is weaker.** Arch's host on a machine with no pacman at all now applies files, units and accounts, and fails only the packages. The comment above the probe in runApply ("must say so once, at the start") no longer describes this. You could still refuse whole when the tool is absent (exec not found) and go partial only when it is present and errors, which was the 233 case. The decision's words allow either reading, so this needs a decision rather than me ruling on it.
**Suggestion.** After a partial apply whose rest wrote files, probe again. If the package manager now answers, apply the packages in the same apply. The 233 replay would then end clean in one apply rather than one partial apply and a silent clean reconcile.
mesh/merge-gate pass: builds mesh-host → ace, g14, novox, shanks; no bus step; every machine composes with the change as it did without (4 of 4 compose)
mesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)
mesh/repo-check: PASS — its merge-check.sh passed
Delivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
tier 0: mesh-host
ace: receives mesh-host
g14: receives mesh-host
novox: receives mesh-host
shanks: receives mesh-host
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 mesh-host is judged on its first machine
read by no module's build: cmd/mesh-host/probe_test.go, cmd/mesh-host/replay233_test.go
Run by the build seat on g14 as build-1791812357081197944 (builds --log build-1791812357081197944).
--- the gate: mesh-host
mesh-host: ok
long-running resources without health: 0
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 0 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T13:33:41Z 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 2026-10-12T13:33:41Z
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-host
notes:
- cmd/mesh-host/probe_test.go and cmd/mesh-host/replay233_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-host/cmd/mesh-bootstrap 1.009s
ok github.com/novox/mesh-host/cmd/mesh-host 7.444s
ok github.com/novox/mesh-host/examples 1.021s
ok github.com/novox/mesh-host/internal/accounts 2.751s
ok github.com/novox/mesh-host/internal/apply 10.949s
ok github.com/novox/mesh-host/internal/bootstrap 1.583s
ok github.com/novox/mesh-host/internal/bundle 1.012s
ok github.com/novox/mesh-host/internal/declaration 1.054s
ok github.com/novox/mesh-host/internal/firewall 1.711s
ok github.com/novox/mesh-host/internal/identity 1.124s
ok github.com/novox/mesh-host/internal/image 1.015s
ok github.com/novox/mesh-host/internal/inventory 1.035s
ok github.com/novox/mesh-host/internal/link 10.554s
ok github.com/novox/mesh-host/internal/liveness 2.085s
ok github.com/novox/mesh-host/internal/meshcli 1.579s
ok github.com/novox/mesh-host/internal/network 1.047s
ok github.com/novox/mesh-host/internal/outward 1.034s
ok github.com/novox/mesh-host/internal/profile 1.535s
ok github.com/novox/mesh-host/internal/reachable 1.011s
ok github.com/novox/mesh-host/internal/store 1.540s
ok github.com/novox/mesh-host/internal/system 1.009s
ok github.com/novox/mesh-host/internal/tunnel 1.018s
ok github.com/novox/mesh-host/internal/units 1.011s
ok github.com/novox/mesh-host/internal/upgrade 1.460s
ok github.com/novox/mesh-host/internal/witness 8.770s
ok github.com/novox/mesh-host/packaging 1.012s
? github.com/novox/mesh-host/rootsearch [no test files]
ok github.com/novox/mesh-host/validate 1.007s
**Merge check** at `74ed7cbe`
- `mesh/merge-gate`: **PASS** — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)
- `mesh/repo-check`: **PASS** — its merge-check.sh passed
**Delivery plan** — builds mesh-host → ace, g14, novox, shanks; no bus step
- tier 0: mesh-host
- ace: receives mesh-host
- g14: receives mesh-host
- novox: receives mesh-host
- shanks: receives mesh-host
- 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 mesh-host is judged on its first machine
- read by no module's build: cmd/mesh-host/probe_test.go, cmd/mesh-host/replay233_test.go
Run by the build seat on g14 as `build-1791812357081197944` (`builds --log build-1791812357081197944`).
```
--- the gate: mesh-host
mesh-host: ok
long-running resources without health: 0
files asking for a setting without saying whether it is trusted: 0
1 manifest(s) checked. Judged against the seats this binary carries; a claim on one of the mesh's own seats is judged fully at registration, and a seat declared by a module not given here reads as unknown. Identities judged on a 6-character machine name, against the bounds of the providers given here
judging 0 module(s) changed against 4 machine(s), as the snapshot of 2026-10-12T13:33:41Z 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 2026-10-12T13:33:41Z
machines:
- a machine (f35): composes
- holds mesh-dns-resolver (idt): composes
- holds git, holds mesh-artifact-store, holds mesh-broker, holds mesh-catalog, holds mesh-delivery, holds mesh-dns-resolver, holds mesh-store, holds mesh-vault, holds npm-package-registry, the control node, the hub (ygcyr): composes
- a machine (yvbpsm): composes
a merge would rebuild 1 module(s) in 1 tier(s): mesh-host
notes:
- cmd/mesh-host/probe_test.go and cmd/mesh-host/replay233_test.go read by no module's build, and rebuild nothing (issue 280)
--- its merge-check.sh (go toolchain)
ok github.com/novox/mesh-host/cmd/mesh-bootstrap 1.009s
ok github.com/novox/mesh-host/cmd/mesh-host 7.444s
ok github.com/novox/mesh-host/examples 1.021s
ok github.com/novox/mesh-host/internal/accounts 2.751s
ok github.com/novox/mesh-host/internal/apply 10.949s
ok github.com/novox/mesh-host/internal/bootstrap 1.583s
ok github.com/novox/mesh-host/internal/bundle 1.012s
ok github.com/novox/mesh-host/internal/declaration 1.054s
ok github.com/novox/mesh-host/internal/firewall 1.711s
ok github.com/novox/mesh-host/internal/identity 1.124s
ok github.com/novox/mesh-host/internal/image 1.015s
ok github.com/novox/mesh-host/internal/inventory 1.035s
ok github.com/novox/mesh-host/internal/link 10.554s
ok github.com/novox/mesh-host/internal/liveness 2.085s
ok github.com/novox/mesh-host/internal/meshcli 1.579s
ok github.com/novox/mesh-host/internal/network 1.047s
ok github.com/novox/mesh-host/internal/outward 1.034s
ok github.com/novox/mesh-host/internal/profile 1.535s
ok github.com/novox/mesh-host/internal/reachable 1.011s
ok github.com/novox/mesh-host/internal/store 1.540s
ok github.com/novox/mesh-host/internal/system 1.009s
ok github.com/novox/mesh-host/internal/tunnel 1.018s
ok github.com/novox/mesh-host/internal/units 1.011s
ok github.com/novox/mesh-host/internal/upgrade 1.460s
ok github.com/novox/mesh-host/internal/witness 8.770s
ok github.com/novox/mesh-host/packaging 1.012s
? github.com/novox/mesh-host/rootsearch [no test files]
ok github.com/novox/mesh-host/validate 1.007s
```
Re-review against my review 63 (at 6e44a27), now at 74ed7cb, together with mesh-controller #263 at d4de454.
What I ran.go vet ./... is clean. go test -race -count=1 ./... passes in every package. TestReplay233, laid over origin/main, still fails there. All four checks are green at this head: merge-gate (4 of 4 compose), repo-check, delivery, and delivery-group ("composed together they pass", 13:50).
The blocking finding is fixed.adoptionFingerprint now carries applyState, which is one of three values:
adoptionWatch.news says a re-apply when its fingerprint differs and it is either about the machine, failing now, or clean after a failing one was last heard. said keeps faulty, and it is called from Heard for deliveries and re-applies alike.
Flooding. An identical failure is silent. The fingerprint compares equal, and TestAReapplyWhoseResultChangedIsSaidUnasked and TestARefusalIsSaidOnceAndNothingIsNotSaid prove it. A failure that changes is said at most once per re-apply (every 5 min).
No time-carrying words in the fingerprint. It holds only keys and declared ids: apply, keeping the declaration, the package manager (<tool>), and resource identities. It holds no error text and no refusal text, and the test feeds a refusal reworded "at another time". Unapplied is json:"-", so the wire format does not change.
The second pass. It runs at most once, as a straight-line call with no loop. It runs only when the probe failed, the first pass moved something (Changed, or Done.Changed on a partial) and a fresh probe answers. It runs over the state the first pass left, so files read unchanged and no restart fires a second time. Actions are guarded by their verify, run-once steps by their record and containers by their spec digest. A module gated in the first pass is retried, which the next re-apply would do anyway. The first pass's outcomes go into moved for the setuid search. The report's own words come from the second pass, which is the pass that is true. Known-good on the manual path is recorded only when the second pass is clean, which is a genuinely clean apply.
Missing tool. The refusal matches only exec.ErrNotFound, through errors.Is. That works because ExecRunner wraps non-exit errors with %w, and arch's and alpine's Confirm wrap with %w too. A pacman that is present but errors, as in the 233 case, still applies in part. TestAPackageManagerThatIsNotThereRefusesWhole proves the whole refusal and that nothing was touched.
Notes, none blocking.
Every failure now reaches the controller once. status's "stuck" line (Times ≥ 3) therefore no longer rises from re-applies. That was already so before this change, and S21 now covers the case.
The fingerprint changes on any shift in Unapplied. A resource that alternates between failing and applying would be said on each change, at most once per re-apply. That is bounded and acceptable.
Verdict: approve at 74ed7cb
Re-review against my review 63 (at 6e44a27), now at 74ed7cb, together with mesh-controller #263 at d4de454.
**What I ran.** `go vet ./...` is clean. `go test -race -count=1 ./...` passes in every package. TestReplay233, laid over origin/main, still fails there. All four checks are green at this head: merge-gate (4 of 4 compose), repo-check, delivery, and delivery-group ("composed together they pass", 13:50).
**The blocking finding is fixed.** `adoptionFingerprint` now carries `applyState`, which is one of three values:
- `apply refused`;
- `apply clean`;
- `apply failed <Failed keys> unapplied <declared ids>`.
`adoptionWatch.news` says a re-apply when its fingerprint differs and it is either about the machine, failing now, or clean after a failing one was last heard. `said` keeps `faulty`, and it is called from Heard for deliveries and re-applies alike.
**Flooding.** An identical failure is silent. The fingerprint compares equal, and TestAReapplyWhoseResultChangedIsSaidUnasked and TestARefusalIsSaidOnceAndNothingIsNotSaid prove it. A failure that changes is said at most once per re-apply (every 5 min).
**No time-carrying words in the fingerprint.** It holds only keys and declared ids: `apply`, `keeping the declaration`, `the package manager (<tool>)`, and resource identities. It holds no error text and no refusal text, and the test feeds a refusal reworded "at another time". `Unapplied` is `json:"-"`, so the wire format does not change.
**The second pass.** It runs at most once, as a straight-line call with no loop. It runs only when the probe failed, the first pass moved something (`Changed`, or `Done.Changed` on a partial) and a fresh probe answers. It runs over the state the first pass left, so files read unchanged and no restart fires a second time. Actions are guarded by their `verify`, run-once steps by their record and containers by their spec digest. A module gated in the first pass is retried, which the next re-apply would do anyway. The first pass's outcomes go into `moved` for the setuid search. The report's own words come from the second pass, which is the pass that is true. Known-good on the manual path is recorded only when the second pass is clean, which is a genuinely clean apply.
**Missing tool.** The refusal matches only `exec.ErrNotFound`, through `errors.Is`. That works because ExecRunner wraps non-exit errors with `%w`, and arch's and alpine's `Confirm` wrap with `%w` too. A pacman that is present but errors, as in the 233 case, still applies in part. TestAPackageManagerThatIsNotThereRefusesWhole proves the whole refusal and that nothing was touched.
**Notes, none blocking.**
- Every failure now reaches the controller once. `status`'s "stuck" line (Times ≥ 3) therefore no longer rises from re-applies. That was already so before this change, and S21 now covers the case.
- The fingerprint changes on any shift in `Unapplied`. A resource that alternates between failing and applying would be said on each change, at most once per re-apply. That is bounded and acceptable.
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.
Replaces #89/#261; same change, history without a built binary. This PR carries #89's commit unchanged (
cb0a07b).What
The pre-apply probe of the package manager (
sys.Confirm) no longer refuses the whole declaration on arch and alpine. When pacman or apk does not answer, the node-engine says so before applying, applies everything else (files, units, accounts, its own upgrade), and every package resource fails with the probe's words. The system'sPackageInstalledalready asks the same probe first. The report names the probe under one stable key,the package manager (pacman), besideapply. That lets the controller count it as one failure that goes on (the condition is in mesh-controller, same branch name).mesh-host applyby hand does the same and exits non-zero with "applied only in part" when no package was declared, and it never records known-good.The android host still refuses whole. Its probe asks the system itself, which nothing the mesh declares can remove, and nothing else guards against the wrong machine.
Issue 234's generation check still runs first. A declaration composed from an older view of the assignments is refused before anything is applied, as the decision of 2026-10-11 requires.
Why
On 2026-10-04 a declaration that left the package manager's module out removed
pacman.conf. From then on, every declaration was refused whole for 90 minutes, including the one that would have written the file back and the node-engine's own upgrade. A person on the machine was the only way back. The decision recorded on the issue (2026-10-11): a failed probe fails only what needs the tool.How it was tested
TestReplay233(cmd/mesh-host/replay233_test.go) is the replay. A pacman on PATH answers as the workstation's did. The test sends a declaration with the configuration file plus a package throughapplyAndKeep, the link's path. It asserts that nothing was refused whole, the file is written and recorded, and the report carries the probe's words and the package. It uses only what the node-engine had before the fix: it fails on main and passes here.mesh-host applyby hand. A further test checks that with no package declared the apply is still said to be partial. Another checks that android still refuses whole.sh merge-check.sh: green.dce0eef, #88 merged): no conflict, and the merged tree builds, vets, and passescmd/mesh-host,internal/applyandinternal/link. Trial merge of that result with #87's branch: no conflict.Replay: R233 (cmd/mesh-host
TestReplay233), to be registered in mesh-lab once this merges.Fixes issue 233
Merge check at
cb0a07bamesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
Run by the build seat on ace as
build-1791809446739400983(builds --log build-1791809446739400983).cb0a07ba7ato6e44a27a63Delivery
novox/mesh-host@74ed7cbe0ab4— delivered since 2026-10-12T14:20:23ZDelivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
Group
fix/233-a-failed-probe-fails-only-what-needs-the-tool-2, in order: novox/mesh-host@74ed7cbe0a → novox/mesh-controller@d4de4543ccMachines
Transitions
The commit's note under
refs/notes/mesh-plankeeps every transition:git log --notes=mesh-plan.Merge check at
6e44a27amesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
Run by the build seat on shanks as
build-1791810799833572095(builds --log build-1791810799833572095).Verdict: request changes at
6e44a27Independent review of the issue 233 delivery group (this and mesh-controller #263), at
6e44a27.What I ran.
go vet ./...is clean.go test -race ./...passes in every package. TestReplay233, laid over origin/main (dce0eef), fails there with "the declaration that restores the package manager's configuration was refused whole: … pacman exited 1: error: config file /etc/pacman.conf could not be read". All four checks are green at this head: merge-gate (4 of 4 compose), repo-check, delivery and delivery-group.What holds.
PackageInstalled, which runs the same probe first, and does so with the probe's words. The report's Failed gainsthe package manager (pacman)under one stable key, and the controller's sameFailure counts by key, so that key works.mesh-host applyreturnsapplyErrorprobe.partial()before RecordKnownGood, ClearAttempts and Retire, so it exits non-zero and records nothing.proving.heard): any taken report about a declaration that was not refused proves the build, including a partial one. That matches ADR 0005, and it is exactly what keeps an upgrade landing on a broken-pacman node from being rolled back. I am noting it so nobody reads "never on a partial apply" as covering the daemon.Blocking: the controller never hears a reconcile's outcome change, so S21 (#263) cannot count or clear.
reconcileKept's report reaches the bus only whenworthSaying(r) && watch.differs(r).adoptionFingerprintcarries no Refused and no Failed, and worthSaying is false for a refusal. A reconcile that is still failing is therefore never said, and neither is a reconcile that turned clean. Two consequences follow:The fix belongs here. A reconcile should be news when its outcome differs from the last one said: refused or not, the failed ids, clean or not. Either it is also said once per interval while it fails, so Times counts as issue 065 assumed, or #263 judges by time since the first report while the machine is heard. A test should defend "a failed reconcile followed by a clean one is said".
Should fix (not blocking): the wrong-machine guard is weaker. Arch's host on a machine with no pacman at all now applies files, units and accounts, and fails only the packages. The comment above the probe in runApply ("must say so once, at the start") no longer describes this. You could still refuse whole when the tool is absent (exec not found) and go partial only when it is present and errors, which was the 233 case. The decision's words allow either reading, so this needs a decision rather than me ruling on it.
Suggestion. After a partial apply whose rest wrote files, probe again. If the package manager now answers, apply the packages in the same apply. The 233 replay would then end clean in one apply rather than one partial apply and a silent clean reconcile.
Merge check at
74ed7cbemesh/merge-gate: PASS — every machine composes with the change as it did without (4 of 4 compose) (modules: mesh-host)mesh/repo-check: PASS — its merge-check.sh passedDelivery plan — builds mesh-host → ace, g14, novox, shanks; no bus step
Run by the build seat on g14 as
build-1791812357081197944(builds --log build-1791812357081197944).Verdict: approve at
74ed7cbRe-review against my review 63 (at
6e44a27), now at74ed7cb, together with mesh-controller #263 at d4de454.What I ran.
go vet ./...is clean.go test -race -count=1 ./...passes in every package. TestReplay233, laid over origin/main, still fails there. All four checks are green at this head: merge-gate (4 of 4 compose), repo-check, delivery, and delivery-group ("composed together they pass", 13:50).The blocking finding is fixed.
adoptionFingerprintnow carriesapplyState, which is one of three values:apply refused;apply clean;apply failed <Failed keys> unapplied <declared ids>.adoptionWatch.newssays a re-apply when its fingerprint differs and it is either about the machine, failing now, or clean after a failing one was last heard.saidkeepsfaulty, and it is called from Heard for deliveries and re-applies alike.Flooding. An identical failure is silent. The fingerprint compares equal, and TestAReapplyWhoseResultChangedIsSaidUnasked and TestARefusalIsSaidOnceAndNothingIsNotSaid prove it. A failure that changes is said at most once per re-apply (every 5 min).
No time-carrying words in the fingerprint. It holds only keys and declared ids:
apply,keeping the declaration,the package manager (<tool>), and resource identities. It holds no error text and no refusal text, and the test feeds a refusal reworded "at another time".Unappliedisjson:"-", so the wire format does not change.The second pass. It runs at most once, as a straight-line call with no loop. It runs only when the probe failed, the first pass moved something (
Changed, orDone.Changedon a partial) and a fresh probe answers. It runs over the state the first pass left, so files read unchanged and no restart fires a second time. Actions are guarded by theirverify, run-once steps by their record and containers by their spec digest. A module gated in the first pass is retried, which the next re-apply would do anyway. The first pass's outcomes go intomovedfor the setuid search. The report's own words come from the second pass, which is the pass that is true. Known-good on the manual path is recorded only when the second pass is clean, which is a genuinely clean apply.Missing tool. The refusal matches only
exec.ErrNotFound, througherrors.Is. That works because ExecRunner wraps non-exit errors with%w, and arch's and alpine'sConfirmwrap with%wtoo. A pacman that is present but errors, as in the 233 case, still applies in part. TestAPackageManagerThatIsNotThereRefusesWhole proves the whole refusal and that nothing was touched.Notes, none blocking.
status's "stuck" line (Times ≥ 3) therefore no longer rises from re-applies. That was already so before this change, and S21 now covers the case.Unapplied. A resource that alternates between failing and applying would be said on each change, at most once per re-apply. That is bounded and acceptable.