ADR 0283: a part waiting for the operator's secret or setting is said as waiting and passes the first-node gate (issue 386) #294

Merged
mesh-admin merged 2 commits from feat/386-a-wait-for-the-operator into main 2026-10-10 19:10:47 +00:00
Contributor

Issue 386 (a module waiting for an operator's secret fails its first-node gate and holds every later walk), with issue 379 behind it.

  • ADR 0283: a health tool answers healthy, unhealthy or waits; the node-engine states waiting; the controller checks the wait (an outside-issued own secret not given on that machine, or a declared setting), the first-node gate reads it as waits for a person (ADR 0254), and the operator gets module.<m>.<machine>.needs-operator naming the act. Own secret families (smb-password-*), never made by the mesh. mounts: username as the setting smb-users, password per source as smb-password-<source>.
  • To-be 48 §4 and §5 amended; to-be 52 amended; ADR 0263 rule 7 gets a mechanism-changed note; glossary: waiting, secret family.
  • Issue 401 opened for the plain outside-issued secret the mesh still makes.

Code follows on the same branch name in mesh-sdk, mesh-controller, mesh-host and mesh-catalog.

Issue 386 (a module waiting for an operator's secret fails its first-node gate and holds every later walk), with issue 379 behind it. - ADR 0283: a health tool answers healthy, unhealthy or waits; the node-engine states `waiting`; the controller checks the wait (an outside-issued own secret not given on that machine, or a declared setting), the first-node gate reads it as waits for a person (ADR 0254), and the operator gets `module.<m>.<machine>.needs-operator` naming the act. Own secret families (`smb-password-*`), never made by the mesh. mounts: username as the setting `smb-users`, password per source as `smb-password-<source>`. - To-be 48 §4 and §5 amended; to-be 52 amended; ADR 0263 rule 7 gets a mechanism-changed note; glossary: waiting, secret family. - Issue 401 opened for the plain outside-issued secret the mesh still makes. Code follows on the same branch name in mesh-sdk, mesh-controller, mesh-host and mesh-catalog.
mesh-admin added 1 commit 2026-10-10 19:05:38 +00:00
ADR 0283: a part waiting for the operator's secret or setting is said as waiting and passes the first-node gate (issue 386)
mesh/merge-gate pass: the change touches no module of the mesh's graph
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery-group group feat/386-a-wait-for-the-operator checking: 1 of 2 member(s) ready
mesh/delivery superseded: a newer head of the same pull request
242a6c548b
A module whose only unhealthy part waited for the operator's secret failed
its gate, held every later walk and told the operator there was nothing to
do. The record adds a waits answer and a waiting state, checked by the
controller, read by the gate as waits for a person, and said as a
needs-operator condition naming the act; and secret families so each SMB
source's password is one hidden line at the desk prompt.
Author
Contributor

Delivery novox/hq@2648d571b347 — delivered since 2026-10-10T19:10:56Z

Delivery plan — builds nothing: the change touches no module of the mesh's graph

Group feat/386-a-wait-for-the-operator, in order: novox/hq@2648d571b3 → novox/mesh-sdk@33917f91ac

Transitions

  • 2026-10-10 19:09 (new) → proposed (announced): novox/hq#294's head announced
  • 2026-10-10 19:10 proposed → checked (checked): the verdict names this commit
  • 2026-10-10 19:10 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
  • 2026-10-10 19:10 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
  • 2026-10-10 19:10 published → delivered (done): nothing for a walk to move

The commit's note under refs/notes/mesh-plan keeps every transition: git log --notes=mesh-plan.

<!-- mesh-delivery:view --> **Delivery** `novox/hq@2648d571b347` — **delivered** since 2026-10-10T19:10:56Z **Delivery plan** — builds nothing: the change touches no module of the mesh's graph **Group** `feat/386-a-wait-for-the-operator`, in order: novox/hq@2648d571b347 → novox/mesh-sdk@33917f91ac3c **Transitions** - 2026-10-10 19:09 (new) → proposed (announced): novox/hq#294's head announced - 2026-10-10 19:10 proposed → checked (checked): the verdict names this commit - 2026-10-10 19:10 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail - 2026-10-10 19:10 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move - 2026-10-10 19:10 published → delivered (done): nothing for a walk to move The commit's note under `refs/notes/mesh-plan` keeps every transition: `git log --notes=mesh-plan`.
mesh-admin left a comment
Author
Contributor

Independent review of 242a6c5. Verdict: REQUEST CHANGES. It is posted as a comment because the forge refuses REQUEST_CHANGES from a tool (ADR 0266). There is one blocking finding, which is quick to fix. merge-check.sh passes. The record format, the How-it-is-checked table, the public-repo rule, the issue named for the deferral (401, open in the tracker), the absence of any secret value, and the to-be frontmatter of 48 and 52 are all fine. The design holds up.

BLOCKING

  1. Pointer notes are missing on ADR 0240 and ADR 0254 (02-DECISIONS/README.md, 'A pointer back from what a record changes'). ADR 0283 changes a mechanism that each of those records states. (a) ADR 0240 rule 4 lists the states as healthy, unhealthy, starting, held and unknown, and says 'The gate needs no new rule'. Rule 2 says failure inside grace does not count and that at least 2 failing looks are needed. ADR 0283 decisions 1, 2 and 4 (lines 102-116) add a fifth state, a third tool answer, a state that one look sets inside grace, and a gate reading. (b) ADR 0254 decision 1 (line 93) says 'Today it has one source', and its narrowing excuses only the move that added the group. ADR 0283 adds a second source and, for that source, excuses any build (options 6/7, decision 4). Add a dated 'The mechanism changed — 2026-10-10, by ADR 0283' note to each, as was done for 0263. The departure from 0254's narrowing is argued well in 0283 itself; the note is what a reader who arrives from a citation needs.

NON-BLOCKING
2. 0283 line 156 says 'Rollout order:'. 'rollout' is retired in hq prose (glossary, walk entry). words.py does not check 02-DECISIONS, but a new record should use the glossary word: 'Order of delivery:'.
3. Decision 6 (around lines 124-128) says a member is 'the prefix followed by a name's characters' and is placed with * replaced by that suffix. State the allowed character set (for example [a-z0-9-], no '/' or '..'). Add 'module check / secret accept refuse a member name outside it' to row 6 of the check table, because the suffix becomes part of a file path on the node.
4. The glossary entry 'secret family' (glossary line 268) says 'one member per part its settings name'. Decision 6 does not tie members to settings: any member name is accepted. Either reword the glossary entry, or say in decision 6 whether a member that no setting names is refused.
5. To-be 48's code: list (line 4) does not include mesh-sdk, although decision 1 puts the answer's type in mesh-sdk go/health.
6. 0283 repeats the 0263 mechanism-changed note inside its own Consequences (line 163, 'by this record'). That is redundant with the note in 0263 and the decision text. Consider dropping it.

DESIGN (checklist item 8)

  • Excusing any build while a secret wait checks out is justified. A missing outside secret is owed by every build alike. The excuse is bounded by four conditions: the secret is declared, it is issued-by outside, the store holds no given value for it, and every other resource of the module is healthy. The cost (a fault in the waiting part shows only after the secret is given) is stated honestly.
  • The setting wait, checked by name only, is acceptably bounded. A setting such as smb-users can be set and still not cover the source, so a value check cannot be generic. A false wait surfaces as a needs-operator warning in front of the operator. Not escalating it by age means a module that wrongly says it waits can pass gates for a long time. Consider recording that limit as a known cost (or opening an issue for it).
  • The secret family rule is coherent with ADR 0228 and ADR 0277. Outside is required, a family member is never made, and an absent file means not given. ADR 0277's refusal of secret-ask for 'one the mesh may make' does not catch a member. Plain outside secrets are left to issue 401.
Independent review of 242a6c5. Verdict: REQUEST CHANGES. It is posted as a comment because the forge refuses REQUEST_CHANGES from a tool (ADR 0266). There is one blocking finding, which is quick to fix. merge-check.sh passes. The record format, the How-it-is-checked table, the public-repo rule, the issue named for the deferral (401, open in the tracker), the absence of any secret value, and the to-be frontmatter of 48 and 52 are all fine. The design holds up. BLOCKING 1. Pointer notes are missing on ADR 0240 and ADR 0254 (02-DECISIONS/README.md, 'A pointer back from what a record changes'). ADR 0283 changes a mechanism that each of those records states. (a) ADR 0240 rule 4 lists the states as healthy, unhealthy, starting, held and unknown, and says 'The gate needs no new rule'. Rule 2 says failure inside grace does not count and that at least 2 failing looks are needed. ADR 0283 decisions 1, 2 and 4 (lines 102-116) add a fifth state, a third tool answer, a state that one look sets inside grace, and a gate reading. (b) ADR 0254 decision 1 (line 93) says 'Today it has one source', and its narrowing excuses only the move that added the group. ADR 0283 adds a second source and, for that source, excuses any build (options 6/7, decision 4). Add a dated 'The mechanism changed — 2026-10-10, by ADR 0283' note to each, as was done for 0263. The departure from 0254's narrowing is argued well in 0283 itself; the note is what a reader who arrives from a citation needs. NON-BLOCKING 2. 0283 line 156 says 'Rollout order:'. 'rollout' is retired in hq prose (glossary, walk entry). words.py does not check 02-DECISIONS, but a new record should use the glossary word: 'Order of delivery:'. 3. Decision 6 (around lines 124-128) says a member is 'the prefix followed by a name's characters' and is placed with * replaced by that suffix. State the allowed character set (for example [a-z0-9-], no '/' or '..'). Add 'module check / secret accept refuse a member name outside it' to row 6 of the check table, because the suffix becomes part of a file path on the node. 4. The glossary entry 'secret family' (glossary line 268) says 'one member per part its settings name'. Decision 6 does not tie members to settings: any member name is accepted. Either reword the glossary entry, or say in decision 6 whether a member that no setting names is refused. 5. To-be 48's code: list (line 4) does not include mesh-sdk, although decision 1 puts the answer's type in mesh-sdk go/health. 6. 0283 repeats the 0263 mechanism-changed note inside its own Consequences (line 163, 'by this record'). That is redundant with the note in 0263 and the decision text. Consider dropping it. DESIGN (checklist item 8) - Excusing any build while a secret wait checks out is justified. A missing outside secret is owed by every build alike. The excuse is bounded by four conditions: the secret is declared, it is issued-by outside, the store holds no given value for it, and every other resource of the module is healthy. The cost (a fault in the waiting part shows only after the secret is given) is stated honestly. - The setting wait, checked by name only, is acceptably bounded. A setting such as smb-users can be set and still not cover the source, so a value check cannot be generic. A false wait surfaces as a needs-operator warning in front of the operator. Not escalating it by age means a module that wrongly says it waits can pass gates for a long time. Consider recording that limit as a known cost (or opening an issue for it). - The secret family rule is coherent with ADR 0228 and ADR 0277. Outside is required, a family member is never made, and an absent file means not given. ADR 0277's refusal of secret-ask for 'one the mesh may make' does not catch a member. Plain outside secrets are left to issue 401.
jschoubben added 1 commit 2026-10-10 19:08:42 +00:00
Point ADRs 0240 and 0254 at ADR 0283, bound a family member's name, and say the cost of a setting wait
mesh/merge-gate pass: the change touches no module of the mesh's graph
mesh/repo-check pass: its merge-check.sh passed
mesh/delivery delivered
mesh/delivery-group group feat/386-a-wait-for-the-operator merged: every member is merged
2648d571b3
From the review of hq#294.
mesh-admin left a comment
Author
Contributor

Independent re-review of 2648d571. Verdict: APPROVE. It is posted as a comment because the forge refuses APPROVED from a tool (ADR 0266). merge-check.sh passes (exit 0). All six findings from the review of 242a6c5 are addressed:

  • ADR 0240 (after rule 4) and ADR 0254 (decision 1) now carry correct 'The mechanism changed — 2026-10-10, by ADR 0283' notes. They leave the decisions standing, and the 0254 note keeps its narrowing for the login.
  • 'Rollout order' now reads 'Order of delivery'.
  • A member's name is bounded to [a-z0-9][a-z0-9-]*, and row 6 of the check table tests that '/', '.' and upper-case letters are refused.
  • The cost of the setting wait, which is never escalated, is stated.
  • The glossary entry 'secret family' is aligned with decision 6.
  • mesh-sdk is in the code: list of to-be 48.
  • The duplicate 0263 note is gone from 0283.

Non-blocking nits, which need not hold the merge: (1) The 0240 note says 'The four states', but rule 4 lists five (healthy, unhealthy, starting, held, unknown); 'The states' would be accurate. (2) In 0283 decision 6, the sentence beginning 'A member is the prefix followed by a name's characters' has become a run-on, with 'any other name is refused' in the middle; splitting it would read better.

Independent re-review of 2648d571. Verdict: APPROVE. It is posted as a comment because the forge refuses APPROVED from a tool (ADR 0266). merge-check.sh passes (exit 0). All six findings from the review of 242a6c5 are addressed: - ADR 0240 (after rule 4) and ADR 0254 (decision 1) now carry correct 'The mechanism changed — 2026-10-10, by ADR 0283' notes. They leave the decisions standing, and the 0254 note keeps its narrowing for the login. - 'Rollout order' now reads 'Order of delivery'. - A member's name is bounded to [a-z0-9][a-z0-9-]*, and row 6 of the check table tests that '/', '.' and upper-case letters are refused. - The cost of the setting wait, which is never escalated, is stated. - The glossary entry 'secret family' is aligned with decision 6. - mesh-sdk is in the code: list of to-be 48. - The duplicate 0263 note is gone from 0283. Non-blocking nits, which need not hold the merge: (1) The 0240 note says 'The four states', but rule 4 lists five (healthy, unhealthy, starting, held, unknown); 'The states' would be accurate. (2) In 0283 decision 6, the sentence beginning 'A member is the prefix followed by a name's characters' has become a run-on, with 'any other name is refused' in the middle; splitting it would read better.
mesh-admin merged commit 536b193c7b into main 2026-10-10 19:10:47 +00:00
mesh-admin deleted branch feat/386-a-wait-for-the-operator 2026-10-10 19:10:47 +00:00
Sign in to join this conversation.
No Reviewers
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: novox/hq#294