Admin "data was restored from a backup" switch for connected devices (data epoch) #904

Closed
opened 2026-10-15 02:18:31 +00:00 by Cordy · 11 comments
Owner

Follow-up to #897 (research comment 9092, spec §5.2a and §11).

After a storage restore, folders return with their old .cairn-folder-<id> markers and older content, so limited devices resume without a question and their sync apps may treat newer files on the device as out of date (Nextcloud's known issue; their fix is data-fingerprint). v0.6.341 documents it: admins should pause or check devices bound under restored paths.

Idea: an admin action "Data was restored from a backup" that bumps an instance data epoch carried in the marker name (or a sidecar), so every bound device latches as Different until its owner confirms ("Use this folder"). Research first (Nextcloud data-fingerprint behaviour and client reactions, oCIS), then mockup, then decisions.

Follow-up to #897 (research comment 9092, spec §5.2a and §11). After a storage restore, folders return with their old `.cairn-folder-<id>` markers and older content, so limited devices resume without a question and their sync apps may treat newer files on the device as out of date (Nextcloud's known issue; their fix is `data-fingerprint`). v0.6.341 documents it: admins should pause or check devices bound under restored paths. Idea: an admin action "Data was restored from a backup" that bumps an instance data epoch carried in the marker name (or a sidecar), so every bound device latches as Different until its owner confirms ("Use this folder"). Research first (Nextcloud data-fingerprint behaviour and client reactions, oCIS), then mockup, then decisions.
Author
Owner

Research (2026-10-15)

Full file: cairn-work/scratch/b7e/batch907/904-research.md.

Core finding: a storage restore rolls back the instance state as well as the files, because state lives with the data (#138). After a restore:

  • devices disconnected after the backup come back to life, and devices created after it get 401;
  • share links revoked after the backup come back, and newer links stop working;
  • legal holds placed after the backup are lost;
  • files deleted after the backup come back, including erasure requests (an nDSG/GDPR problem: the deletions must be done again).

No server signal can stop a third-party two-way sync app from treating the older server copy as truth. rclone bisync's documented rule, for example, lets an older unchanged server file overwrite a newer local one. The only thing that works for every WebDAV client is refusing service until a person decides, which #897's pause already does.

Market:

  • Nextcloud data-fingerprint: only Nextcloud/ownCloud desktop clients read it; they then keep local files as conflicts. It is a manual step people forget; ownCloud had a data-loss report, and the all-in-one installer had to automate it.
  • oCIS/OpenCloud: their docs are silent on it.
  • Dropbox, OneDrive, Google: restores are forward changes, so the problem doesn't arise.
  • Syncthing and Microsoft Sync Framework: require a new ID after a restore.

Bug found: after a restore of a backup taken while Cairn was stopped, the change feed reuses its old epoch with a lower sequence, so clients can silently miss changes (internal/changes/changes.go:179).

Detection: only partial is possible. It works on S3 if the pod's local volume survives (a local "witness" counter). It does not work for the documented recovery drill (new pod plus restored bucket), a whole-root POSIX restore or a partial restore.

Recommendation:

  1. A "Storage was restored from a backup" action in the admin UI, plus a cairnd CLI step that can run before first start (devices sync within seconds, so a button after boot is too late). The action:
    • creates a new data epoch and an audit entry;
    • pauses every device with a new state restored (the owner resumes it, with advice per client type);
    • starts a new change-feed epoch and exposes the epoch on the WebDAV root and the feed;
    • shows the admin a checklist: share links, revived devices, holds, erasures, peering.
  2. Auto-detection as a safety net only: it pauses devices and asks the admin, never declares a restore itself.
  3. Not in ETags.

Decisions (my recommendation in brackets):

  1. Manual, automatic or both [both, manual first].
  2. Pause scope [all devices by default, the admin can narrow it].
  3. UI, CLI or both [both; the CLI is the documented restore step].
  4. Share links [optional "pause public links until reviewed", off by default].
  5. Client signal [a Cairn property and header; oc:data-fingerprint only if Nextcloud desktop support is planned].
  6. Ship the change-feed fix now, on its own [yes].
  7. New restored state [yes].
  8. Who resumes [the owner, plus an admin bulk resume with a warning].

Mockup needed:

  • the admin declaration dialog, the checklist card and the auto-detect banner;
  • the restored row, bell item and Resume dialog with per-client advice;
  • the user banner.

Size: about 2 weeks, splittable:

  • the change-feed fix: under 1 day;
  • the declaration and CLI: 3-4 days;
  • the UI: about 3 days;
  • S3 auto-detection: 1-2 days.
## Research (2026-10-15) Full file: `cairn-work/scratch/b7e/batch907/904-research.md`. **Core finding:** a storage restore rolls back the instance state as well as the files, because state lives with the data (#138). After a restore: - devices disconnected after the backup come back to life, and devices created after it get 401; - share links revoked after the backup come back, and newer links stop working; - legal holds placed after the backup are lost; - files deleted after the backup come back, including erasure requests (an nDSG/GDPR problem: the deletions must be done again). No server signal can stop a third-party two-way sync app from treating the older server copy as truth. rclone bisync's documented rule, for example, lets an older unchanged server file overwrite a newer local one. The only thing that works for every WebDAV client is refusing service until a person decides, which #897's pause already does. **Market:** - **Nextcloud `data-fingerprint`:** only Nextcloud/ownCloud desktop clients read it; they then keep local files as conflicts. It is a manual step people forget; ownCloud had a data-loss report, and the all-in-one installer had to automate it. - **oCIS/OpenCloud:** their docs are silent on it. - **Dropbox, OneDrive, Google:** restores are forward changes, so the problem doesn't arise. - **Syncthing and Microsoft Sync Framework:** require a new ID after a restore. **Bug found:** after a restore of a backup taken while Cairn was stopped, the change feed reuses its old epoch with a lower sequence, so clients can silently miss changes (`internal/changes/changes.go:179`). **Detection:** only partial is possible. It works on S3 if the pod's local volume survives (a local "witness" counter). It does not work for the documented recovery drill (new pod plus restored bucket), a whole-root POSIX restore or a partial restore. **Recommendation:** 1. **A "Storage was restored from a backup" action** in the admin UI, plus a `cairnd` CLI step that can run before first start (devices sync within seconds, so a button after boot is too late). The action: - creates a new data epoch and an audit entry; - pauses every device with a new state `restored` (the owner resumes it, with advice per client type); - starts a new change-feed epoch and exposes the epoch on the WebDAV root and the feed; - shows the admin a checklist: share links, revived devices, holds, erasures, peering. 2. **Auto-detection as a safety net only:** it pauses devices and asks the admin, never declares a restore itself. 3. **Not in ETags.** **Decisions** (my recommendation in brackets): 1. Manual, automatic or both [both, manual first]. 2. Pause scope [all devices by default, the admin can narrow it]. 3. UI, CLI or both [both; the CLI is the documented restore step]. 4. Share links [optional "pause public links until reviewed", off by default]. 5. Client signal [a Cairn property and header; `oc:data-fingerprint` only if Nextcloud desktop support is planned]. 6. Ship the change-feed fix now, on its own [yes]. 7. New `restored` state [yes]. 8. Who resumes [the owner, plus an admin bulk resume with a warning]. **Mockup needed:** - the admin declaration dialog, the checklist card and the auto-detect banner; - the `restored` row, bell item and Resume dialog with per-client advice; - the user banner. **Size:** about 2 weeks, splittable: - the change-feed fix: under 1 day; - the declaration and CLI: 3-4 days; - the UI: about 3 days; - S3 auto-detection: 1-2 days.
Author
Owner

Decisions (Nikola, 2026-10-15): all recommendations accepted (comment 9156): both manual and auto-detect (manual first; auto only pauses and asks), pause all devices by default (admin can narrow), UI + CLI, optional "pause public links" (off), Cairn property/header as client signal, new restored state, owner resumes + admin bulk resume. The change-feed epoch fix ships now in batch A; the rest after its mockup is approved.

**Decisions (Nikola, 2026-10-15): all recommendations accepted** (comment 9156): both manual and auto-detect (manual first; auto only pauses and asks), pause all devices by default (admin can narrow), UI + CLI, optional "pause public links" (off), Cairn property/header as client signal, new `restored` state, owner resumes + admin bulk resume. The change-feed epoch fix ships now in batch A; the rest after its mockup is approved.
Author
Owner

Mockup answers (Nikola, 2026-10-15) and research on Q1

  • Q2: "Not a restore" resumes all devices the guess paused, after one confirm. Accepted.
  • Q3: bulk resume preselects "Only devices nobody resumed for 7 days". Accepted.
  • Q1 (how long the user banner stays): Nikola asked for research first. File: cairn-work/scratch/b7e/batch907/904-q1-banner-research.md.

Q1 finding: institutions would reject both options.

  • 14 days is too short. People on leave miss it: Art. 329c OR means a two-week block is common, and three to four weeks or parental leave happens. Their pre-leave changes may be exactly the lost ones.
  • "Until the checklist is closed" is the wrong trigger. That is admin work and has nothing to do with whether staff have been told.
  • A lost change can be a lost official record. GEVER traceability and cantonal file rules apply, and only the author can re-file it.
  • Data-loss incidents need proof of who was told when. This follows from nDSG Art. 24 and the FDPIC guide, GDPR Art. 33/34 and EDPB 01/2021, ISO 27002 5.24-5.27, BSI DER.2.1.A4 and NIST 800-61r3.
  • GitLab and Confluence: their banners have admin-set end dates.
  • WCAG 2.2.1: an expiring message is fine only if the information stays available elsewhere.

Code finding:

  • The bell can't be the record: items are pruned after 30 days and capped at 20 per person.
  • A per-person list of affected files is unreliable: on POSIX the audit log is restored too.

Recommendation:

  1. The banner stays until each person acknowledges it. "Got it" is stored per person on the server; until then the banner comes back at every sign-in.
  2. End: the notice ends when an admin presses "End the notice" or after 90 days (adjustable 14-365, or "until I end it"). This is separate from closing the checklist.
  3. Wording: the text names the gap as a date range ("between {backup} and {declared}").
  4. Admin message: an optional one-line message (contact or ticket).
  5. Permanent history: "What this means" lists every past declaration.
  6. Proof: "Seen by N of M" with a CSV export for admins.
  7. E-mail: optional "Also e-mail everyone", once #896 exists, off by default.
  8. Later: a per-person list of changed files, only where the audit data survived.

Decisions (my recommendation in brackets):

  1. End rule [per-person acknowledgement plus admin end or 90 days].
  2. Default period [90 days].
  3. Admin message [one plain-text line].
  4. Seen-by [counts plus admin-only export].
  5. E-mail [opt-in].
  6. Per-person file list [later].
  7. Notice on public link pages [no].

Mockup changes if accepted:

  • Frame G: "Got it", the date range, the admin line.
  • Frame B: message field, active period, e-mail option.
  • Frame C: "Seen by", export, "End the notice".
  • New: a list of past restore notices.

Size: about 2 days on top of the #904 core.

## Mockup answers (Nikola, 2026-10-15) and research on Q1 - **Q2:** "Not a restore" resumes all devices the guess paused, after one confirm. Accepted. - **Q3:** bulk resume preselects "Only devices nobody resumed for 7 days". Accepted. - **Q1 (how long the user banner stays):** Nikola asked for research first. File: `cairn-work/scratch/b7e/batch907/904-q1-banner-research.md`. **Q1 finding: institutions would reject both options.** - **14 days is too short.** People on leave miss it: Art. 329c OR means a two-week block is common, and three to four weeks or parental leave happens. Their pre-leave changes may be exactly the lost ones. - **"Until the checklist is closed" is the wrong trigger.** That is admin work and has nothing to do with whether staff have been told. - **A lost change can be a lost official record.** GEVER traceability and cantonal file rules apply, and only the author can re-file it. - **Data-loss incidents need proof of who was told when.** This follows from nDSG Art. 24 and the FDPIC guide, GDPR Art. 33/34 and EDPB 01/2021, ISO 27002 5.24-5.27, BSI DER.2.1.A4 and NIST 800-61r3. - **GitLab and Confluence:** their banners have admin-set end dates. - **WCAG 2.2.1:** an expiring message is fine only if the information stays available elsewhere. **Code finding:** - **The bell can't be the record:** items are pruned after 30 days and capped at 20 per person. - **A per-person list of affected files is unreliable:** on POSIX the audit log is restored too. **Recommendation:** 1. **The banner stays until each person acknowledges it.** "Got it" is stored per person on the server; until then the banner comes back at every sign-in. 2. **End:** the notice ends when an admin presses "End the notice" or after 90 days (adjustable 14-365, or "until I end it"). This is separate from closing the checklist. 3. **Wording:** the text names the gap as a date range ("between {backup} and {declared}"). 4. **Admin message:** an optional one-line message (contact or ticket). 5. **Permanent history:** "What this means" lists every past declaration. 6. **Proof:** "Seen by N of M" with a CSV export for admins. 7. **E-mail:** optional "Also e-mail everyone", once #896 exists, off by default. 8. **Later:** a per-person list of changed files, only where the audit data survived. **Decisions** (my recommendation in brackets): 1. End rule [per-person acknowledgement plus admin end or 90 days]. 2. Default period [90 days]. 3. Admin message [one plain-text line]. 4. Seen-by [counts plus admin-only export]. 5. E-mail [opt-in]. 6. Per-person file list [later]. 7. Notice on public link pages [no]. **Mockup changes if accepted:** - Frame G: "Got it", the date range, the admin line. - Frame B: message field, active period, e-mail option. - Frame C: "Seen by", export, "End the notice". - New: a list of past restore notices. **Size:** about 2 days on top of the #904 core.
Author
Owner

Decisions (Nikola, 2026-10-15) on comment 9179

All accepted as recommended.

  1. End rule: each person acknowledges with "Got it" (stored on the server). Until then the banner returns at every sign-in. The notice ends when an admin presses "End the notice" or after the active period.
  2. Active period: 90 days by default, adjustable from 14 to 365 days or "until I end it".
  3. Banner text: names the gap as a date range. The admin can add an optional one-line plain-text message.
  4. Seen by N of M: shown to admins, with a CSV export for admins only.
  5. E-mail: opt-in "Also e-mail everyone", only once #896 exists. Noted on #896.
  6. Per-person list of changed files: later, and only where the audit data survived.
  7. Public link pages: no notice.
  8. History: a permanent list of past restore notices under "What this means".

The mockup is being updated now: frames B, C and G, plus a new history page.

## Decisions (Nikola, 2026-10-15) on comment 9179 All accepted as recommended. 1. **End rule:** each person acknowledges with "Got it" (stored on the server). Until then the banner returns at every sign-in. The notice ends when an admin presses "End the notice" or after the active period. 2. **Active period:** 90 days by default, adjustable from 14 to 365 days or "until I end it". 3. **Banner text:** names the gap as a date range. The admin can add an optional one-line plain-text message. 4. **Seen by N of M:** shown to admins, with a CSV export for admins only. 5. **E-mail:** opt-in "Also e-mail everyone", only once #896 exists. Noted on #896. 6. **Per-person list of changed files:** later, and only where the audit data survived. 7. **Public link pages:** no notice. 8. **History:** a permanent list of past restore notices under "What this means". The mockup is being updated now: frames B, C and G, plus a new history page.
Author
Owner

Mockup approved (Nikola, 2026-10-15)

Nikola approved 904-storage-restored-mockup.html with the 2026-10-15 update:

  • Frame B: message line, how long the notice stays, e-mail option (greyed out until #896).
  • Frame C: "Seen by N of M", CSV export, "End the notice".
  • Frame G: "Got it", the date range, the admin's line.
  • Frame I (new): the "Restore notices" page.

The rest of #904 builds in batch B, from this mockup.

## Mockup approved (Nikola, 2026-10-15) Nikola approved `904-storage-restored-mockup.html` with the 2026-10-15 update: - Frame B: message line, how long the notice stays, e-mail option (greyed out until #896). - Frame C: "Seen by N of M", CSV export, "End the notice". - Frame G: "Got it", the date range, the admin's line. - Frame I (new): the "Restore notices" page. The rest of #904 builds in batch B, from this mockup.
Author
Owner

Change-feed part shipped in v0.6.343 (PR #910)

  • Cursors are hash-chained. A cursor from a history that a restore from an older backup rewrote now expires, instead of silently skipping changes.
  • First start after the upgrade: a fresh epoch, so every client re-scans once.

Live check (local instance), PASS:

  • After restoring the journal from a cold backup, a newer cursor answers expired.
  • The cursor taken at the backup point stays valid.
  • A cursor kept over a clean restart stays valid.

This issue stays open for batch B: the restore declaration, the paused devices and the notice. The mockup was approved on 2026-10-15.

## Change-feed part shipped in v0.6.343 (PR #910) - **Cursors are hash-chained.** A cursor from a history that a restore from an older backup rewrote now expires, instead of silently skipping changes. - **First start after the upgrade:** a fresh epoch, so every client re-scans once. **Live check (local instance), PASS:** - After restoring the journal from a cold backup, a newer cursor answers expired. - The cursor taken at the backup point stays valid. - A cursor kept over a clean restart stays valid. This issue stays open for batch B: the restore declaration, the paused devices and the notice. The mockup was approved on 2026-10-15.
Author
Owner

Decisions (Nikola, 2026-10-27), from the R5 spec's open questions and the Q1 research (scratch/b7e/backlog/research-904-q1-basic-sync.md):

  • Q1, sync apps using the account password: paused like devices. They get 403 device-paused and appear on Connected devices as paused entries, named after the app Cairn recorded (rclone, Finder and so on). The person resumes them there with the per-app advice, and the admin's bulk resume includes them. This needs a mockup addition before the UI is built.
    • Config-file accounts get a Resume line on their Account page.
    • Accounts first seen after the declaration are paused too.
  • Q2: a "Resume public links" button on the checklist's Share links row, audited.
  • Q3: cairnd restored takes --message and --notice-days (defaults: no message, 90 days).
  • Q12: checklist ticks, closing the checklist and resuming links are each audited.
  • Other open questions: the spec's interim answers are adopted (spec section 10).
  • New research, requested by Nikola: can local-account passwords survive a backup restore, and what do best practice and standards say? Reported back before the checklist wording is decided.
**Decisions (Nikola, 2026-10-27), from the R5 spec's open questions and the Q1 research** (`scratch/b7e/backlog/research-904-q1-basic-sync.md`): - **Q1, sync apps using the account password:** paused like devices. They get 403 `device-paused` and appear on Connected devices as paused entries, named after the app Cairn recorded (rclone, Finder and so on). The person resumes them there with the per-app advice, and the admin's bulk resume includes them. This needs a mockup addition before the UI is built. - Config-file accounts get a Resume line on their Account page. - Accounts first seen after the declaration are paused too. - **Q2:** a "Resume public links" button on the checklist's Share links row, audited. - **Q3:** `cairnd restored` takes `--message` and `--notice-days` (defaults: no message, 90 days). - **Q12:** checklist ticks, closing the checklist and resuming links are each audited. - **Other open questions:** the spec's interim answers are adopted (spec section 10). - **New research, requested by Nikola:** can local-account passwords survive a backup restore, and what do best practice and standards say? Reported back before the checklist wording is decided.
Author
Owner

Decisions (Nikola, 2026-10-27), after the credentials-restore research (scratch/b7e/backlog/research-904-credentials-restore.md):

  • Local-account credentials and a restore: a targeted checklist.
    • At a declaration, Cairn uses the audit log, where it survived, to list local accounts changed after the backup: password changed, disabled, removed, demoted.
    • A new Accounts checklist row proposes "Disable again", "Remove administrator again" and "Set a one-time password". Nothing is re-applied automatically.
    • Where the audit log went back with the restore, one fallback sentence replaces the list.
  • One-time admin resets everywhere: any admin reset of a local account sets a password that must be changed at the next sign-in (OWASP ASVS 6.4.1).
  • "This restore follows a security incident" box: in R5. It requires a new password from every local account, and devices stay paused until resumed.
  • Password policy, pre-existing:
    • New local passwords need at least 15 characters; existing ones keep working.
    • A self-service password change gets its own password-changed audit entry.

These extend the #904 batch. The mockup gets additions (declaration form, checklist Accounts row, the set-a-new-password page) for approval before the UI is built. No standard or product keeps local passwords outside the restored data; the research has the sources.

**Decisions (Nikola, 2026-10-27), after the credentials-restore research** (`scratch/b7e/backlog/research-904-credentials-restore.md`): - **Local-account credentials and a restore: a targeted checklist.** - At a declaration, Cairn uses the audit log, where it survived, to list local accounts changed after the backup: password changed, disabled, removed, demoted. - A new **Accounts** checklist row proposes "Disable again", "Remove administrator again" and "Set a one-time password". Nothing is re-applied automatically. - Where the audit log went back with the restore, one fallback sentence replaces the list. - **One-time admin resets everywhere:** any admin reset of a local account sets a password that must be changed at the next sign-in (OWASP ASVS 6.4.1). - **"This restore follows a security incident" box:** in R5. It requires a new password from every local account, and devices stay paused until resumed. - **Password policy, pre-existing:** - New local passwords need at least 15 characters; existing ones keep working. - A self-service password change gets its own `password-changed` audit entry. These extend the #904 batch. The mockup gets additions (declaration form, checklist Accounts row, the set-a-new-password page) for approval before the UI is built. No standard or product keeps local passwords outside the restored data; the research has the sources.
Author
Owner

Decisions (Nikola, 2026-10-27), round 3:

  • Mockup: 904-storage-restored-mockup-v2.html is approved in full.
    • New frames: K/K2 "Choose a new password", L/L2 one-time reset, B2, C2, F2, J/J2.
    • Changed frames: CLI, B, C, D, E, G, H, I.
  • Q17: with the incident box ticked, the restored password signs in once, and a new password must be chosen right away.
  • Q21: an unused one-time password expires after 7 days.
  • Q24: after an admin reset, the person gets a bell item at the next sign-in, and an e-mail once #896 lands.
  • Interim answers adopted:
    • Q18: the CLI gets --security-incident.
    • Q19: a new account's first password is not one-time.
    • Q20: "Disable again" for accounts removed after the backup.
    • Q22: the incident box does not force the scope.
    • Q23: no extra Frame I sentence.
    • Q25: one paused entry per app.

Release plan:

  • v0.6.362 = #900 (in CI now).
  • v0.6.363 = the account policy: one-time admin resets, at least 15 characters for new local passwords, a password-changed audit entry, frames K and L.
  • v0.6.364 = this issue.
**Decisions (Nikola, 2026-10-27), round 3:** - **Mockup:** `904-storage-restored-mockup-v2.html` is approved in full. - New frames: K/K2 "Choose a new password", L/L2 one-time reset, B2, C2, F2, J/J2. - Changed frames: CLI, B, C, D, E, G, H, I. - **Q17:** with the incident box ticked, the restored password signs in once, and a new password must be chosen right away. - **Q21:** an unused one-time password expires after 7 days. - **Q24:** after an admin reset, the person gets a bell item at the next sign-in, and an e-mail once #896 lands. - **Interim answers adopted:** - Q18: the CLI gets `--security-incident`. - Q19: a new account's first password is not one-time. - Q20: "Disable again" for accounts removed after the backup. - Q22: the incident box does not force the scope. - Q23: no extra Frame I sentence. - Q25: one paused entry per app. **Release plan:** - v0.6.362 = #900 (in CI now). - v0.6.363 = the account policy: one-time admin resets, at least 15 characters for new local passwords, a `password-changed` audit entry, frames K and L. - v0.6.364 = this issue.
Author
Owner

Decision (Nikola, 2026-10-27): Q19 revised. An admin-created local account's first password is one-time too, with the same flow as a reset: 16 generated characters, valid for 7 days, and the person chooses their own at first sign-in. The first admin created in setup keeps the password as typed. This brings Cairn in line with OWASP ASVS 6.4.1 for new accounts. It ships with the next release, after v0.6.363.

The implementers' strings for #900 and the account policy are approved as shipped.

**Decision (Nikola, 2026-10-27): Q19 revised.** An admin-created local account's first password is one-time too, with the same flow as a reset: 16 generated characters, valid for 7 days, and the person chooses their own at first sign-in. The first admin created in setup keeps the password as typed. This brings Cairn in line with OWASP ASVS 6.4.1 for new accounts. It ships with the next release, after v0.6.363. The implementers' strings for #900 and the account policy are approved as shipped.
Author
Owner

Decisions (Nikola, 2026-10-27):

  • Account-password app pauses hold until resumed. A paused app stays paused until its person resumes it (or an admin bulk-resumes it), also after the notice ends or is ended early, and after a later declaration with another scope. Nothing stale syncs silently months later.
  • Paused public link page text approved: "This link is paused. The owner's administrator is checking the server after a restore. Try again later." (DE/FR/IT as drafted).
**Decisions (Nikola, 2026-10-27):** - **Account-password app pauses hold until resumed.** A paused app stays paused until its person resumes it (or an admin bulk-resumes it), also after the notice ends or is ended early, and after a later declaration with another scope. Nothing stale syncs silently months later. - **Paused public link page text approved:** "This link is paused. The owner's administrator is checking the server after a restore. Try again later." (DE/FR/IT as drafted).
Cordy closed this issue 2026-10-27 20:53:50 +00:00
Sign in to join this conversation.
No labels
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: Cordy/Cairn#904
No description provided.