#742 batch B: admin switch for local sign-in during a coordination outage (strict mode) #1179

Closed
opened 2026-11-16 03:18:03 +00:00 by Cordy · 2 comments
Owner

Follow-up from #742 batch A (decided 2026-11-16).

Batch A behaviour during a majority loss (fx9b, B2):

  • Local accounts without a two-step record (as last seen on the member, list refreshed every 10 s) sign in with the password.
  • Accounts with one are refused with 503 login-coord-unavailable.

Residual: a two-step setup made on another member in the 10-15 s before an outage is not yet on the list, so that account can sign in with the password alone during the outage. Batch A documents this.

Market (verified 2026-11-16): the most common MFA proxy offers failmode=safe (default: allow if primary auth succeeds when the MFA service is unreachable) and failmode=secure (reject everyone). NIST SP 800-53 SC-24 leaves the failure state to the organization.

Batch B: add an admin switch, off by default and compliance-friendly, that refuses every local password sign-in during a coordination outage ("fail secure"). Include docs, the audit/settings surface and tests. Research per the playbook first.

Follow-up from #742 batch A (decided 2026-11-16). Batch A behaviour during a majority loss (fx9b, B2): - Local accounts **without** a two-step record (as last seen on the member, list refreshed every 10 s) sign in with the password. - Accounts **with** one are refused with 503 `login-coord-unavailable`. **Residual:** a two-step setup made on another member in the 10-15 s before an outage is not yet on the list, so that account can sign in with the password alone during the outage. Batch A documents this. **Market (verified 2026-11-16):** the most common MFA proxy offers `failmode=safe` (default: allow if primary auth succeeds when the MFA service is unreachable) and `failmode=secure` (reject everyone). NIST SP 800-53 SC-24 leaves the failure state to the organization. **Batch B:** add an admin switch, off by default and compliance-friendly, that refuses every local password sign-in during a coordination outage ("fail secure"). Include docs, the audit/settings surface and tests. Research per the playbook first.
Author
Owner

Research: research-b10-r1-auth.md. Decided for batch 11 (v0.6.412), together with the shared sign-in records.

  • One rule: "new sign-ins are paused" during a coordination outage. It covers every new sign-in, including identity-provider sign-ins.
  • Exemption: the emergency account is exempt.
  • Default: off.
  • With the switch off: batch A behaviour stays. Sessions are held on that member and written to the store on recovery.
  • Mockup frames before the build: the Two-step sign-in admin block, and the sign-in page refusal.
Research: `research-b10-r1-auth.md`. Decided for batch 11 (v0.6.412), together with the shared sign-in records. - **One rule:** "new sign-ins are paused" during a coordination outage. It covers every new sign-in, including identity-provider sign-ins. - **Exemption:** the emergency account is exempt. - **Default:** off. - **With the switch off:** batch A behaviour stays. Sessions are held on that member and written to the store on recovery. - **Mockup frames before the build:** the Two-step sign-in admin block, and the sign-in page refusal.
Author
Owner

Shipped in v0.6.413 (#742 batch B part 2b, PR #1195). Users & access › Access › Two-step sign-in, "During a coordination outage" (shown with several instances only): "Keep signing people in on the instance they reach" (default) or "Pause all new sign-ins" (asks first). With the pause on, during an outage every new sign-in with a password answers 503 signin-paused (the sign-in form, the identity-provider start and callback, and Basic with an account password); existing sessions, app passwords, API keys and emergency accounts keep working; an unknown setting counts as paused. Audit verb signin-outage-strict-changed. The B2 window (a two-step setup made on another member just before an outage) is closed while the pause is on. Verified on the VM cluster (run-r16 step 11) and by real-main tests. Closing.

Shipped in v0.6.413 (#742 batch B part 2b, PR #1195). Users & access › Access › Two-step sign-in, "During a coordination outage" (shown with several instances only): "Keep signing people in on the instance they reach" (default) or "Pause all new sign-ins" (asks first). With the pause on, during an outage every new sign-in with a password answers 503 `signin-paused` (the sign-in form, the identity-provider start and callback, and Basic with an account password); existing sessions, app passwords, API keys and emergency accounts keep working; an unknown setting counts as paused. Audit verb `signin-outage-strict-changed`. The B2 window (a two-step setup made on another member just before an outage) is closed while the pause is on. Verified on the VM cluster (run-r16 step 11) and by real-main tests. Closing.
Cordy closed this issue 2026-11-22 01:15:32 +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#1179
No description provided.