circagraph
Sign inSign up

openchronology/timeline-app on GitHub

The OpenChronology repository on GitHub, mirrored by och-import-github: issues and pull requests as durations (ongoing while open), what happened to them as moments, plus releases, milestones, commits and branches. The Branches view shows the most recently active branches each in a track of its own.

Source: https://github.com/openchronology/timeline-app

3,131 moments · 2026-10-05 CE 20:21:49 UTC – 2026-11-16 CE 12:10:45 UTC · by @athan · @athan

#410 Sign-in: verification instead of a per-account limit, and queued password checks

–

  • github
  • github/openchronology/timeline-app
  • github/openchronology/timeline-app#410
  • pull request
  • state:merged

## Summary **Per-account sign-in handling** - Removes the hard per-username sign-in limit. After three failures in fifteen minutes for an account (or from an address), each further attempt must pass human verification. This existing check is now the only per-account measure, so the account's owner can always sign in by passing it. - With `CHALLENGE_PROVIDER=off`, the same failures delay each further sign-in for that account by 1.5 seconds. They are never refused. - Directory (LDAP) sign-in follows the same rule. Linking a directory account is now limited per signed-in account and per address, not per directory name. - The per-address limits are unchanged. **Password-check capacity** - The concurrent scrypt cap is now sized from `os.availableParallelism()`: one slot per core, at least two, and below libuv's worker threads (`UV_THREADPOOL_SIZE`, default 4). That gives three slots with the defaults. - The new setting `PASSWORD_HASH_CONCURRENCY` (1–64) overrides it. - A check that finds every slot busy waits in a first-come, first-served line for up to five seconds before getting `503`. Before, it got `503` at once. The line is bounded at 32 waiters per slot. - Sign-in, registration, password reset and change, and the admin password tools all share these slots. - Unknown usernames still do the same derivation through the same line and slot. Timing equalisation and the scrypt parameters are unchanged. - A password reset link is checked to exist before the breach lookup and the hashing. **Docs** - `docs/authentication.md`, the OpenAPI descriptions (sign-in, register, reset), `.env.example`, `compose.yml` and the Debian env template are updated. ## Tests - `test/challenge-postgres.mjs` (`pow`): a dozen challenged failures for an account, each from its own address, then its owner signs in. Without this change this gets 429. - `test/postgres.mjs` (`off`): - A dozen failures for an account, then its owner signs in, delayed but not refused. Without this change this gets 429. - A burst of unknown-username sign-ins plus the owner's sign-in against two slots: all are answered. Without this change some get 503. - `test/auth.test.mjs`: parsing of the slot setting; checks queue and finish; a wait that runs out gives 503. - `test/challenge.test.mjs`: `signInGate` delays under `off` and asks for verification (428) under `pow`. - Also passing: `npm test` (404 tests), `test/account-security-postgres.mjs` and `test/ldap-postgres.mjs`. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_011fFY8eruEi1xAkkMBni13B https://github.com/openchronology/timeline-app/pull/410

Related

Open #410 Sign-in: verification instead of a per-account limit, and queued password checks in the timeline

#410 Sign-in: verification instead of a per-account limit…