[maxmad64bis]

Problem

The credential health scheduler tests every active provider connection on a single global cadence (CREDENTIAL_HEALTH_CHECK_INTERVAL, default 5 min). There is no way to pace heavy or low-value connections individually, and no per-connection opt-out. Issue #8443 reports the probe volume this generates (~6 000 probe requests/day for ~20 providers, as corrected in the thread).

Fix

• The sweep reads provider_connections.healthCheckInterval (minutes) per connection: > 0 overrides the global cadence, <= 0 excludes the connection from the sweep (opt-out), absent falls back to the env default. The column already exists (the token-refresh scheduler uses it) — no migration, no API or UI change (the field is already editable per connection).
• On success the scheduler keeps the per-connection timing (nextAttemptAt = lastAttemptAt + interval) instead of clearing it, so a healthy connection is retested on its own schedule, not every sweep.
• Failure backoff (5 min → 10 → 30 → 2 h) is unchanged and takes priority.
• Removed dead code in the same file: OAUTH_INTERVAL_MULTIPLIER (never used — the inert multiplier called out in the issue thread; the header comment claiming OAuth is tested 2x less often was false), getMaxFailuresAcrossConnections (never called), and a leftover interval variable in sweep().

Note on the shared column: a value set for token-refresh purposes also paces credential tests (and 0 disables both) — one knob, two schedulers. A per-connection interval shorter than the global sweep cadence is capped by it (the sweep timer itself is unchanged).

⚠️ base-red inherited: #9985

Verification

The sweep predicates are replicated in the unit tests, following the existing house pattern for this file (the scheduler auto-initializes on import); the replica and the implementation were aligned line by line.

$ node --import tsx/esm --test tests/unit/credential-health-interval.test.ts

# tests 6
# pass 6
# fail 0

$ node --import tsx/esm --test tests/unit/credential-health-backoff-retry.test.ts

# tests 6
# pass 6
# fail 0

The full credential-health family (5 files: active-connections-9180, backoff-retry, boot-wiring, interval, search-providers) passes, 0 fail; the three pre-existing files are unchanged and pass as at base.

$ npx eslint tests/unit/credential-health-interval.test.ts tests/unit/credential-health-backoff-retry.test.ts src/lib/credentialHealth/scheduler.ts

/home/.../src/lib/credentialHealth/scheduler.ts
  21:1  error  '@/lib/localDb' import is restricted ...  no-restricted-imports

✖ 1 problem (1 error, 0 warnings)

The full npm run lint is red on the base branch (25 errors in untouched test files, byte-identical to upstream/release/v3.8.50); the one error above (no-restricted-imports on the @/lib/localDb import) also exists at base. This branch introduces no new lint error.

Tests added/modified: tests/unit/credential-health-interval.test.ts (new), tests/unit/credential-health-backoff-retry.test.ts (adapted: success now spaces the next sweep by the per-connection interval instead of clearing the timing).

[diegosouzapw]

Solid, focused feature: per-connection sweep cadence via the already-existing provider_connections.healthCheckInterval column, with <=0 as an opt-out and absent falling back to the env default. Correctly reuses the existing column (no migration/API/UI change), removes genuinely dead code (OAUTH_INTERVAL_MULTIPLIER was never used; getMaxFailuresAcrossConnections never called), and ships real unit tests that pass (I verified the full credential-health family green: 6+6+8). Cleanup observations only, none blocking: (1) the isOAuth parameter on testConnection is now fully unused after removing OAUTH_INTERVAL_MULTIPLIER — worth dropping to avoid confusion; (2) the sweep predicates are replicated in the tests rather than imported (documented house pattern because the scheduler auto-inits on import) — keep the replica/implementation aligned as you have been; (3) note the shared-column coupling is real and worth an explicit callout in the release notes: a healthCheckInterval set for token refresh also paces credential health, and 0 disables both (the EditConnectionModal defaults to 60, so configured OAuth connections will now be credential-checked at 60min instead of ~10min — aligned with the probe-volume goal of #8443, but a deliberate behavior change). No new lint errors vs base; the single no-restricted-imports on the @/lib/localDb import predates this PR. ready to merge.

[maxmad64bis]

Thanks for the review! Both optional cleanups are in (commit e2160f7): dropped the unused isOAuth param on testConnection, and added the release-notes callout about the shared healthCheckInterval column (60min default for OAuth connections, 0 disables both). Typecheck and tests are clean — only the known omniglyph base-red remains. Ready to merge whenever.