Summary

• search service, request, server, build, and rollout logs in VictoriaLogs instead of filtering only browser-loaded entries
• add validated 1h, 6h, 24h, and 7d query ranges, defaulting continuous log views to 24h
• preserve search and range parameters through polling and older-log pagination
• strictly validate RFC 3339 pagination cursors and numeric limits before constructing LogsQL, with query-layer validation as a second boundary
• keep existing logs visible during debounced searches and transient polling failures
• fix same-key pagination request ownership and highlight every case-insensitive match
• document the distinction between query range and the existing storage retention setting

Why

The log search box filtered only the records already loaded in the browser. Continuous views initially load up to 500 records, so matching older logs could not be found even though VictoriaLogs retained them.

Impact

Users can now search retained backend data from every log viewer. Continuous service, request, and server feeds include a time-range selector, while build and rollout searches remain scoped to their identifiers. Existing log retention remains unchanged.

Malformed before and after values are rejected before VictoriaLogs is queried, preventing cursor text from changing tenant filters or injecting pipeline stages. Upstream failures still return an error response, but cached logs remain on screen while the viewer reports the refresh failure.

Validation

• Vitest: 17 files, 83 tests passed, including cursor-injection, deployment cursor direction, route validation, upstream failure, OR grouping, and highlight regressions
• TypeScript: tsc --noEmit
• Biome: all changed files pass
• Next.js 16 production build passed
• git diff --check

Tracking

• Basecamp todo