Keeping a process running is only part of keeping work reliable. Cloudflare now delays some Durable Object code transitions, D1 has an approaching result-size limit, and Vercel Daemons can restart containers that fail health checks. Together, these changes make deployment timing, bounded data access and meaningful health signals worth reviewing.
Coverage: three primary announcements dated 30 October 2026, the preceding calendar day in the preferred 24-hour scan. Exact publication hours are unspecified; no 72-hour expansion was needed. These are vendor announcements, not capabilities tested in a HexOrbit deployment.
Durable Objects: a transition window, not a completion guarantee
Cloudflare now defers code updates by default. An active Durable Object can keep its current code for up to five minutes while waiting to hibernate. Connections using the WebSocket Hibernation API remain connected when new code is adopted after hibernation. If the delay expires first, the object resets and adopts the update.
The delay is configurable, and an immediate mode is available. Deferral is best effort: an object may still reset early, including during a runtime update. HexOrbit interpretation: test deployment checks that expect new behavior immediately, and retain interruption handling for long tasks. A grace period does not make a task durable by itself.
Primary source — Cloudflare, 30 October 2026: https://developers.cloudflare.com/changelog/post/2026-10-30-durable-objects-deferred-code-updates/
D1: prepare large-result queries before 4 November
Cloudflare says D1 will enforce a 50 MB maximum query result starting 4 November 2026. Results exceeding it will fail. REST API import and export operations are exempt from this limit.
The announcement recommends selecting necessary columns and paginating large results. HexOrbit interpretation: review administrative reports and bulk-read paths before enforcement, including their handling of failed or partial retrieval. This is a query-result limit, not an announced database-capacity limit. No local query workload was inspected or changed.
Primary source — Cloudflare, 30 October 2026: https://developers.cloudflare.com/changelog/post/2026-10-30-query-result-size-limit/
Vercel Daemons: detect processes that run but no longer work
Vercel Daemons now accept HTTP liveness checks. A lightweight endpoint should report an error when the daemon is unhealthy; the configuration points livenessProbe at that endpoint. Vercel periodically checks it and restarts the container on failure.
HexOrbit interpretation: define what unhealthy means for the actual worker, then test both detection and restart behavior. A process answering HTTP is not necessarily completing useful work. Restarting also does not establish that interrupted work is safe to repeat; consider duplicate handling and task recovery separately. No probe or infrastructure setting was changed here.
Primary source — Vercel, 30 October 2026: https://vercel.com/changelog/add-liveness-checks-to-your-daemons
Practical next step
Choose one long-running workflow and document three boundaries: what happens during a code transition, how much data one read can return, and which signal proves the worker is healthy. Use those answers to design a small maintenance test before adopting the changes.