← ~/changelog / 2026-08-27

uSync: fixed and finished background processing mode (opt-in), then added load-balanced backoffice support

Repo: KevinJump/uSync

• Fixed the background sync path (SyncProcessingMode.Background): it was passing the run's requestId as the long-running-operation type (so the concurrency guard never worked) and returning an id the client could never use to check on the run
• Added Status/Running polling endpoints, backed by a per-run progress cache, so the client can poll progress and reattach to a run after a page reload
• Fixed export-to-file downloading before the export had actually finished writing
• Fixed a shared Stopwatch in SyncActionService that made elapsed-time logging a cross-run/cross-user race
• Diagnosed and fixed a self-inflicted SQLite 'table is locked' issue: fixed-interval status polling was itself enough concurrent DB traffic to trigger it during a background import; polling is now a SignalR fallback only, with backoff
• Fixed status polling to be staleness-aware - a crashed/restarted server used to leave the client stuck reporting "running" forever; now self-heals within a few minutes, plus a manual dismiss button in the UI as an immediate escape hatch
• Kept ProcessingMode defaulted to Normal (opt-in only, not a default change mid-version); gave the new ISyncManagementService members default interface implementations so it stays non-breaking for external implementers
• Merged via PR #1054
• Made background processing mode work correctly behind a load-balanced backoffice: progress/status now resolves from any server via the shared long-running-operation record, not just the one that ran the job
• Fixed the client's polling fallback - it used to only kick in when the SignalR socket was fully disconnected, so a socket pinned to the wrong server (no sticky sessions) left the UI hung on a finished run; now it falls back based on message liveness instead
• Fixed the startup import running concurrently on every server under a load-balanced setup - its only guard assumed a role that topology doesn't use; now elects a single server via the long-running-operation service instead
• Tightened the background-mode poll interval from 5s to 1s now that polling is gated on liveness rather than being the sole progress mechanism
• Surfaced a progress message during polling fallback too - it previously only ever came from a live SignalR push, so the message line stayed blank whenever polling was doing the work
• Open as PR #1055