Check recall-memory pressure before assuming a File Sync read warmed the cache

Source facts

Azure File Sync reserves ten percent of total memory for persisting recalls to disk. Microsoft says that when sixty percent of this reserved memory is in use, recalls are not persisted to disk. Heavy access to many tiered files can reach this threshold and cause extra egress and degraded performance.

Cloud tiering maintains its own access heatmap. It does not depend on the NTFS last-access-time tracking feature, and Microsoft advises against manually enabling that NTFS feature for this purpose. Microsoft Learn.

Applicability

Identify the tiered-file population, concurrent access pattern and server memory context. Keep the percentage of reserved recall memory distinct from the percentage of total system memory.

DSE recommendation

DSE recommends checking whether recalled content actually remains local before expanding a bulk warm-up job. Review memory pressure and application demand with the file-service owner. Do not attempt to fix the heatmap by enabling unrelated NTFS tracking. If repeated access is expected, establish an approved cache-readiness test rather than equating a completed read with persistence.

Verification

Use representative harmless files and a bounded access pattern. Observe local file state, repeated access behavior and relevant server pressure during the test. Record whether content persisted and whether concurrency changes the result. Retain the workload conditions with observations so a successful low-load trial is not presented as assurance for a much larger recall burst.

Official references

Microsoft Learn: Understand Azure File Sync Cloud Tiering. Source retrieved September 9, 2026.