Issue Summary

We suspect data loss can occur when sweeping a layered dhandle is supposed to be drained during step up. This was identified following the investigation in WT-16703.

Context

• The root cause is sweeping layered dhandles during concurrent draining.
• Multiple solutions were discussed:
  • Short term:
    • Scanning the entire metadata and draining each table.
    • Preventing the sweep server from sweeping layered dhandles.

• • Long term:
    • Tie the lifetime of a layered dhandle to its ingest dhandle.

• The immediate next step is to implement the following short-term solution: the sweep server will not sweep layered dhandles. This approach was chosen due to its low performance impact and minimal code changes.

Proposed Solution

• Implement the change so that the sweep server never sweeps layered dhandles.
• Consider creating a ticket for the long-term solution to tie dhandle lifetimes.

This ticket was generated by AI from a Slack thread.