Problem

WT_PAGE_COMPACTION_WRITE is set in bt_compact.c when compaction marks an in-memory page for rewriting, and unconditionally cleared in rec_write.c after __reconcile() returns. No code between those two points checks the flag or emits a stat.

The existing compaction write byte stat only covers the direct on-disk block relocation path (__wt_block_compact_page_rewrite). The reconciliation path — where compaction dirties an in-memory page and lets it flow through normal reconciliation — is entirely unaccounted for, leaving a gap in compaction I/O observability.

Fix

Add a new connection-level stat incremented in rec_write.c when the WT_PAGE_COMPACTION_WRITE flag is set at reconciliation time, before the flag is cleared. This closes the observability gap and makes compaction-driven writes fully visible regardless of which path triggered them.

References

• Flag set: bt_compact.c — compaction marks in-memory page for rewriting
• Flag cleared: rec_write.c — unconditionally after __reconcile() returns
• Defined: btmem.h