Always capture segments flusher first for consistency (#7882)

agourlay · web-flow · commit e62d447ef5b5 · 2026-01-08T11:50:30.000+01:00
diff --git a/lib/segment/src/segment/entry.rs b/lib/segment/src/segment/entry.rs
@@ -637,6 +637,7 @@ impl SegmentEntry for Segment {
             (_, _) => {}
         }
 
+        // Capture all flushers first to improve data consistency
         let vector_storage_flushers: Vec<_> = self
             .vector_data
             .values()
diff --git a/lib/shard/src/segment_holder/flush.rs b/lib/shard/src/segment_holder/flush.rs
@@ -74,16 +74,17 @@ impl SegmentHolder {
         // as it is exclusive
         let mut background_flush_lock = self.lock_flushing()?;
 
+        // Capture all flushers first to improve data consistency
+        let flushers: Vec<_> = segment_reads
+            .iter()
+            .filter_map(|read_segment| read_segment.flusher(force))
+            .collect();
+
         if sync {
-            for read_segment in segment_reads.iter() {
-                read_segment.flush(force)?;
+            for flusher in flushers {
+                flusher()?;
             }
         } else {
-            let flushers: Vec<_> = segment_reads
-                .iter()
-                .filter_map(|read_segment| read_segment.flusher(force))
-                .collect();
-
             *background_flush_lock = Some(
                 std::thread::Builder::new()
                     .name("background_flush".to_string())

Original file line number	Diff line number	Diff line change
`@@ -637,6 +637,7 @@ impl SegmentEntry for Segment {`
`637`	`637`	`(_, _) => {}`
`638`	`638`	`}`
`639`	`639`
	`640`	`+ // Capture all flushers first to improve data consistency`
`640`	`641`	`let vector_storage_flushers: Vec<_> = self`
`641`	`642`	`.vector_data`
`642`	`643`	`.values()`