Refresh returns stale data after large-scale UPDATE with asynchronous indexing

[wyfanxiao]

Environment

• Version: seekdb (OceanBase seekdb)
• Revision: 1-4a092697e4a0f07ff7fb77e62b453489811acf9e
• Build Branch: master
• Build Time: Apr 13 2026 23:38:32
• Build Flags: RelWithDebInfo|Sanity
• Build Info: obbu...

Description

After performing a large-scale UPDATE operation, a refresh of the asynchronous index returns stale (outdated) data.

Steps to Reproduce

1. Perform a large-scale UPDATE operation on a table with an asynchronous index.
2. Trigger a refresh of the asynchronous index.
3. Query the data via the index.
4. Observed result: The returned data is stale and does not reflect the updates.

Impact

Queries relying on the refreshed asynchronous index will receive incorrect, outdated data, leading to data consistency issues for applications.

[hnwyllmm]

This can also happen with synchronous indexes.

=== Conclusion ===

PASS: async=10/10, immediate=0/10 both normal.

=== Bug #2 Minimal Reproduction ===

[async] Insert 50,000 rows...
[async] UPDATE 15,000 rows → [0,0,0,0]...
[async] UPDATE 1,000 rows → [99,99,99,99]...
After the first async refresh, top-10 hits updated rows: 10/10

[immediate] Insert 50,000 rows...
[immediate] UPDATE 15,000 rows → [0,0,0,0]...
[immediate] UPDATE 1,000 rows → [99,99,99,99]...
After immediate UPDATE, top-10 hits updated rows: 5/10

[hnwyllmm]

This also happens with synchronous indexes.

=== Bug #2 Minimal Reproduction ===

[async] Insert 50,000 rows...
[async] UPDATE 15,000 rows → [0,0,0,0]...
[async] UPDATE 1,000 rows → [99,99,99,99]...
After the first async refresh, top-10 hits on updated rows: 10/10

[immediate] Insert 50,000 rows...
[immediate] UPDATE 15,000 rows → [0,0,0,0]...
[immediate] UPDATE 1,000 rows → [99,99,99,99]...
After immediate UPDATE, top-10 hits on updated rows: 5/10

=== Conclusion ===

PASS: Both async=10/10 and immediate=5/10 are normal.