I’ve tested the patch across several thousand test cases, and no faults of any kind have been observed.
Additionally, I independently built a closed banking transaction system to verify consistency during REINDEX CONCURRENTLY while multiple backends were writing simultaneously. The results showed no missing transactions, and the validation logic worked exactly as expected. On large tables, I observed a significant speedup—often several times faster.
I believe this patch is highly valuable, as REINDEX CONCURRENTLY is a common maintenance operation. I also noticed that there is a separate thread working on adding support for concurrent reindexing of partitioned indexes. Without this patch, that feature would likely suffer from serious performance issues due to the need to reindex many indexes in one go—making the process both time-consuming and lock-intensive.
