On 2020-Nov-26, Huan Ruan wrote:
> Hi All
>
> We cannot work out a reproducible case but we have a copy of the offending
> database. I was hoping to know
I think the way to debug this would be to see what WAL records have been
emitted for the index, using pageinspect to find the problem index
tuple.
Use 'select ctid rrom large_table_with_623m_records where ...' to
pinpoint the unindexed tuple's page number; see when (in LSN) was that
tuple written; inspect WAL surroundings looking for updates (or lack
thereof) for the BRIN index. Use pageinspect to examine raw brin data.