https://www.youtube.com/watch?v=8mKpfutwD0U&t=1741s
Somewhere around 13th minute, Chris Tavers mentions this:
1 - In the queuing table, the application deletes lot of rows (typical for a
queuing table).
2 - Query trying to find out rows to be picked by the queue, accesses them via
indexes.
3 - Vacuum took lot of time to clean up dead rows.
4 - Indexes do not reach out to the latest visible tuple and has to traverse
lot of dead
rows before hitting the row required by the sql. This is because of (3).
My Question:
Is the (4) true for all updates. AFAIK, if an update occurs on even one index
col, the
index itself creates a new version in MVCC. Only HOT updates will end up in
situation
described in (3).
