Vitaliy Garnashevich wrote:
>
> > Have not worked through all of the above, but a first draft suggestion:
> >
> > Move the SELECT minus the aggregation functions into a sub-query that
> > uses FOR UPDATE. Then do the aggregation on the results of the
> > sub-query.
>
> The aggregated table has hundreds of millions of rows, and the query runs
> for many hours (which is one of the reasons why it's better not to fail). I
> really doubt that row level locking would work. That would be a lot of RAM
> just to hold all the locks.
Row locks are not stored in memory.
Of course, a FOR KEY SHARE lock would block DELETEs that try to remove
the locked row.
I think your proposed strategy of trying to merge what other processes
did while you were copying is very problematic.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
