Godfrey,
Glad I could help! I suspected that was what the problem was. I have made a
view in my postgres database to perform the inner lateral join, so that should
let me work around this for the time being.
Thanks,
Dylan
From: godfrey he
Date: Friday, November 20, 2020 at 1:09 AM
To: Dylan F
Hi Harshit,
the cluster.evenly-spread-out-slots strategy works the following way. If
you schedule a task w/o preferred inputs (e.g. no inputs or too many inputs
(I think the threshold is 8 inputs)), then it will pick a slot from a
TaskManager with the lowest utilization. If, however, the task has
Hi Rex,
The join key already has been used to organize records. As I said before,
"the join key is the key of the keyed states". So an iterate on the
MapState actually is a range scan (scan the join key prefix). However, this
will perform "seek" operation which is rather slow than "get" operation.
