Thnk you all for the reply. Maybe I could set up some metrics and count the
keys per subtasks/slot by myself.
However in the example of the playground there are 6 keys and they get
distributed in the 2 slots as 4 and 2: is this a bug (since Piotr said that key
groups can have sizes +/- 1 and in thi
Currently, we don't work on trying to ensure that the number of key groups
is as evenly spread as possible. As a workaround I would suggest to
increase the number of key groups or to change the key function.
Cheers,
Till
On Wed, Oct 23, 2019 at 1:42 PM Piotr Nowojski wrote:
> Hi,
>
> This is a
Hi,
This is a known issue of Flink. For example key groups can have sizes +/- 1 and
they are currently randomly distributed across the cluster, so some machines
will get more keys to handle then the others. If the number of keys is
relatively small, like 3 keys per key group, the load differenc
Hi to all,
I was looking into the Flink example of the Flink training trying to
understand why in the ClickEventCount[1] one task manager was reading
twice the speed of the other.
I had to debug a lot of internal code of Flink to understand that it
depends on the adopted hash function (used by Fl
