Hi Can,
We are using 0.6.1, cca 160 vertices, mostly cooperative ones. We have reviewed the metrics collected, it generates 30k vertex executions across the cluster, cca 10k per node.
So I have set up additional metrics, especially in places where Processor.tryProcess returns false. It helped us to detect slow Vertex implementation.
What exactly is happenning when tryProcess returns false repeatedly? Is the execution engine trying to work somewhere else when some part of the flow is overloaded?
Is this case covered by 0.7 metrics?
Regards
Lukas
Dne čtvrtek 20. září 2018 13:50:36 UTC+2 Can Gencer napsal(a):