a few questions:
- what amount of RAM have you allocated to KNIME? It should be enough but maybe not too close to the absolute limit you have
- have you tried to make a clean start of KNIME once?
- then you could try to use Parquet as internal storage and
- configure some nodes that do heavy work to store only small amounts of data in RAM and otherwise use the disc
Another thing you could try is to organize your workflow so only one node is executed at the same time thru Flow variables. I think in the current configuration there could be parallel executions.
And maybe if this does not help you could send a larger portion of the relevant LOG file.