OOM means you have too much data for the node. And the node disconnect will be related.
High level, your options are;
- Close or delete some indices
- Add more heap to the existing nodes
- Add more nodes to handle the data.
Upgrading can help, but you will likely run into this again. Rivers have been known to cause issues.
Monitoring, using Marvel, or similar, is important to prevent this and also to watch general resource consumption.