1) Setup standalone DJ with 2GB heap and 4 PDB backends of 10% db-cache, on a server with 8 CPUs. Have some entries in at least one of the backends.
2) Run online rebuild index
Server should run out of memory
The above configuration (40%/2GB/8CPUS) seems like a very reasonable configuration and should not lead to memory problems.
Depending on other parameters this can be reproduced even more easily, for example if there are more CPUs/worker threads or a few more indexes on the backends etc.