Out of memory during non-completed viewChange process (under load)

Prehistory:
1) use AWS Perfomance pool
2) create new pool with 20 nodes
3) write about 1k txns
4) demote node2 and node3
5) write about 97k txns

After all this start the load test
Used load settings: sudo python3.5 perf_processes.py -g pool_transactions_genesis_performance -c 100 -n 5000 (from single agent)

Actual results:

• Pool stops to write after ~139k ledger size
• Many nodes have been blacklisted (the first time it happened an hour after load start)
• Current viewNo: 411 (in progress), but we have "INSTANCE_CHANGE {'viewNo': 959, 'reason': 28}

  " in validator-info output

• From some point "cannot allocate memory" began to appear every ~10 minutes

Logs will be attached later