Cluster is broken apart after finish upgrade from 2.5.0 to 4.5.0-2585 due to memcached disconnected

Run test newupgradetests.MultiNodesUpgradeTests.online_upgrade_rebalance_in_out,initial_version=2.5.0-1059-rel,items=1000000,upgrade_version=4.5.0-2585,stop-on-failure=True,get-cbcollect-info=true,skip_init_check_cbserver=true,skip_cleanup=True

Test failed due to not able to verify items after upgrade is done. Check cluster (143 and 108), I saw that after DCP upgrade rebalance was done, and in few minutes, memcached on one node was disconnected (172.23.105.108) and down. When node 108 was up, this node was not part of original cluster any more.

This test was passed on build 4.5.0-2468