[ALTER INDEX] reducing number of replicas does not always honor server group constraints

4 node cluster -2 kv,data,index nodes, 2 index nodes

2 server groups -
group one - 1 kv,data,index node, 1 index node
group two - 1 kv,data,index node, 1 index node

1.load 400k docs with workloadgen

/opt/couchbase/bin/cbworkloadgen -n 172.23.104.110:8091 -r 100 -i 400000 -b default -s 1024 -j -u Administrator -p password

May need to repeat the following steps several times:
2. create index
CREATE INDEX random_index_929791 ON default(age) USING GSI
3. alter index to increase copies across groups
ALTER INDEX default.random_index_929791 WITH

{"action":"replica_count","num_replica": 2}

4.after starting step 3, wait a few seconds then start step 4, alter index again to reduce number of copies
ALTER INDEX default.random_index_929791 WITH

{"action":"replica_count","num_replica": 1}

The indexes that remain will both be on the same server group. This only seems to happen if the second alter index happens before the first alter index replicas can be finished building. For me the server goup nodes were 173 and 175 (group 2) which the indexes stayed on.