Rocksdb version in memo release v0.1.2 is low.
It doesn't support rocksdb ingest,ssttabelwriter api.
So in v0.1.2 it is hard to dump or ingest raw sst table file.
Is there a stabilize nemo version with rocksdb version higher than 5.0?

There is no api to release snapshot of rocksdb instance when use nemo iterator(HIterator KIterator ...)with snapshot.

dbnemo column family has a compact filter class(CompactionFilter) with a private pointer name db pointer to memo db instance.In constructor of CompactionFilter,pointer db is null.
After memo is opened,the pointer will be set .
However, if the pointer has not been set, compact function maybe excecute.
Then, in CompactionFilter method ShouldDrop call DBNemoImpl::GetVersionAndTS, there will be a null ptr reference of private member "db".

If k is set to zero,then the wile statement will be pass and idx_flag[0] will be set to 1;
If k is set to zero the second time, idx_flag[0] will be set to 2;
The first member of set will be return twice in the case above.
Further more , the biggest member will be returned with higher probability if the rand() function return the same value.
//......
int k = rand() % remain_card + 1;
int i = 0;
int cnt = 0; // the valid k-th number
while (i < card && cnt < k) {
if (idx_flag.find(i) == idx_flag.end()) {
cnt++;
}

            if (cnt == k) break;
            i++;
        }
        idx_flag[i]++;
        remain_card--;

In multi keys operation like SUnion SInter ZUnion ZInter RpopLpush,
nemo code requests record lock with the input argument order.
For example,thread 1 locks on "Key1" then requests for "Key2",thread 2 locks on "Key2" requests for "Key1".Then dead lock occurs.
Thread mutex of record lock must be locked in the same order.

zexpire接口看起来是只修改了meta中的ttl字段，member to score/score to member中的过期value如何在compcation中被清掉？