Bigdata SQL: Why Is DML Difficult on HDFS?

HDFS, the distributed file system on which most big data tools and frameworks are based, is architected to be WORM (write once read many). HDFS support appends but performs no updates. Modifying data becomes an inherent limitation of HDFS, hence, most SQL solutions do not support any DML operations on Hadoop. Some vendors come up with novel ways of supporting updates by logging modifications and then merging the modifications with the original data.