DML refers to "Data Manipulation Language", a subset of SQL statements that modify the data stored in tables. Because Impala focuses on query performance and leverages the append-only nature of HDFS storage, currently Impala only supports a small set of DML statements:
INSERT in Impala is primarily optimized for inserting large volumes of data in a single statement, to make effective use of the multi-megabyte HDFS blocks. This is the way in Impala to create new data files. If you intend to insert one or a few rows at a time, such as using the INSERT ... VALUES syntax, that technique is much more efficient for Impala tables stored in HBase. See Using Impala to Query HBase Tables for details.
LOAD DATA moves existing data files into the directory for an Impala table, making them immediately available for Impala queries. This is one way in Impala to work with data files produced by other Hadoop components. (CREATE EXTERNAL TABLE is the other alternative; with external tables, you can query existing data files, while the files remain in their original location.)
To simulate the effects of an UPDATE or DELETE statement in other database systems, typically you use INSERT or CREATE TABLE AS SELECT to copy data from one table to another, filtering out or changing the appropriate rows during the copy operation.
Although Impala currently does not have an UPDATE statement, you can achieve a similar result by using Impala tables stored in HBase. When you insert a row into an HBase table, and the table already contains a row with the same value for the key column, the older row is hidden, effectively the same as a single-row UPDATE.
Currently, Impala cannot perform DML operations for tables or partitions stored in the Amazon S3 filesystem. See Using Impala to Query the Amazon S3 Filesystem for details.