When a Special character like '$' or '#' are present in column name, sqoop/avro removes those special character. In some cases it leads to duplicate column.
e.g. If we have COL$1 and COL1$ in the schema, it removes both of them and creates the duplicate column as COL1 and it results in failure of the SQOOP import job as a avro data. The same table can be loaded without --as-avarodata flag.
The similar issue is raised in, https://issues.apache.org/jira/browse/SQOOP-1361 - which i suppose is fixed and the fix is creating this new issue.