how to load a Kafka topic to HDFS?

we use camus.

Camus is a simple MapReduce job developed by LinkedIn to load data from Kafka into HDFS. It is capable of incrementally copying data from Kafka into HDFS such that every run of the MapReduce job picks up where the previous run left off. At LinkedIn, Camus is used to load billions of messages per day from Kafka into HDFS.

But it looks like it's replaced with gobblin

Gobblin is a universal data ingestion framework for extracting, transforming, and loading large volume of data from a variety of data sources, e.g., databases, rest APIs, FTP/SFTP servers, filers, etc., onto Hadoop. Gobblin handles the common routine tasks required for all data ingestion ETLs, including job/task scheduling, task partitioning, error handling, state management, data quality checking, data publishing, etc. Gobblin ingests data from different data sources in the same execution framework, and manages metadata of different sources all in one place. This, combined with other features such as auto scalability, fault tolerance, data quality assurance, extensibility, and the ability of handling data model evolution, makes Gobblin an easy-to-use, self-serving, and efficient data ingestion framework.

You have several other options as well:

• Use Apache Flume to read messages from Kafka and write them to your HDFS. There are several examples of how you can set it up, but one article from Cloudera covers that topic quite well. They even named the solution Flafka ;)
• Use Kafka HDFS Connector, which is quite simple to set up. However, it would require Confluent Kafka (which still is open sourced).

We tested both quite successfully.