The HDFS plugin lets you use Hadoop Districbuted File System (HDFS) to store and retrieve GPDB backups when you run gpbackup and gprestore. For more information on gpbackup plugin refer to https://github.com/greenplum-db/gpbackup/tree/master/plugins

To use the HDFS plugin, you specify the location of the plugin. When you run gpbackup or gprestore, you specify the configuration file with the option --plugin-config.

If you perform a backup operation with the gpbackup option --plugin-config, you must also specify the --plugin-config option when you restore the backup with gprestore.

• Hadoop HDFS client installed on the greenplum cluster (coorindator and segments).
• Make sure you modify the executablepath in the hdfs_plugin_config.yaml file based on the absolute path of the hdfs_plugin.bash script.

This is a Hack and only meant to serve as a reference implementation for hdfs plugin for gpbackup. The recommended approach, would be to do away with the hadoop client cli dependancy and instead go with hdfs java or golang api for the plugin endpoints.

Right now the hdfs url isn't configuralbe(would only work against colocated hadoop cluster). This is something that can easily be supported by simply adding to the hdfs_plugin_config.yaml and consumed by the hdfs plugin.

Below command backups database testdb from gpdb to HDFS.

gpbackup --dbname testdb --plugin-config hdfs_plugin_config.yaml

Below command restores data from HDFS into gpdb database restoredb

gprestore --timestamp <YOUR_BACKUP_TIMESTAMP> --plugin-config hdfs_plugin_config.yaml --redirect-db restoredb --create-db