At TripAdvisor we use Hadoop and Hive for our warehousing needs. Processing the daily logs takes a long time, and re-processing them would be prohibitive. As we couldn't find a backup solution, we put one together. We open sourced it in hopes that it might be useful to others as well. You can find it on github:
https://github.com/TAwarehouse/backup-hadoop-and-hive The backup app traverses the hdfs filesystem looking for all files with mtime in a given range, then copying (a'la copyToLocal) the files to local directory. If hdfs were to crash, you can use "hadoop fs -copyFromLocal" to restore the filesystem contents. The backup can be invoked incrementally to keep updating the local copy. Files that would be overwritten get copied to a "preserved" area, so that older versions remain available. This project also includes a dump of the hive schema, along with hql statements to reassociate the tables with hdfs partitions. This portion came in very handy when we migrated our Hive backing db from derby to mysql. Thanks go to Josh Patterson, Edward Capriolo, and Rapleaf for letting us use their hdfs-style checksum, hive show-create-table, hdfs traversal code. For more info see the README: https://github.com/TAwarehouse/backup-hadoop-and-hive/blob/master/README.txt tom. tpalka<at>tripadvisor<dot>com
