HBase directory Structure

The first part of the file is the Write-ahead log file that is processed by Hlog, and these log files are saved in http://www.aliyun.com/zixun/aggregation/13713.html "> The. Logs folder under the HBase root directory. Logs directory create a separate folder for each hregionserver, with several Hlog files under each folder (because of log rotation). Each hregionserver region shares a Hlog file.


when a log file is no longer needed (because its containing edit information is persisted to store files), the log file is saved to the database root. oldlogs folder. The files in the Oldlogs directory are deleted by master in 10 minutes (the time can be set by the Hbase.master.logcleaner.ttl parameter). Master checks these old log files every minute (the time is set by the Hbase.master.cleaner.interval parameter).


the second part of the file is hbase.id and hbase.version files, Hbase.id records the unique identity of the cluster; Hbase.version records the version number of the file format.


the third part, over time, some other directories will be created in the root directory. The split and. Corrupt directories are used during log splitting to hold some intermediate results and corrupted logs.


table-level files (table-level files)

Each table in the
HBase has a separate folder in the root directory (here we call the table directory). There is a file named. tableinfo in the table directory that holds the serialized htabledescriptor that corresponds to the table. Htabledescriptor contains table and column accessibility patterns. In addition to. Tableinfo, there is a. tmp directory. The. tmp directory has many effects, for example, that need to be used during the. Tableinfo update process.


domain-level files (region-level files)


In the table directory, create a separate directory for each region of the table, whose name is the MD5 hash value of region name. The entire directory structure looks like this:


/<hbase-root-dir>/<tablename>/<encoded-regionname>/<column-family>/<filename >


stores real data files in every column-family directory. These data files are named after random numbers and are generated by the Java built-in random number generator. The HBase program is very intelligent and can find random number repeats to prevent name duplication until an unused random number is found.

The
region directory contains a. regioninfo file that contains serialized information for the hregioninfo corresponding to the region. In addition to the file, there is an optional directory. TMP is created as requirements occur, such as rewriting files during the merge process.