A pure-Python implementation of Basho's bitcask key-value store.

GNU General Public License version 3

This is the current feature implementation status:

• Get
• Set
• Delete
• Create/read hint file
• Merge old files
• Rotation of data files

• A Bitcask instance is a directory;
• Only one system process will open a Bitcask instance for writing at a given time;
• The Bitcask instance directory consists of many files but only one is the "active". The other files are fixed, read-only (there is a rotation process to create a new active file when a threshold is exceeded);
• Once a file is closed, either purposefully or due to server exit, it is considered immutable and will never be opened for writing again;
• The active file is only written by appending (sequential writes do not require disk seeking);

To be done.

The data file consists on entries in this format:

[crc][timestamp][key_size][value_size][key][value]

Where:

• crc: 32-bit int with crc32 of [timestamp]...[value]
• timestamp: 32-bit int with UNIX timestamp
• key_size: 16-bit int with key size (in bytes)
• value_size: 32-bit int with value size (in bytes)

Some highlights:

• Everything is written in big endian;
• With each write, a new entry is appeneded to the active file (even if it is an update or delete operation)
• Deletion is simply a write of a kv-pair with the "value" field filled with a "tombstone" special value (the duplication of kv-pair will be removed on the merging process);

To be done.

To be done.

To be done.

To be done.