Skip to content

Fast version controlled database based upon memory mapped containers

License

Notifications You must be signed in to change notification settings

golos-blockchain/chain-db

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

81 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ChainBase - a fast version controlled, transactional database

ChainBase is designed to meet the demanding requirments of blockchain applications, but is suitable for use in any application that requires a robust transactional database with the ability have near-infinate levels of undo history.

While chainbase was designed for blockchain applications, it is suitable for any program that needs to persist complex application state with the ability to undo.

Features

  • Supports multiple objects (tables) with multiple indicies (based upon boost::multi_index_container)
  • State is persistant and sharable among multiple processes
  • Nested Transactional Writes with ability to undo changes

Dependencies

  • c++11
  • Boost
  • CMake Build Process
  • Supports Linux, Mac OS X (no Windows Support)

Example Usage

enum tables {
   book_table
};

/**
 * Defines a "table" for storing books. This table is assigned a 
 * globally unique ID (book_table) and must inherit from chainbase::object<> which
 * decorates the book type by defining "id_type" and "type_id" 
 */
struct book : public chainbase::object<book_table, book> {

   /** defines a default constructor for types that don't have
     * members requiring dynamic memory allocation.
     */
   CHAINBASE_DEFAULT_CONSTRUCTOR( book )
   
   id_type          id; ///< this manditory member is a primary key
   int pages        = 0;
   int publish_date = 0;
};

struct by_id;
struct by_pages;
struct by_date;

/**
 * This is a relatively standard boost multi_index_container definition that has three 
 * requirements to be used withn a chainbase database:
 *   - it must use chainbase::allocator<T> 
 *   - the first index must be on the primary key (id) and must be unique (hashed or ordered)
 */
typedef multi_index_container<
  book,
  indexed_by<
     ordered_unique< tag<by_id>, member<book,book::id_type,&book::id> >, ///< required 
     ordered_non_unique< tag<by_pages>, BOOST_MULTI_INDEX_MEMBER(book,int,pages) >,
     ordered_non_unique< tag<by_date>, BOOST_MULTI_INDEX_MEMBER(book,int,publish_date) >
  >,
  chainbase::allocator<book> ///< required for use with chainbase::database
> book_index;

/**
    This simple program will open database_dir and add two new books every time
    it is run and then print out all of the books in the database.
 */
int main( int argc, char** argv ) {
   chainbase::database db;
   db.open( "database_dir", database::read_write, 1024*1024*8 ); /// open or create a database with 8MB capacity
   db.add_index< book_index >(); /// open or create the book_index 


   const auto& book_idx = db.get_index<book_index>().indicies();

   /**
      Returns a const reference to the book, this pointer will remain
      valid until the book is removed from the database.
    */
   const auto& new_book300 = db.create<book>( [&]( book& b ) {
       b.pages = 300+book_idx.size();
   } );
   const auto& new_book400 = db.create<book>( [&]( book& b ) {
       b.pages = 300+book_idx.size();
   } );

   /**
      You modify a book by passing in a lambda that receives a
      non-const reference to the book you wish to modify. 
   */
   db.modify( new_book300, [&]( book& b ) {
      b.pages++;
   });

   for( const auto& b : book_idx ) {
      std::cout << b.pages << "\n";
   }

   auto itr = book_idx.get<by_pages>().lower_bound( 100 );
   if( itr != book_idx.get<by_pages>().end() ) {
      std::cout << itr->pages;
   }

   db.remove( new_book400 );
   
   return 0;
}

Concurrent Access

By default ChainBase provides no synchronization and has the same concurrency restrictions as any boost::multi_index_container. This means that two or more threads may read the database at the same time, but all writes must be protected by a mutex.

Multiple processes may open the same database if care is taken to use interpocess locking on the database.

Persistance

By default data is only flushed to disk upon request or when the program exits. So long as the program does not crash in the middle of a call to db.modify(), or db.create() the content of the database should remain in a consistant state. This means that you should minimize the complexity of the lambdas used to create and/or modify state.

If the operating system crashes or the computer loses power, then the database will be left in an undefined state depending upon which memory pages that operating system was able to sync to disk.

ChainBase was designed to be used with blockchain applications where an append-only log of blocks is used to secure state in the event of power loss. This block log can be replayed to regenerate the full database state. Dealing with OS crashes, loss of power, and logs, is beyond the scope of ChainBase.

Portability

The contents of the database file is dependent upon the memory layout of the computer and process that created the database. Moving the database to a machine that uses a different compiler, operating system, libraries, or build type (release vs debug) will result in undefined behavior.

If portability is desired, the developer will have to export the database to a suitable format.

Background

Blockchain applications depend upon a high performance database capable of millions of read/write operations per second. Additionally blockchains operate on the basis of "eventually consistant" which means that any changes made to the database are potentially reversible for an unknown amount of time depending upon the consenus protocol used.

Existing database such as libbitcoin Database achieve high peformance using similar techniques (memory mapped files), but they are heavily specialised and do not implement the logic necessary for multiple indicies or undo history.

Databases such as LevelDB provide a simple Key/Value database, but suffer from poor performance relative to memory mapped file implementations.

About

Fast version controlled database based upon memory mapped containers

Resources

License

Stars

Watchers

Forks

Languages

  • C++ 95.0%
  • CMake 5.0%