Using the JE JE Collections API

Using Transactions
Transaction Rollback
Access Method Restrictions

An Environment manages the resources for one or more data stores. A Database object represents a single database and is created via a method on the environment object. SecondaryDatabase objects represent an index associated with a primary database. Primary and secondary databases are then used to create stored collection objects, as described in Using Stored Collections .

Using Transactions

Once you have an environment, one or more databases, and one or more stored collections, you are ready to access (read and write) stored data. For a transactional environment, a transaction must be started before accessing data, and must be committed or aborted after access is complete. The JE JE Collections API provides several ways of managing transactions.

The recommended technique is to use the TransactionRunner class along with your own implementation of the TransactionWorker interface. TransactionRunner will call your TransactionWorker implementation class to perform the data access or work of the transaction. This technique has the following benefits:

  • Transaction exceptions will be handled transparently and retries will be performed when deadlocks are detected.

  • The transaction will automatically be committed if your TransactionWorker.doWork() method returns normally, or will be aborted if doWork() throws an exception.

  • TransactionRunner can be used for non-transactional environments as well, allowing you to write your application independently of the environment.

If you don't want to use TransactionRunner, the alternative is to use the CurrentTransaction class.

  1. Obtain a CurrentTransaction instance by calling the CurrentTransaction.getInstance method. The instance returned can be used by all threads in a program.

  2. Use CurrentTransaction.beginTransaction(), CurrentTransaction.commitTransaction() and CurrentTransaction.abortTransaction() to directly begin, commit and abort transactions.

If you choose to use CurrentTransaction directly you must handle the LockConflictException exception and perform retries yourself. Also note that CurrentTransaction may only be used in a transactional environment.

The JE JE Collections API supports transaction auto-commit. If no transaction is active and a write operation is requested for a transactional database, auto-commit is used automatically.

The JE JE Collections API also supports transaction dirty-read via the StoredCollections class. When dirty-read is enabled for a collection, data will be read that has been modified by another transaction but not committed. Using dirty-read can improve concurrency since reading will not wait for other transactions to complete. For a non-transactional container, dirty-read has no effect. See StoredCollections for how to create a dirty-read collection.

Transaction Rollback

When a transaction is aborted (or rolled back) the application is responsible for discarding references to any data objects that were modified during the transaction. Since the JE JE Collections API treats data by value, not by reference, neither the data objects nor the JE JE Collections API objects contain status information indicating whether the data objects are 1- in sync with the database, 2- dirty (contain changes that have not been written to the database), 3- stale (were read previously but have become out of sync with changes made to the database), or 4- contain changes that cannot be committed because of an aborted transaction.

For example, a given data object will reflect the current state of the database after reading it within a transaction. If the object is then modified it will be out of sync with the database. When the modified object is written to the database it will then be in sync again. But if the transaction is aborted the object will then be out of sync with the database. References to objects for aborted transactions should no longer be used. When these objects are needed later they should be read fresh from the database.

When an existing stored object is to be updated, special care should be taken to read the data, then modify it, and then write it to the database, all within a single transaction. If a stale data object (an object that was read previously but has since been changed in the database) is modified and then written to the database, database changes may be overwritten unintentionally.

When an application enforces rules about concurrent access to specific data objects or all data objects, the rules described here can be relaxed. For example, if the application knows that a certain object is only modified in one place, it may be able to reliably keep a current copy of that object. In that case, it is not necessary to reread the object before updating it. That said, if arbitrary concurrent access is to be supported, the safest approach is to always read data before modifying it within a single transaction.

Similar concerns apply to using data that may have become stale. If the application depends on current data, it should be read fresh from the database just before it is used.

Access Method Restrictions

The BTREE access method is always used for JE Databases. Sorted duplicates — more then one record for a single key — are optional.

The restrictions imposed by the access method on the database model are:

  • If duplicates are allowed then more than one value may be associated with the same key. This means that the data store cannot be strictly considered a map — it is really a multi-map. See Using Stored Collections for implications on the use of the collection interfaces.

  • If duplicate keys are allowed for a data store then the data store may not have secondary indices.

  • With sorted duplicates, all values for the same key must be distinct.

See Using Stored Collections for more information on how access methods impact the use of stored collections.