Archive
Archive
As you can see from Chapter 4, Designing the Database Schema, defining the database schema and the partitioning plan go hand in hand with understanding how the data is accessed. The two must be coordinated to ensure optimum performance. Your stored procedures must use the same attribute for partitioning as the table being accessed. Proper partitioning ensures that the table rows the stored procedure requests are in the same partition in which the procedure executes, thereby ensuring maximum efficiency.
It doesn't matter whether you design the partitioning first or the data access first, as long as in the end they work together. However, for the sake of example, we will use the schema and partitioning outlined in Chapter 4, Designing the Database Schema when discussing how to design the data access.
The key to designing the data access for VoltDB applications is that complex or performance sensitive access to the database should be done through stored procedures. It is possible to perform ad hoc queries on a VoltDB database. However, ad hoc queries do not benefit as fully from the performance optimizations VoltDB specializes in and therefore should not be used for frequent, repetitive, or complex transactions.
Within the stored procedure, you access the database using standard SQL syntax, with statements such as SELECT, UPDATE, INSERT, and DELETE. You can also include your own code within the stored procedure to perform calculations on the returned values, to evaluate and execute conditional statements, or to perform many other functions your applications may need.
In VoltDB, a stored procedure and a transaction are one and the same. Thus when you define a stored procedure, VoltDB automatically provides ACID transaction guarantees for the stored procedure. This means that stored procedures fully succeed or automatically roll back as a whole if an error occurs (atomic). When stored procedures change the data, the database is guaranteed to remain consistent. Stored procedures execute and access the database completely isolated from each other, including when they execute concurrently. Finally, stored procedure changes to the database are guaranteed to be saved and available for subsequent database access (durable).
Because the transaction is defined in advance as a stored procedure, there is no need for your application to manage transactions using specific transaction commands such as BEGIN, ROLLBACK, COMMIT or END.[1]
To ensure data consistency and durability, VoltDB procedures must be deterministic. That is, given specific input values, the outcome of the procedure is consistent and predictable. Determinism is critical because it allows the same stored procedure to run in multiple locations and give the same results. It is determinism that makes it possible to run redundant copies of the database partitions without impacting performance. (See Chapter 10, Availability for more information on redundancy and availability.)
One key to deterministic behavior is avoiding ambiguous SQL queries in stored procedures. Specifically, performing
unsorted queries can result in a non-deterministic outcome. VoltDB does not guarantee a consistent order of results
unless you use a tree index to scan the records in a specific order or you specify an ORDER BY clause in the query
itself. In the worst case, a limiting query, such as SELECT TOP 10 Emp_ID FROM Employees without an
index or ORDER BY clause, can result in a different set of rows being returned. However, even a simple query such as
SELECT * from Employees can return the same rows in a different order.
The problem is that even if a non-deterministic query is read-only, its results might be used as input to an INSERT, UPDATE, or DELETE statement elsewhere in the stored procedure. For clusters with a K-safety value greater than zero, this means unsorted query results returned by two copies of the same partition, which may not match, could be used for separate update queries. If this happens, VoltDB detects the mismatch, reports it as a potential source of data corruption, and shuts down all but one copy of each partition.
By switching to this reduced K-safety mode, VoltDB avoids the threat of data corruption due to non-determinism. However, it also means that the cluster is no longer K-safe; there is only one copy of each partition and any node failure will crash the database. So, although the database continues to operate after a mismatch, it is critically important you determine the cause of the non-deterministic behavior, correct the affected procedures, take a final snapshot, and restart the database to restore full K-safety.
The risk of mismatched results at run time is why VoltDB issues a warning for any non-deterministic queries in read-write stored procedures when you load the schema or classes. This is also why use of an ORDER BY clause or a tree index in the WHERE constraint is strongly recommended for all SELECT statements.
Another key to deterministic behavior is avoiding calls within your stored procedures to external functions or procedures that can introduce arbitrary data. External functions include file and network I/O (which should be avoided any way because they can impact latency), as well as many common system-specific procedures such as Date and Time.
However, this limitation does not mean you cannot use arbitrary data in VoltDB stored procedures. It just means you must either generate the arbitrary data before the stored procedure call and pass it in as input parameters or generate it in a deterministic way. For example, if you need to load a set of records from a file, you can open the file in your application and pass each row of data to a stored procedure that loads the data into the VoltDB database. This is the best method when retrieving arbitrary data from sources (such as files or network resources) that would impact latency.
The other alternative is to use data that can be generated deterministically. For two of the most common cases, timestamps and random values, VoltDB provides methods for this:
VoltProcedure.getTransactionTime() returns a timestamp that can be used in place of the
Java Date or Time classes.
VoltProcedure.getSeededRandomNumberGenerator() returns a pseudo random number that can be
used in place of the Java Util.Random class.
These procedures use the current transaction ID to generate a deterministic value for the timestamp and the random number. See the VoltDB Java Stored Procedure API for more.
Finally, even seemingly harmless programming techniques, such as static variables can introduce nondeterministic behavior. VoltDB provides no guarantees concerning the state of the stored procedure class instance across invocations. Any information that you want to persist across invocations must either be stored in the database itself or passed into the stored procedure as a parameter.
[1] One side effect of transactions being precompiled as stored procedures is that external transaction management frameworks, such as Spring or JEE, are not supported by VoltDB.
V9.x Documentation