CHECK and FOREIGN KEY constraints are not supported.

VoltDB does not support AUTO_INCREMENT, the automatic incrementing of column values.

A table can have up to 1024 columns. Each column has a maximum size of 1 megabyte and the total declared size of all of the columns in a table cannot exceed 2 megabytes. For VARCHAR columns where the length is specified in characters, the declared size is calculated as 4 bytes per character to allow for the longest potential UTF-8 string.

If you intend to use a column to partition a table, that column cannot contain null values. You must specify NOT NULL in the definition of the column or VoltDB issues an error when compiling the schema.

To specify an index — either for an individual column or as a table constraint — that is globally unique across the database, use the standard SQL keywords UNIQUE and PRIMARY KEY. However, for partitioned tables, VoltDB can only ensure uniqueness if the index includes the partitioning column. Otherwise, these keywords are not allowed.

It can be a performance advantage to define indexes or constraints on non-partitioning columns that you, as the developer, know are going to contain unique values. Although VoltDB cannot ensure uniqueness across the entire database, it does allow you to define indexes that are assumed to be unique by using the ASSUMEUNIQUE keyword.

When you define an index on a partitioned table as ASSUMEUNIQUE, VoltDB verifies uniqueness within the current partition when creating an index entry. However, it is your responsibility as developer or administrator to ensure that the values are actually globally unique. If the database is repartitioned due to adding new nodes or restoring a snapshot to a different cluster configuration, non-unique ASSUMEUNIQUE index entries may collide. When this occurs it results in a constraint violation error and the database will not be able to complete its current action.

Therefore, ASSUMEUNIQUE should be used with caution. Also, it is not necessary and should not be used with replicated tables or indexes that contain the partitioning column, which can be defined as UNIQUE.

EXPORT TO TARGET and EXPORT TO TOPIC allow you to connect a table to either an export target or a topic, so that by default data written into the table is also sent to the topic or the export connector for delivery to the specified target. By default, only insert operations ( INSERT and UPSERT when it inserts a new row) initiate export records. However, you can use the ON clause to specify which actions you want to trigger export. For example, the following table declaration generates export records whenever rows are created or modified.

CREATE TABLE RESERVATION
   EXPORT TO TARGET airlines ON INSERT, UPDATE_NEW
     (reserv_id INT NOT NULL,
      flight_id INT NOT NULL,
      . . . );

The following table defines the actions that you can specify in the ON clause.

EXPORT TO TOPIC (and MIGRATE TO TOPIC) also allow you to specify which column values are used as the topic key and which as the body of the topic message using the WITH KEY/VALUE clause. See the description of topic streams in the description of CREATE STREAM for more information.

The length of VARCHAR columns can be specified in either characters (the default) or bytes. To specify the length in bytes, include the BYTES keyword after the length value; for example VARCHAR(16 BYTES).

Specifying the VARCHAR length in characters is recommended because UTF-8 characters can require a variable number of bytes to store. By specifying the length in characters you can be sure the column has sufficient space to store any string of the specified length. Specifying the length in bytes is only recommended when all values contain only single byte (ASCII) characters or when conserving space is required and the strings are less than 64 bytes in length.

The VARBINARY datatype provides variable storage for arbitrary strings of binary data and operates similarly to VARCHAR(n BYTES) strings. You assign byte arrays to a VARBINARY column when passing in variables, or you can use a hexidecimal string for assigning literal values in the SQL statement.

The VoltDB TIMESTAMP datatype is a long integer representing the number of microseconds since the epoch. Two important points to note about this timestamp:

For TIMESTAMP columns, you can define a default value using the NOW or CURRENT_TIMESTAMP keywords in place of a specific value. For example:

CREATE TABLE Event (
    Event_Id INTEGER UNIQUE NOT NULL,
    Event_Timestamp TIMESTAMP DEFAULT NOW,
    Event_Description VARCHAR(128)
);

The default value is evaluated at runtime as an approximation, in milliseconds, of when the transaction begins execution.

There must be a usable index on the TTL column for the table. VoltDB uses that index to optimize the evaluation of the TTL values. If not, the USING TTL clause is accepted, but no automated deletion will occur at runtime until a usable index is defined.

The CREATE TABLE... USING TTL statement is not rejected if the index is missing. This way you can define the index in a subsequent DDL statement. However, a warning message is issued if the USING TTL clause has no supporting index available. A similar warning is issued if you delete the last usable index.

When the table definition includes both USING TTL and MIGRATE TO TARGET or MIGRATE TO TOPIC, there must be an index including the TTL column for the USING TTL clause and a separate index including only the TTL column and a WHERE NOT MIGRATING clause. This index is required to effectively find and schedule the migration of expired records. For example, the sessions table in the previous example would require the following index. If the index is not present, records for the table will neither be deleted nor migrated and a warning will be logged on the server:

CREATE INDEX sessions_migrate_index ON sessions 
   (last_update) WHERE NOT MIGRATING;

TTL clauses are most effective when used on partitioned tables. Defining TTL for a replicated table, especially a large replicated table, can have a significant impact on database performance because the TTL delete actions must be processed as multi-partition transactions.

You can specify the frequency and maximum size of the TTL processing cycle.

Under extreme loads or sudden bursts of inserts, it is possible for TTL processing to fall behind. Or if the records are extremely large, attempting to delete too many records at one time can cause the TTL process to exceed the temporary table limit. The BATCH_SIZE and MAX_FREQUENCY clauses let you customize the TTL processing per table to meet the specific requirements of your application. The TTL selector for the @Statistics system procedure can help you evaluate TTL performance against your application workload to determine what settings you need.

Evaluation of the time to live is made against the current value of the TTL column, not its initial value. So if a subsequent transaction alters the column value (either increasing or decreasing it) that modification will impact the subsequent lifetime of the record.

When using database replication (DR), it is possible for the TTL transaction to exceed the 50MB limit on the DR binary log. If this happens, a warning is issued and TTL processing is suspended.

When using MIGRATION TO TARGET or MIGRATE TO TOPIC, there is an interval after the TTL value is triggered and before the record is successfully migrated and deleted from the VoltDB database. During this interval, the record is available for read access from SELECT queries. You can also update or delete the record; but modifying the record will cancel the pending delete. So if, for example, you update the record to extend the TTL column, the record will remain in the database until the new TTL column value is reached. However the update does not cancel the export of the original data to the specified target or topic that had already been triggered. So two records will eventually be migrated.

In most cases, you can ignore whether a record is currently being migrated and scheduled for delete or not. For example, if you delete a record that is currently being migrated, you cancel the pending delete but you delete the record anyway, so the results end up the same. However, if you do want to distinguish between currently active and currently migrating records, you can use the MIGRATING function, that identifies records that are currently "in flight". For example, to select records for a specific user ID and only those records that are not being migrated, you can use the following query:

SELECT user_id, login FROM sessions WHERE user_id = ? AND NOT MIGRATING;