Cassandra storage engine
This page is a documentation for Cassandra storage engine which is a work in progress.
This page describes a feature that's under development. The feature has not been released (even in beta), its interface and function may change, etc.
Cassandra table
MariaDB's table represents a column family in Cassandra. The table must follow this pattern:
create table cassandra_tbl -- name can be chosen at will ( rowkey char(N), -- First field must be named 'rowkey'. This -- is what Cassandra's row key is mapped to.type must be the same as for column1 varchar(N), -- columns go here ... PRIMARY KEY(rowkey) -- Primary key over 'rowkey' is mandatory ) engine=cassandra thrift_host='192.168.1.0' keyspace= 'cassandra_key_space' column_family='column_family_name';
MariaDB's table is a view of Cassandra's column family.
- column family must exist before the MariaDB table can be created
- dropping MariaDB's table will not drop the column family.
Mapping Cassandra columns to SQL
Cassandra has three kinds of columns
- The row key. It is not considered to be a column by Cassandra, but MySQL/MariaDB can only store data in columns, so we consider it a special kind of column.
- Columns that were present in static column family declaration. These columns have a defined name/data type.
- "Ad-hoc" columns that can be encountered in individual rows.
note: Cassandra's supercolumns are not handled at the moment. If we need to handle them, we'll make them dynamic-column blob. Cassandra's counter columns will be handled, and will be mapped to regular SQL columns.
When these three kinds of columns are mapped to SQL:
- The row key is mapped into a regular column named
rowkey
- Static column family members are mapped to regular SQL columns (i.e, column named 'foo1' in Cassandra will map to column named 'foo' in SQL. If SQL's CREATE TABLE doesn't match cassandra's columns, a warning is issued.
- Ad-hoc columns are all put into a regular SQL column which has a Dynamic Columns blob. This allows us to return arbitrary sets of columns within one row.
Column datatypes
There is no direct 1-to-1 mapping between Cassandra data types and MySQL datatypes. Also, Cassandra's limitations are generally more lax than MySQL/MariaDB's: for example, Cassandra's limit on rowkey is about 2G, while MySQL's is about 1.5Kb.
Currently, we use this mapping:
Cassandra | MariaDB |
---|---|
blob | VARBINARY(n) |
ascii | VARCHAR(n) .. use charset=latin1 |
text | VARCHAR(n) .. use charset=utf8 |
varint | BIGINT, VARCHAR(N) |
int | INT |
bigint | BIGINT |
uuid, | ?? todo: figure out thrift representation?? |
timestamp | DATETIME, TIMESTAMP |
boolean | BOOL, BOOLEAN |
float | FLOAT |
double | DOUBLE |
decimal | ?? todo: figure out thrift representation?? |
counter | BIGINT |
{ Hence, one must responsibly pick datatypes himself }
{ TODO: equality (and ordering?) relationship for SQL's `rowkey` must be the same as for Cassandra's primary key }
{TODO: put here a suggested mapping for Cassandra's datatypes}
TODO
More details here.
See also
- MDEV-431 - jira task for the cassandra storage engine
- cassandra-storage-engine-future-plans
- hbase-storage-engine