RandomQueryGeneratorPartitioning
Contents |
[edit] Generating Partitioned Tables
It is possible to generate partitioned tables from a .ZZ file, as follows:
$tables = {
rows => [10],
partitions => [ undef , 'KEY (pk) PARTITIONS 2' ]
};
$fields = {
types => [ 'int' ],
indexes => [undef, 'key']
};
When the tables are generated, the PARTITION BY clause will be wrapped in /*!50100 */ comments. This allows for comparison tests, where one of the servers is a 5.0. The tables will be created without partitions on such a server.
[edit] Existing Partitioning Tests
[edit] Comparison Tests
It is possible to compare the result set returned by a query on a partitioned table against the result set returned by a non-partitioned table having the same structure. For example:
perl runall.pl \ --grammar=conf/partitions-ddl.yy \ --basedir1=/path/to/mysql-with-partitions \ --basedir2=/path/to/mysql-without-partitions \ --threads=1
[edit] Partition DDL and Key Cache
The conf/partitions-ddl.yy grammar contains a wide range of partitioning-related DDL statements, along with operations on the key cache and assigning a particular key cache to a particular partition.
No Validator is available to go along with this grammar (except a possible comparison between two servers), so this test would catch mostly deadlocks and crashes/assertions.
When using partitions-ddl.yy for comparison tests, please note that this grammar includes DROP PARTITION and TRUNCATE PARTITION which have no corresponding query against a non-partition table, and thus will cause the two servers to diverge if one does not support partitions.
[edit] Partition Pruning
A specific test is TBD for testing the accuracy of partition pruning. Such a test would target complex/creative WHERE clauses.