-
Mattias Jonsson authored
Implementing partitioning natively in InnoDB, so it does not need ha_partition generic partition engine for supporting partitioning. ha_partition uses one ha_innobase handler for each partition which does not share common data with each other resulting in high resource usage which this WL fix. It also makes it easier to support other InnoDB features that partitioning currently does not support. A new handler ha_innopart is added, inheriting both ha_innobase (for InnoDB access) and Partition_helper (for partitioning support, see wl#4807). And to avoid a proxy object for Partition_handler (see wl#4807) it is also inherited. Also ha_partition is changed to inherit Partition_handler directly. As a result of this there is no longer any need for .par files for partitioned InnoDB tables, since InnoDB can use its internal data dictionary for finding partitions during rename and delete. The optimizer estimate is also changed, especially for records_in_range, where all used partitions are checked instead of only the biggest ones. rb#7636 Approved by Krunal and Annamalai
Mattias Jonsson authoredImplementing partitioning natively in InnoDB, so it does not need ha_partition generic partition engine for supporting partitioning. ha_partition uses one ha_innobase handler for each partition which does not share common data with each other resulting in high resource usage which this WL fix. It also makes it easier to support other InnoDB features that partitioning currently does not support. A new handler ha_innopart is added, inheriting both ha_innobase (for InnoDB access) and Partition_helper (for partitioning support, see wl#4807). And to avoid a proxy object for Partition_handler (see wl#4807) it is also inherited. Also ha_partition is changed to inherit Partition_handler directly. As a result of this there is no longer any need for .par files for partitioned InnoDB tables, since InnoDB can use its internal data dictionary for finding partitions during rename and delete. The optimizer estimate is also changed, especially for records_in_range, where all used partitions are checked instead of only the biggest ones. rb#7636 Approved by Krunal and Annamalai
Loading