-
Ole John Aske authored
ASSERTION: !USES_BLOB_VALUE(TABLE->READ_SET) IN MULTI_RANGE_START_RETRIEVALS Starting with MySQL-5.6-cluster-7.3.1, (bug 16020945) columns in 'read_set' required due to implementation specific details of each handler is not any more calculated (read: 'guessed') by the optimizer. Instead additional columns may be added to the read_set by each handler before the table access is init'ed. ha_ndbcluster::get_read_set() was introduced for this purpose. However, the native MRR implementation in Cluster called get_read_set() too late. Thus we failed to detect Blob colums being part of the read set, which should have forced is to use use the default MRR impl. instead of the native NDB MRR impl.. This caused us to later hit the assert mentioned in the bug synopsis
Ole John Aske authoredASSERTION: !USES_BLOB_VALUE(TABLE->READ_SET) IN MULTI_RANGE_START_RETRIEVALS Starting with MySQL-5.6-cluster-7.3.1, (bug 16020945) columns in 'read_set' required due to implementation specific details of each handler is not any more calculated (read: 'guessed') by the optimizer. Instead additional columns may be added to the read_set by each handler before the table access is init'ed. ha_ndbcluster::get_read_set() was introduced for this purpose. However, the native MRR implementation in Cluster called get_read_set() too late. Thus we failed to detect Blob colums being part of the read set, which should have forced is to use use the default MRR impl. instead of the native NDB MRR impl.. This caused us to later hit the assert mentioned in the bug synopsis
Loading