Fix update duplication after crashfix #69
Conversation
Issue: with the fix in Percona-Lab#67, pgtde decrypts tuples during update into a new memory region, and changes the t_data pointer to this new region. Because of this, later updates to tuple flags also happen in the new data, and the original persisted tuple flags are never updated. Fix: after the update statement is done with the decrypted data, restore the t_data pointer to the original. This way, flag changes happen where they should. Fixes Percona-Lab#68
src/access/pg_tdeam.c
Outdated
| @@ -3012,6 +3012,7 @@ pg_tde_update(Relation relation, ItemPointer otid, HeapTuple newtup, | |||
| ItemId lp; | |||
| HeapTupleData oldtup; | |||
| HeapTupleData oldtup2; | |||
| void* oldtupptr; | |||
There was a problem hiding this comment.
Tiny suggestion. I think we should use more descriptive name for oldtupptr (may be oldtup_data). since the variable refers to the tuple data instead of the tuple. Similarly not related to this particular PR, but oldtup2 may be renamed as decrypted_tup or something to make it easier to understand when we revisit the code in the future.
There was a problem hiding this comment.
Not related to this particular PR. But I figured out we are copying the complete tuple data in new_ptr in the destination tuple before decrypting the data in that.
/* decrypt the old tuple */
{
char* new_ptr = NULL;
new_ptr = MemoryContextAlloc(CurTransactionContext, oldtup.t_len);
memcpy(new_ptr, oldtup.t_data, oldtup.t_len); <==== Here
// only neccessary field
oldtup2.t_data = (HeapTupleHeader)new_ptr;
}
This is fine for small size tuples, but for large tuples we can gain some performance by saving the memory copy by only copying the tuple header ( since that is all what we need for destination tuple)
/* decrypt the old tuple */
{
char* new_ptr = NULL;
new_ptr = MemoryContextAlloc(CurTransactionContext, oldtup.t_len);
memcpy(new_ptr, oldtup.t_data, oldtup.t_data->t_hoff); <==== Here
// only neccessary field
oldtup2.t_data = (HeapTupleHeader)new_ptr;
}
Issue: with the fix in #67, pgtde decrypts tuples during update into a new memory region, and changes the t_data pointer to this new region.
Because of this, later updates to tuple flags also happen in the new data, and the original persisted tuple flags are never updated.
Fix: after the update statement is done with the decrypted data, restore the t_data pointer to the original. This way, flag changes happen where they should.
Fixes #68