fix 1:#818 Master slave synchronization - There will be too many tuples problem
Cause:
Tianmu::dbhandler::TianmuHandler::current_position This variable is not initialized. In some cases,
a large value may cause too many tuples problem.
Solution: Initialize the variable
fix 2:#819 Master slave synchronization - Primary key conflict problem
Solution: Modify the modification logic of master slave synchronization, so that the delete and update operations do not follow the primary key logic
Supplement
sql/sql_insert.cc
Fix the problem that the insert statement will not generate binlog in the delayed insert mode
storage/tianmu/handler/tianmu_handler.cpp
Fix the binlog error of the line format generated by the tianmu engine
The root cause is that the tianmu syntax tree conversion
cannot support continuous null-value elimination,
and the cleaned equal-value elimination items are retained to be compatible with the tianmu execution plan
The cause of an error is
that you need to identify the returned
result of the exists subquery for sl->join->zero_result_cause
Determines whether a return value exists.
and format query.cpp
:
The exists subquery determines whether a value exists during the query optimization phase
result is not set to zero only when a matching value is found in the query optimization phase
When a field has an index, the optimization phase scans the table through the index
The primary key implementation of the current column storage engine has a problem with the primary key index to scan the table for data
As a result, the primary key index is used to obtain the exists subquery result in the query optimization phase. Therefore, the exists subquery is not executed in the query execution phase
scan the table for data Remove the following temporary practices after primary key indexing is complete
replace always true
for examples:
select * from t1 where b>2 or 1=2;
select * from t1 where b>2 or 1<2;
select * from t1 where b>2 or 1>2;
select * from t1 where b>2 or 1=1;
select * from t1 where b>2 or 1;
select * from t1 where b>2 or 0;
select * from t1 where ((((1=1) or (3=3)) and (3=3)) or ((1=100) and (5=a)));
select * from t1 where ((((1=1) or (3=3)) and (3=4)) or ((1=100) and (5=a)));
select * from t1 where ((((1=2) or (3=3)) and (3=3)) or ((1=100) and (5=a)));
select * from t1 where ((((1>2) or (3=3)) and (3=3)) or ((1=1) and (5=a)));
select * from t1 where ((((1=2) or (3<3)) and (3=3)) or ((1=1) and (5=a)));
select * from t1 where ((((1=2) or (3<3)) and (3>1)) or ((1=1) and (5=a)));
select * from t1 where ((((1=2) or (3=3)) and (3=1)) or ((1>1) and (5=a)));
Cause:
At present, the primary key of the tianmu engine does not fully support delete and update statements.
Temporary solution:
Modify the execution plan of the SQL layer so that the delete and update statements do not follow the primary key logic
The reason for the error lies in that in Item_cond_or condition,
sub-conditions similar to a=a will be cleared after
elimination of equivalent conditions,
but the tianmu engine does not do additional scene
processing for the remaining conditions after elimination
of equivalent conditions of Item_cond_or.
Results in a scenario where
only the remaining predicates are processed
and the or predicates are not considered.
The corresponding issuse:
https://github.com/stoneatom/stonedb/issues/301https://github.com/stoneatom/stonedb/issues/733
revert code style
adofsauron