Bug #16869534 QUERYING SUBSET OF COLUMNS DOESN'T USE TABLE CACHE; OPENED_TABLES I
Description: When querying a subset of columns from the information_schema.TABLES Analysis: When information about tables is collected for statements like "SELECT ENGINE FROM I_S.TABLES" we do not perform full-blown table opens in SE, instead we only use information from table shares from the Table Definition Cache or .FRMs. Still in order to simplify I_S implementation mock TABLE objects are created from TABLE_SHARE during this process. This is done by calling open_table_from_share() function with special arguments. Since this function always increments "Opened_tables" counter, calls to it can be mistakingly interpreted as full-blown table opens in SE. Note that claim that "'SELECT ENGINE FROM I_S.TABLES' statement doesn't use Table Cache" is nevertheless factually correct. But it misses the point, since such statements a) don't use full-blown TABLE objects and therefore don't do table opens b) still use Table Definition Cache. Fix: We are now incrementing the counter when db_stat(i.e open flags for ha_open( we have considered an optimization which would use TABLE objects from Table Cache when available instead of constructing mock TABLE objects, but found it too intrusive for stable releases.
Showing
Please register or sign in to comment