-
Dmitry Lenev authored
Alternative fix for #22811659 "ACCESSES TO NEW DATA-DICTIONARY ADD CONFUSING STAGES TO P_S.EVENTS_STAGES_*." Also fixes failure of of group_replication.gr_majority_loss_restart which popped-up after WL#7743 "New data dictionary: changes to DDL-related parts of SE API" was implemented. Do not start "Opening/Locking system tables" stages when we are opening/ locking system tables. Stages are supposed to be coarse phases of statement execution which are useful for users to know about (either because they involve lot of work or some waiting). OTOH, opening/locking system tables are operations which are normally executed immediately and typically as part of coarser phase of statement execution. By entering these stages we will either wipe out or split information about coarser/higher-level stage (there is no concept of sub-stage yet) without providing much additional value to users. Therefore this patch removes these two stages which were introduced by original fix for #22811659. This also solves problem with group_replication.gr_majority_loss_restart where access to system tables wiped out information about high-level "After create" phase, during which commit of CREATE TABLE statement happens after WL#7743 and which name is used by the test to detect that CREATE TABLE is blocked during commit.
Dmitry Lenev authoredAlternative fix for #22811659 "ACCESSES TO NEW DATA-DICTIONARY ADD CONFUSING STAGES TO P_S.EVENTS_STAGES_*." Also fixes failure of of group_replication.gr_majority_loss_restart which popped-up after WL#7743 "New data dictionary: changes to DDL-related parts of SE API" was implemented. Do not start "Opening/Locking system tables" stages when we are opening/ locking system tables. Stages are supposed to be coarse phases of statement execution which are useful for users to know about (either because they involve lot of work or some waiting). OTOH, opening/locking system tables are operations which are normally executed immediately and typically as part of coarser phase of statement execution. By entering these stages we will either wipe out or split information about coarser/higher-level stage (there is no concept of sub-stage yet) without providing much additional value to users. Therefore this patch removes these two stages which were introduced by original fix for #22811659. This also solves problem with group_replication.gr_majority_loss_restart where access to system tables wiped out information about high-level "After create" phase, during which commit of CREATE TABLE statement happens after WL#7743 and which name is used by the test to detect that CREATE TABLE is blocked during commit.
Loading