MDEV-31541: Optimize calls to build_table_filenames for CREATE TABLE

[an3l]

• In order to optimize the call of ha_table_exists() for CREATE statement that for single statement has 2 paths from mysql_create_table(() that are called sequentually:

sequenceDiagram
    box create table query
        participant mysql_create_table as mysql_create_table
        participant open_and_lock_tables as open&lock tbl
        participant open_tables as open tbl
        participant lock_table_names as lock tbl names
        participant upgrade_lock_if_not_exists as upgrade lock if not exists
        participant ha_table_exists as ha_table_exists
        participant mysql_create_table_no_lock as create without lock
        participant build_table_filename as build tbl filename
        participant create_table_impl as implicit create
    end
    autonumber
    Note left of mysql_create_table:Query
    mysql_create_table->>+open_and_lock_tables: 
    open_and_lock_tables->>open_tables: 
    open_tables->>lock_table_names: 
    lock_table_names->>+upgrade_lock_if_not_exists: 
    upgrade_lock_if_not_exists->>+ha_table_exists: 
    Note over upgrade_lock_if_not_exists, ha_table_exists: pdate HA_CREATE_INFO <br>with table_path and table_exists <br> call here <br> build_table_filename
    ha_table_exists->>+build_table_filename: Optimized
    activate build_table_filename
    deactivate build_table_filename
    build_table_filename-->>ha_table_exists: 
    ha_table_exists-->>-mysql_create_table: Return to caller
    Note over ha_table_exists, mysql_create_table: Path 1
    mysql_create_table->>+mysql_create_table_no_lock: 
    mysql_create_table_no_lock->>+build_table_filename: Optimized
    Note over mysql_create_table_no_lock, build_table_filename: Use HA_CRETE_INFO struct
    activate build_table_filename
    deactivate build_table_filename
    mysql_create_table_no_lock->>create_table_impl: 
    create_table_impl--xha_table_exists: Optimized (no call)
    Note over create_table_impl, ha_table_exists: Use HA_CRETE_INFO struct
    ha_table_exists--x+build_table_filename: Optimized (no call)
    create_table_impl-->+mysql_create_table: Return to caller
    Note over create_table_impl, mysql_create_table: Path 2

• We are creating the new members of HA_CREATE_INFO that are updated after first call of ha_table_exists() and used in the place where second invocation of function was done.

  • Handling of case for CREATE ...SELECT FROM is added too with early call of ha_table_exists instead from ha_create.
  • Since or replace [like] , if not exists and lock tables, have different paths invocation that are different from Path 1, we are leaving the ha_table_exists() call for that cases, without optimization.

• To optimize build_table_filename call is done in Path 1 from upgrade_lock_if_not_exists by saving the path and length, that is reused in ha_table_exists(), while there is special handling for views that still need to call the function from within ha_table_exists(). Optimization is done by eliminating the call for Path 2.

  • Special cases are replace or if not exists ddl that do not use Path 1 but Path2 only, without optimization
  • Special case is CREATE..SELECT without optimization.

Reviewer: [email protected]

• [x] The Jira issue number for this PR is: MDEV-31541

Description

How can this PR be tested?

I have tested the PR using debug traces and noted optimization of calls for CREATE TABLE in ha_table_exists(), as well as build_table_filenames() by comparing the trace files with/without PR. Here is the test case

--echo # 
--echo # MDEV-31541: Optimize calls to build_table_filenames for CREATE TABLE
--echo # 

--source include/have_debug.inc
--source include/have_debug_sync.inc
select @@debug_dbug;
set @old_debug= @@debug_dbug;
select @old_debug;
set debug_dbug= "d:t:i:o,/tmp/anel.trace";
select @@debug_dbug;
#set debug_sync='create_table_before_check_if_exists SIGNAL parked WAIT_FOR go';
Create table t(t int);
drop table t;

Basing the PR against the correct MariaDB version

• [ x] This is a new feature and the PR is based against the latest MariaDB development branch.
• [ ] This is a bug fix and the PR is based against the earliest maintained branch in which the bug can be reproduced.

PR quality check

• [ x] I checked the CODING_STANDARDS.md file and my PR conforms to this where appropriate.
• [ ] For any trivial modifications to the PR, I am ok with the reviewer making the changes themselves.