MDEV-39548 Cleanup MDL_request boilerplate with ACQUIRE_LOCK macro by longjinvan · Pull Request #5075 · MariaDB/server

longjinvan · 2026-05-13T23:11:48Z

Description

MDL (Metadata Lock) is MariaDB's locking subsystem that protects database objects (tables, schemas, stored procedures, etc.) from concurrent DDL — for example, preventing a table from being dropped while another thread is reading from it. MDL_request is the "lock request form" that callers must create, initialize, submit, and then extract the resulting MDL_ticket from.

Currently, every MDL lock acquisition requires repetitive boilerplate:

> MDL_request mdl_request;
> MDL_REQUEST_INIT(&mdl_request, namespace, db, object, lock_type, duration);
> if (mdl_context.acquire_lock(&mdl_request, lock_wait_timeout))
>   /* error */;
> ticket = mdl_request.ticket;

This patch introduces MDL_context::ACQUIRE_LOCK() — a set of overloaded methods (wrapped in a macro for FILE/LINE tracking) that combine the above steps into a single call returning MDL_ticket* directly:

> if (!(ticket = mdl_context.ACQUIRE_LOCK(namespace, db, object, lock_type, lock_wait_timeout)))
>   /* error */;
>

This is a step toward making MDL_request a private implementation detail of the MDL subsystem, reducing its exposure to callers and preventing the class of bugs where stale or misused MDL_request objects cause issues (MDEV-39241, MDEV-39184).

Release Notes

N/A

How can this PR be tested?

All MTR tests pass, confirming no regressions introduced by this refactoring.

Basing the PR against the correct MariaDB version

This is a refactoring change, and the PR is based against the latest MariaDB development branch.

Copyright

All new code of the whole pull request, including one or several files that are either new files or modified ones, are contributed under the BSD-new license. I am contributing on behalf of my employer Amazon Web Services, Inc.

gemini-code-assist

Code Review

This pull request streamlines the process of acquiring Metadata Locks (MDL) by adding new acquire_lock methods to the MDL_context class and introducing ACQUIRE_LOCK and ACQUIRE_LOCK_BY_KEY macros. These additions allow for more concise code by removing the need to manually declare and initialize MDL_request objects in many parts of the server, such as in backup operations, DDL logging, and table handling. The reviewer recommended adding an MDL_ prefix to the new macros to ensure consistency with existing naming patterns and to prevent potential namespace conflicts.

gkodinov

Thank you for your contribution! This is a preliminary review.

Nice cleanup. Some more changes and it will be perfect.

svoj

Looks very nice, a few comments inline.

svoj · 2026-05-14T18:29:14Z

+            MDL_SHARED_NO_WRITE, thd->variables.lock_wait_timeout,
+            MDL_TRANSACTION)))
      goto exit;
-    table->mdl_ticket= tl->mdl_request.ticket;


Dunno, I'd avoid this change as we can't trivially verify that subsequently executed code doesn't rely on initialized tl->mdl_request. Or did you verify this?

Reverted — keeping the original MDL_REQUEST_INIT + acquire_lock pattern here since subsequent code may rely on tl->mdl_request being initialized.

I don't see this reverted. Please double check.

Sorry about that — I forgot to copy the revert from my test workspace to the submission package. Fixed now.

gkodinov

LGTM. Please keep working on it for the final review.

svoj

There're a few more cases that can be updated:

dict_acquire_mdl_share() - replace both acquire and try_acquire
backup_lock() it is not right that sql_yacc.yy makes decision which lock type to use, change to MDL_ACQIRE_LOCK() please
Item_func_get_lock::val_int() - pre-initialize ull_key instead of ull_request like in Item_func_release_lock::val_int()
open_and_lock_for_insert_delayed() - protection_request can be replaced
trx_purge_table_acquire()
lock_table_names() - global_request can be moved

svoj · 2026-05-15T11:56:24Z

-                                      thd->variables.lock_wait_timeout))
+    if (!(mdl_ticket= thd->mdl_context.MDL_ACQUIRE_LOCK(
+            MDL_key::BACKUP, "", "", MDL_BACKUP_COMMIT,
+            MDL_TRANSACTION, thd->variables.lock_wait_timeout)))


You don’t need to use mdl_ticket here. Checking the result of MDL_ACQUIRE_LOCK() is sufficient, as in the original code.

Fixed. Now just checks the return value without assigning to mdl_ticket.

longjinvan · 2026-05-15T17:08:22Z

There're a few more cases that can be updated:

dict_acquire_mdl_share() - replace both acquire and try_acquire

backup_lock() it is not right that sql_yacc.yy makes decision which lock type to use, change to MDL_ACQIRE_LOCK() please

Item_func_get_lock::val_int() - pre-initialize ull_key instead of ull_request like in Item_func_release_lock::val_int()

open_and_lock_for_insert_delayed() - protection_request can be replaced

trx_purge_table_acquire()

lock_table_names() - global_request can be moved

Done. Updated in the latest revision:

dict_acquire_mdl_shared() — split trylock/blocking paths; trylock uses a new try_acquire_lock() overload, blocking uses MDL_ACQUIRE_LOCK().
backup_lock() — now uses MDL_ACQUIRE_LOCK() directly with MDL_SHARED_HIGH_PRIO, no longer relies on the lock type set in sql_yacc.yy.
Item_func_get_lock::val_int() — pre-initializes MDL_key ull_key (same pattern as Item_func_release_lock::val_int()), then uses MDL_ACQUIRE_LOCK_BY_KEY().
open_and_lock_for_insert_delayed() — protection_request replaced with MDL_ACQUIRE_LOCK(); delayed_get_table() now takes MDL_ticket* directly.
trx_purge_table_acquire() — uses the new try_acquire_lock() overload, MDL_request removed.
lock_table_names() — global_request removed; loop now uses try_acquire_lock() overload and MDL_ACQUIRE_LOCK() inline.

svoj · 2026-05-19T16:44:52Z

@longjinvan no merge commits please, makes less convenient to work with your branch. Please revert.

svoj

A few comments inline.

svoj · 2026-05-17T19:01:59Z

+MDL_context::try_acquire_lock(MDL_key::enum_mdl_namespace mdl_namespace,
+                              const char *db, const char *name,
+                              enum_mdl_type mdl_type,
+                              enum_mdl_duration mdl_duration)


You've missed src_file and src_line.

Fixed. Added src_file/src_line parameters to the try_acquire_lock overload and introduced MDL_TRY_ACQUIRE_LOCK macro to pass them from the caller site.

svoj · 2026-05-20T10:23:21Z

-                   MDL_SHARED_NO_WRITE, MDL_EXPLICIT);
-  MDL_key *ull_key= &ull_request.key;
+  MDL_key ull_key;
+  ull_key.mdl_key_init(MDL_key::USER_LOCK, res->c_ptr_safe(), "");


I'd suggest

const MDL_key ull_key(MDL_key::USER_LOCK, res->c_ptr_safe(), "");

svoj · 2026-05-20T10:34:29Z

+  MDL_REQUEST_INIT(&mdl_request, mdl_namespace, db, name, mdl_type,
+                   mdl_duration);
+  if (try_acquire_lock_impl(&mdl_request, nullptr))
+    return reinterpret_cast<MDL_ticket*>(-1);


I'd rather expose it via function parameter bool &busy, such code is easier to follow.

Done. Replaced the -1 sentinel with a bool &busy output parameter.

svoj · 2026-05-20T11:34:12Z

+                                             global_system_variables
+                                             .lock_wait_timeout));
+      if (!*mdl)
        return nullptr;


I'm unsure why, but original code would return nullptr only if trylock is true. This looks wrong, but let's keep original behaviour.

Fixed — the non-trylock path no longer returns nullptr on failure, preserving the original behavior.

Add new acquire_lock() overloads on MDL_context that combine MDL_request initialization and lock acquisition into a single call, returning MDL_ticket* directly. Convert applicable call sites across the codebase to use the new ACQUIRE_LOCK macro. All new code of the whole pull request, including one or several files that are either new files or modified ones, are contributed under the BSD-new license. I am contributing on behalf of my employer Amazon Web Services, Inc.

svoj

A few more issues outlined.

svoj · 2026-05-20T20:45:34Z

+  MDL_request mdl_request;
+  mdl_request.init_with_source(mdl_namespace, db, name, mdl_type, mdl_duration,
+                               src_file, src_line);
+  busy= try_acquire_lock_impl(&mdl_request, nullptr);


You have inverted the busy meaning. It must be either

busy= !try_acquire_lock_impl(...);

or replace it with

error= try_acquire_lock_impl(...);

svoj · 2026-05-20T20:49:17Z

+    MDL_ticket *backup_ticket= thd->mdl_context.MDL_TRY_ACQUIRE_LOCK(
+        MDL_key::BACKUP, "", "", MDL_BACKUP_DDL, MDL_STATEMENT, busy);
+    if (busy)
+      break;


It is better to check busy after backup_ticket.

if (backup_ticket) ... else if (busy) break;

svoj · 2026-05-20T20:51:01Z

+    bool busy;
+    *mdl= mdl_context->MDL_TRY_ACQUIRE_LOCK(MDL_key::TABLE, db_buf, tbl_buf,
+                                             MDL_SHARED, MDL_EXPLICIT, busy);
+    if (busy || !*mdl)


It is enough to check !*mdl, there's no difference in handling these conditions. You don't need to reset *mdl.

svoj · 2026-05-20T20:53:33Z

+      bool busy;
+      *mdl= mdl_context->MDL_TRY_ACQUIRE_LOCK(MDL_key::TABLE, db_buf, tbl_buf,
+                                               MDL_SHARED, MDL_EXPLICIT, busy);
+      if (busy || !*mdl)


It is enough to check !*mdl, there's no difference in handling these conditions. You don't need to reset *mdl.

gemini-code-assist Bot reviewed May 13, 2026

View reviewed changes

Comment thread sql/mdl.h Outdated

gkodinov added the External Contribution All PRs from entities outside of MariaDB Foundation, Corporation, Codership agreements. label May 14, 2026

svoj self-requested a review May 14, 2026 09:08

gkodinov requested changes May 14, 2026

View reviewed changes

Comment thread sql/mdl.h Outdated

Comment thread sql/ddl_log.cc Outdated

gkodinov self-assigned this May 14, 2026

longjinvan force-pushed the MDEV-39548 branch 2 times, most recently from c2c8e8e to a0f3c13 Compare May 14, 2026 18:19

svoj requested changes May 14, 2026

View reviewed changes

longjinvan force-pushed the MDEV-39548 branch 2 times, most recently from 106f9e7 to 433b5cf Compare May 14, 2026 22:23

gkodinov approved these changes May 15, 2026

View reviewed changes

gkodinov assigned svoj and unassigned gkodinov May 15, 2026

svoj requested changes May 15, 2026

View reviewed changes

longjinvan force-pushed the MDEV-39548 branch from 433b5cf to 053874c Compare May 15, 2026 17:13

longjinvan force-pushed the MDEV-39548 branch from 2a60fec to 053874c Compare May 19, 2026 17:09

svoj requested changes May 20, 2026

View reviewed changes

longjinvan force-pushed the MDEV-39548 branch from 053874c to e55a536 Compare May 20, 2026 19:16

svoj requested changes May 20, 2026

View reviewed changes

svoj mentioned this pull request May 21, 2026

MDEV-36025 Parallel slave concurrent BACKUP MDL locking with ongoing … #5103

Open

Uh oh!

Conversation

longjinvan commented May 13, 2026

Description

Release Notes

How can this PR be tested?

Basing the PR against the correct MariaDB version

Copyright

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

gkodinov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

svoj left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

gkodinov left a comment

Choose a reason for hiding this comment

Uh oh!

svoj left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

longjinvan commented May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

svoj commented May 19, 2026

Uh oh!

svoj left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

svoj left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

longjinvan commented May 15, 2026 •

edited

Loading