MDEV-16335: Include deadlock detail information in SHOW WARNINGS #4492

varundeepsaini · 2025-12-22T19:30:20Z

The Jira issue number for this PR is: MDEV-16335

Description

Currently, deadlock information is only available globally via SHOW ENGINE INNODB STATUS, so a client can never be certain that the information shown is for their own session's deadlock.

This patch stores deadlock information in the victim transaction's structure when a deadlock is detected, making it accessible via SHOW WARNINGS.

Implementation

Added deadlock_info fields to trx_lock_t to store per-session deadlock information
In Deadlock::report(), copy deadlock info to the victim transaction (only when srv_print_all_deadlocks is set)
In convert_error_code_to_mysql(), push deadlock info as a warning via push_warning_printf()
Clear deadlock info when a new transaction starts or when the transaction is freed

Release Notes

When innodb_print_all_deadlocks=ON, deadlock detail information is now available per-session via SHOW WARNINGS after a deadlock error (ER_LOCK_DEADLOCK).

How can this PR be tested?

./mtr --suite=innodb deadlock_session_info

Basing the PR against the correct MariaDB version

This is a new feature and the PR is based against the main branch.

PR quality check

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.

…etection error

CLAassistant · 2025-12-22T19:30:26Z

All committers have signed the CLA.

mysql-test/suite/innodb/t/deadlock_session_info.test

mysql-test/suite/innodb/r/deadlock_session_info.result

storage/innobase/include/trx0trx.h

mysql-test/suite/innodb/t/deadlock_session_info.test

Signed-off-by: Varun Deep Saini <varun.23bcs10048@ms.sst.scaler.com>

storage/innobase/handler/ha_innodb.cc

mysql-test/suite/innodb/r/deadlock_session_info.result

storage/innobase/lock/lock0lock.cc

FooBarrior · 2025-12-24T19:10:24Z

storage/innobase/lock/lock0lock.cc

+                                       static_cast<size_t>(len),
+                                       lock_latest_err_file);
+              if (deadlock_info_len && deadlock_info[deadlock_info_len - 1] == '\n')
+                --deadlock_info_len;


We cannot read more than ftell returned: deadlock_info is a temporary file and we access it under lock.
We can put assert

ut_ad(len >= deadlock_info_len);

maybe +1 somewhere -- please check.

Then this if is not needed, as well as zero assignment.

What we know is that fread can return only less bytes than ftell did:
https://en.cppreference.com/w/c/io/fread#:~:text=stream%20to%20read-,Return%20value,-Number%20of%20objects

storage/innobase/lock/lock0lock.cc

FooBarrior

Looks good to me.

dr-m

The test for this functionality is only executed in debug-instrumented builds. I think that new functionality had better tested with all builds, so that the functionality will be better covered. For example, GNU/Linux distributions that package MariaDB themselves will likely not use CMAKE_BUILD_TYPE=Debug at all. We should test what we ship.

dr-m · 2026-01-19T06:40:56Z

storage/innobase/lock/lock0lock.cc

+      deadlock_info= static_cast<char*>(ut_malloc_nokey(
+        static_cast<size_t>(len) + 1));
+      if (!deadlock_info)
+        goto func_exit;


The memory will be leaked if !deadlock_info.

Why would we do all this busy work if !current_trx || victim != trx || !trx->mysql_thd?

The memory will be leaked if !deadlock_info.

Which memory?

I mean in case of !deadlock_info it's not allocated, and besides ut_free is called unconditionally after func_exit. So do you mean some other memory?

Oh, sorry, I take my words back regarding deadlock_info == nullptr.

janlindstrom · 2026-01-19T07:12:39Z

storage/innobase/lock/lock0lock.cc

+
+      if (!srv_print_all_deadlocks)
+        goto func_exit;
+      long len= ftell(lock_latest_err_file);


This is confusing, why you do these file operations? IMHO it would be better to change current print functions to produce string that is then written to both to the deadlock error file and pushed to warning.

The temporary file is a relic from MySQL 5.5 or earlier when InnoDB was written in C and there was no formatted output facility into a heap-allocated buffer. That could be fixed separately. I seemed to remember that there is an option to expose the lock_latest_err_file in the file system, but there only is the Boolean option innodb_status_file that covers the entire SHOW ENGINE INNODB STATUS output.

@janlindstrom the nearly same is done from the side of SHOW INNODB STATUS -- the file is opened, ftell'd, etc. I think this work may remain local, wihtout the requirement to refactor.

FooBarrior · 2026-01-19T13:41:09Z

storage/innobase/lock/lock0lock.cc

+                                      static_cast<size_t>(len),
+                                      lock_latest_err_file);
+      ut_ad(deadlock_info_len <= static_cast<size_t>(len));
+      if (deadlock_info[deadlock_info_len - 1] == '\n')


this will crash if deadlock_info_len is 0

FooBarrior · 2026-01-19T14:10:18Z

You didn't check

I checked the CODING_STANDARDS.md file and my PR conforms to this where appropriate.

Signed-off-by: Varun Deep Saini <varun.23bcs10048@ms.sst.scaler.com>

Include deadlock detail information in SHOW WARNINGS after deadlock d…

77ce06a

…etection error

FooBarrior requested changes Dec 22, 2025

View reviewed changes

varundeepsaini marked this pull request as draft December 22, 2025 19:46

FooBarrior requested a review from Thirunarayanan December 22, 2025 19:55

FooBarrior requested changes Dec 22, 2025

View reviewed changes

mysql-test/suite/innodb/r/deadlock_session_info.result Outdated Show resolved Hide resolved

storage/innobase/include/trx0trx.h Outdated Show resolved Hide resolved

FooBarrior reviewed Dec 22, 2025

View reviewed changes

mysql-test/suite/innodb/t/deadlock_session_info.test Outdated Show resolved Hide resolved

FooBarrior reviewed Dec 22, 2025

View reviewed changes

mysql-test/suite/innodb/t/deadlock_session_info.test Outdated Show resolved Hide resolved

varundeepsaini force-pushed the MDEV-16335-deadlock-info-show-warnings branch 2 times, most recently from cc12800 to 67920e9 Compare December 22, 2025 21:07

Added the global env condition and simplified the tests

19592c6

Signed-off-by: Varun Deep Saini <varun.23bcs10048@ms.sst.scaler.com>

varundeepsaini force-pushed the MDEV-16335-deadlock-info-show-warnings branch from 67920e9 to 19592c6 Compare December 22, 2025 21:23

dr-m reviewed Dec 23, 2025

View reviewed changes

storage/innobase/handler/ha_innodb.cc Outdated Show resolved Hide resolved

grooverdan reviewed Dec 23, 2025

View reviewed changes

mysql-test/suite/innodb/r/deadlock_session_info.result Outdated Show resolved Hide resolved

storage/innobase/lock/lock0lock.cc Outdated Show resolved Hide resolved

Resolved comments

0cf376f

varundeepsaini force-pushed the MDEV-16335-deadlock-info-show-warnings branch from 63a7517 to 0cf376f Compare December 24, 2025 05:04

remove non functional changes

948f9a8

varundeepsaini marked this pull request as ready for review December 24, 2025 05:22

FooBarrior requested changes Dec 24, 2025

View reviewed changes

storage/innobase/lock/lock0lock.cc Outdated Show resolved Hide resolved

storage/innobase/lock/lock0lock.cc Outdated Show resolved Hide resolved

Updated tests

1eadd6b

FooBarrior requested changes Dec 24, 2025

View reviewed changes

varundeepsaini and others added 2 commits December 25, 2025 01:49

resolved comments

8d4c835

Merge branch 'main' into MDEV-16335-deadlock-info-show-warnings

2b12fdc

FooBarrior reviewed Dec 25, 2025

View reviewed changes

storage/innobase/lock/lock0lock.cc Outdated Show resolved Hide resolved

refactoring

b940c6c

varundeepsaini requested a review from FooBarrior December 26, 2025 18:09

gkodinov added the External Contribution All PRs from entities outside of MariaDB Foundation, Corporation, Codership agreements. label Dec 29, 2025

FooBarrior approved these changes Jan 19, 2026

View reviewed changes

dr-m requested changes Jan 19, 2026

View reviewed changes

janlindstrom reviewed Jan 19, 2026

View reviewed changes

FooBarrior requested changes Jan 19, 2026

View reviewed changes

Added a new test for non debug builds

4f63d11

Signed-off-by: Varun Deep Saini <varun.23bcs10048@ms.sst.scaler.com>

Uh oh!

MDEV-16335: Include deadlock detail information in SHOW WARNINGS #4492

Are you sure you want to change the base?

MDEV-16335: Include deadlock detail information in SHOW WARNINGS #4492

Conversation

varundeepsaini commented Dec 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Implementation

Release Notes

How can this PR be tested?

Basing the PR against the correct MariaDB version

PR quality check

Uh oh!

CLAassistant commented Dec 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

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!

Uh oh!

Uh oh!

FooBarrior left a comment

Choose a reason for hiding this comment

Uh oh!

dr-m 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

FooBarrior commented Jan 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

7 participants

varundeepsaini commented Dec 22, 2025 •

edited

Loading

CLAassistant commented Dec 22, 2025 •

edited

Loading