MDEV-8235: Expose check_slave_start_position as SQL function GTID_CHECK_POS() by ayush-jha123 · Pull Request #4956 · MariaDB/server

ayush-jha123 · 2026-04-18T15:42:24Z

I’ve been diving deep into the replication source code for the last two weeks,
investigating MDEV-8235. I noticed this task has been open for a while and realized
how useful it would be for DBAs to have a simple way to check GTID availability
directly from the SQL layer.

After studying how check_slave_start_position() works internally, I’ve implemented
a user-facing version called GTID_CHECK_POS().

What this PR does:

Extracted the internal GTID binlog-lookup logic into a shared helper function
in sql_repl.cc to avoid code duplication.
Created the Item_func_gtid_check_pos class to wrap the logic for the SQL parser.
Added input validation: it returns 1 (exists), 0 (purged), or NULL (invalid input)
which makes it robust for automation scripts.
Included an MTR test main.gtid_check_pos covering standard and edge cases.

I’m really excited to contribute this to MariaDB and would appreciate feedback
on the implementation.

gkodinov

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

First, the formalities:

please keep a single commit, with a commit message in it.
please make sure all of the buildbot hosts compile and run successfully

Below are some of my thoughts on the functionality and its implementation and test coverage. You can ignore all of the below and take it with the final reviewer. Or address it and have a faster final review with these fixed.

This commit implements the GTID_CHECK_POS() SQL function, which allows validating if a given GTID position is reachable within the current set of binary logs. This is useful for external tools or orchestration layers to verify if a node is viable for replication start without actually initiating a slave connection. Features: - Returns 1 if all GTIDs in the requested state are reachable. - Returns 0 if any GTID in the requested state has been purged. - Propagates parsing errors for malformed GTID strings. - Derived from Item_bool_func to ensure native boolean handling. - Protected by HAVE_REPLICATION for safe embedded builds. Implemented via a thin wrapper over the engine-internal gtid_find_binlog_pos() logic.

gkodinov

It's looking much better. Please make sure to address the comments below. This is my last batch of preliminary review comments.
Thanks for your efforts.

gkodinov · 2026-05-13T10:34:59Z

revert these please

gkodinov · 2026-05-13T10:35:08Z

revert this please.

ayush-jha123 · 2026-05-14T17:02:23Z

Hi @gkodinov,

I've just pushed the updated commit addressing your latest feedback:

Compilaton: Added #include "sql_repl.h" to sql/item_func.cc to fix the scope error you observed on Buildbot.
Submodules: Reverted all changes to the libmariadb and storage/columnstore/columnstore submodules.
Whitespace: I've ensured that no space-only changes are included in the diff; the comment lines and empty lines now match the original source exactly.

Thanks for the review, let me know if there's anything else!

gkodinov

Thank you for working on this. Some of the test results do not make a lot of sense it seems.

It'd be good to have even more test coverage: more edge cases, more positive and negative outcomes, scenarios producing all the possible return values and/or errors etc.

And also a better design definition of what the SQL function does is going to help a lot: e.g. possible return values, when are they returned, formal description or parameters, what determines the return value (just the parameter values or some other state, etc.

gkodinov · 2026-06-03T12:56:42Z

+  if ((null_value= args[0]->null_value))
+    return 0;
+
+#ifndef HAVE_REPLICATION


I'd reverse the condition for readability. But that's optional.

gkodinov

It is looking much better now! Thanks for the updated tests. This makes it look good. Some final remarks below.

gkodinov · 2026-06-04T12:35:44Z

+SELECT GTID_CHECK_POS('NEW_GTID_POS');
+GTID_CHECK_POS('NEW_GTID_POS')
+1
+# The old position (now in a purged log) returns 0


this one fails on buildbot:

main.gtid_check_pos w7 [ fail ] Test ended at 2026-06-03 19:55:16 CURRENT_TEST: main.gtid_check_pos --- C:/a/server/server/mysql-test/main/gtid_check_pos.result 2026-06-03 19:12:08.051490800 +0000 +++ C:\a\server\server\mysql-test\main\gtid_check_pos.reject 2026-06-03 19:55:13.752902600 +0000 @@ -30,7 +30,7 @@ # The old position (now in a purged log) returns 0 SELECT GTID_CHECK_POS('OLD_GTID_POS'); GTID_CHECK_POS('OLD_GTID_POS') -0 +1

gkodinov · 2026-06-04T12:36:31Z

@@ -37,6 +37,7 @@
 #include "sql_acl.h"                            // EXECUTE_ACL
 #include "mysqld.h"                             // LOCK_short_uuid_generator
 #include "rpl_mi.h"
+


no white-space only changes please.

gkodinov · 2026-06-04T12:38:27Z

+  and verifies if they exist and are reachable (viable) within the current
+  set of binary logs.
+
+  @param gtid_str A string representation of GTIDs (e.g. '0-1-1,0-1-2').


This will make doxygen return an error if one day it was run on these files. This is a method documentation block. And yet, you're talking about the SQL function itself. I'd move this particular block towards the class definition (?), reformat it a bit so that there's no @return or @param directives and then add a real one here explaining the return value and what the function does in specific.

gkodinov · 2026-06-04T12:39:21Z

+
+  @param gtid_str A string representation of GTIDs (e.g. '0-1-1,0-1-2').
+
+  @return


fwiw, use @RetVal if you want to do good doxygen. Or don't do doxygen.

gkodinov · 2026-06-04T12:40:34Z

@@ -5838,4 +5839,113 @@ int compare_log_name(const char *log_1, const char *log_2) {
  return res;
 }

+/**


that's what good doxygen looks like! Nice! I'd consider doing some @sa too towards the bigger picture/subsystem. But this is optional.

ayush-jha123 · 2026-06-06T18:18:23Z

Thnks again for the review, I pushed an update addressing your final remarks:

Buildbot w7 failure (gtid_check_pos.test):
The test was failing on Windows because the active binary log was occasionally locked by a background thread, preventing PURGE BINARY LOGS TO from successfully removing it before the subsequent GTID_CHECK_POS() check was made. I have refactored this block to use --source include/wait_for_purge.inc (wrapped in --disable_result_log), which safely polls until the log is genuinely purged.
Whitespace only changes: Removed the empty newline after #include "rpl_mi.h" in item_func.cc.
Doxygen Cleanup:
- I moved the general overview documentation for the SQL function into item_cmpfunc.h right above the class definition (and removed @param and @return).
- I replaced the documentation block for val_bool() in item_func.cc with a specific method-level explanation using the proper @retval tags.
- I added the @sa cross-references in sql_repl.cc to tie it into the bigger picture.

gkodinov added the External Contribution All PRs from entities outside of MariaDB Foundation, Corporation, Codership agreements. label Apr 20, 2026

gkodinov requested changes Apr 20, 2026

View reviewed changes

gkodinov self-assigned this Apr 20, 2026

ayush-jha123 force-pushed the mdev-8235-gtid-pos branch 2 times, most recently from 81aa4a1 to c4377de Compare April 24, 2026 17:57

ayush-jha123 force-pushed the mdev-8235-gtid-pos branch from c4377de to bd30bf6 Compare April 26, 2026 18:30

vuvova requested a review from knielsen May 13, 2026 10:22

gkodinov requested changes May 13, 2026

View reviewed changes

ayush-jha123 force-pushed the mdev-8235-gtid-pos branch from fb66443 to beeba7c Compare May 14, 2026 16:50

MDEV-8235: Fix symbol visibility and embedded build for GTID_CHECK_POS

ebb7195

ayush-jha123 force-pushed the mdev-8235-gtid-pos branch from beeba7c to ebb7195 Compare May 14, 2026 16:59

ayush-jha123 force-pushed the mdev-8235-gtid-pos branch 4 times, most recently from ca84483 to 75568df Compare May 25, 2026 05:13

gkodinov requested changes Jun 3, 2026

View reviewed changes

ayush-jha123 added 2 commits June 4, 2026 00:09

MDEV-8235: Fix gtid_check_pos compilation and validation logic

7c26bf7

MDEV-8235: Fix test determinism and clean up sql_repl.cc

cf68c81

ayush-jha123 force-pushed the mdev-8235-gtid-pos branch from 75568df to cf68c81 Compare June 3, 2026 19:11

gkodinov requested changes Jun 4, 2026

View reviewed changes

MDEV-8235: Address final PR review feedback

2ff319c


		@param gtid_str A string representation of GTIDs (e.g. '0-1-1,0-1-2').

		@return

Uh oh!

Conversation

ayush-jha123 commented Apr 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gkodinov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

gkodinov left a comment

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!

Uh oh!

ayush-jha123 commented May 14, 2026

Uh oh!

gkodinov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gkodinov 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!

ayush-jha123 commented Jun 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

ayush-jha123 commented Apr 18, 2026 •

edited

Loading