Add PostgreSQL support for rollback backend #16757

johnnyjoy · 2025-12-17T19:06:00Z

Summary

Adds PostgreSQL support to the rollback backend, allowing servers that already use PostgreSQL as their primary database backend to use rollback without relying on SQLite.

Details

Implements a PostgreSQL-backed rollback storage alongside the existing SQLite implementation
Wires backend selection through existing server configuration
Adds documentation and configuration examples for PostgreSQL rollback usage
Includes a CMake preset for building a server-only PostgreSQL configuration

Testing

Built server with PostgreSQL enabled
Verified rollback recording and lookup using a PostgreSQL-backed server
Confirmed existing SQLite-based rollback behavior remains unchanged

sfan5

The rollback code is a bit neglected, but surprisingly many people still use it and I guess adding another backend is okay.

CMakePresets.json

doc/world.mt.rollback_postgresql.example

minetest.conf.example

src/server.cpp

src/server/rollback_postgresql.cpp

src/server/rollback.cpp

src/server/rollback.h

sfan5 · 2025-12-25T13:08:19Z

src/server/rollback.cpp

+		endSaveActions();
+		action_todisk_buffer.clear();
+	} catch (...) {
+		rollbackSaveActions();


this seems somewhat assuming of a database-like backend. I would have expected for flush to be fully implemented in the child class.

Made "flush()" pure virtual in the base class and extracted the common buffer iteration logic to a "flushBufferContents()" helper method. Each backend now implements its own "flush()" method with backend-specific transaction handling, allowing for future backends that don't use database transactions.

sfan5 · 2025-12-25T13:12:18Z

src/server/rollback_postgresql.cpp

+
+RollbackManagerPostgreSQL::~RollbackManagerPostgreSQL()
+{
+	flush();


is this safe if connectToDatabase fails?

Good catch. Thanks. Added an "if (initialized())" check before calling "flush()" in the destructor. This ensures that "flush()" is called only if the database connection was successfully established. The same safety check was also added to the SQLite backend for consistency.

src/server/rollback_postgresql.cpp

src/server/rollback_sqlite3.cpp

Extract the large cache loading blocks from RollbackManagerPostgreSQL constructor into loadActorCache() and loadNodeCache() helper methods. This reduces the constructor from ~33 lines to 5 lines, addressing sfan5's feedback about large code blocks.

builtin/settingtypes.txt

doc/lua_api.md

src/server.cpp

src/server/rollback_postgresql.h

src/server/rollback_sqlite3.cpp

SmallJoker · 2025-12-28T11:00:13Z

src/server/rollback_sqlite3.cpp

+				// Invalid format - bind 0,0,0 as fallback
+				SQLOK(sqlite3_bind_int(stmt_do, 10, 0));
+				SQLOK(sqlite3_bind_int(stmt_do, 11, 0));
+				SQLOK(sqlite3_bind_int(stmt_do, 12, 0));


This was not part of the original code. Did you ensure that these code blocks are functionally identical?

I have tested it and it works fine.

SQLOK expands to.
if (sqlite3_bind_int(stmt_do, 10, 0) != SQLITE_OK) {
throw FileNotGoodException(...);
}

This might be a little paranoid. I can remove it if you like.

My point was the fallback to 0,0,0. I think the previous code would throw an std::out_of_range exception if npos was returned from one of the find calls. If that's the case, then this new code would be unused - even worse - it would become code that barely anyone would want to touch due to the rick of breaking things.

EDIT: Also 0,0,0 is a valid node position, potentially causing unexpected side-effects.

Sorry, I was being blind there. That is a brain fart. Damn.

parseNodemetaLocation() should throw a std::invalid_argument on invalid format instead of returning false?

Yes, I believe that would be appropriate. Of course, having a few unittest lines for that would be greatly appreciated.

Added unittest TestRollbackBackendBase

src/server/rollback_sqlite3.cpp

Co-authored-by: SmallJoker <[email protected]>

Improve error handling in Server::createRollbackManager - Throw exception if world.mt cannot be read, matching openModStorageDatabase pattern - Remove unused backend_source variable and simplify logging - Simplify backend selection using world_mt.getNoEx() instead of if-else block Enhance error messages and help output - Add Server::getRollbackBackends() to dynamically list available backends - Use dynamic backend list in error message for unsupported backends - Add rollback backends to print_help() output in main.cpp - Remove redundant USE_POSTGRESQL=OFF specific error message Improve destructor safety and consistency - Move db != nullptr check from SQLite destructor to flush() method - Move initialized() check from PostgreSQL destructor to flush() method - Ensures consistent safety checks across both backends Restore spacing alignment for readability - Add space padding after sqlite3_bind_* function names - Matches existing pattern in rollback_sqlite3.cpp Rename RollbackManager to RollbackMgr for line length compliance - Use RollbackMgr abbreviation following existing codebase patterns - Rename RollbackManager → RollbackMgr - Rename RollbackManagerSQLite3 → RollbackMgrSQLite3 - Rename RollbackManagerPostgreSQL → RollbackMgrPostgreSQL - Break long function signatures to ensure all lines are under 95 characters

…ng bool The original code would throw std::out_of_range when substr() was called with npos, making the fallback code that bound 0,0,0 unreachable dead code. This change makes parseNodemetaLocation() throw std::invalid_argument on invalid format, matching the original behavior and eliminating the dead code. Changes: Change parseNodemetaLocation() signature from bool to void Throw std::invalid_argument exceptions instead of returning false Remove unreachable fallback blocks in SQLite and PostgreSQL backends Add #include <stdexcept> to rollback.cpp This addresses SmallJoker's feedback.

Address SmallJoker's request for unit tests to verify that parseNodemetaLocation() correctly throws std::invalid_argument on invalid format. Tests cover: Valid format parsing (basic, negative, zero coordinates) Invalid format exceptions (missing prefix, missing commas, empty string)

Add PostgreSQL support for rollback backend

9975b66

Zughy added @ Startup / Config / Util Feature ✨ PRs that add or enhance a feature Roadmap: Needs approval The change is not part of the current roadmap and needs to be approved by coredevs beforehand labels Dec 17, 2025

sfan5 requested changes Dec 18, 2025

View reviewed changes

Made changes suggest by sfan5 and addition cleanup

9cf21dd

sfan5 self-requested a review December 19, 2025 19:36

sfan5 added Roadmap: supported by core dev PR not adhering to the roadmap, yet some core dev decided to take care of it and removed Roadmap: Needs approval The change is not part of the current roadmap and needs to be approved by coredevs beforehand labels Dec 19, 2025

Removed edit of minetest.conf.example, and removed support for PG < v9.5

fa5718d

sfan5 reviewed Dec 25, 2025

View reviewed changes

sfan5 self-requested a review December 27, 2025 21:27

SmallJoker reviewed Dec 28, 2025

View reviewed changes

Update builtin/settingtypes.txt

685ece0

Co-authored-by: SmallJoker <[email protected]>

Zughy added the Action / change needed Code still needs changes (PR) / more information requested (Issues) label Dec 28, 2025

johnnyjoy and others added 2 commits December 28, 2025 10:20

Update doc/lua_api.md

6301a7d

Co-authored-by: SmallJoker <[email protected]>

SmallJoker self-requested a review December 28, 2025 20:24

johnnyjoy added 2 commits December 28, 2025 17:40

Zughy removed the Action / change needed Code still needs changes (PR) / more information requested (Issues) label Dec 29, 2025

Add PostgreSQL support for rollback backend #16757

Are you sure you want to change the base?

Add PostgreSQL support for rollback backend #16757

Conversation

johnnyjoy commented Dec 17, 2025

Summary

Details

Testing

Uh oh!

sfan5 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!

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!

Choose a reason for hiding this comment

Uh oh!

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!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SmallJoker Dec 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

SmallJoker Dec 28, 2025 •

edited

Loading