The three companies Continuent, Codership, and Monty Program are planning to start working on some enhancements to the replication system in MariaDB, together with anyone interested in joining in. At this stage, there are no fixed directions for the project, and to do this in as open a way possible with the maximum community involvement and interest, we agreed to start with an email discussion on the maria-developers@ mailing list. So consider it started! The plan so far is: 1) The parties taking this initiative, MP, Continuent, and Codership, present their own ideas in this thread on maria-developers@ (and everyone else who wants to chime in at this stage). 2) Once we have some concrete suggestions as a starting point, we use this to reach out in a broader way with some blog posts on planetmysql / planetmariadb to encourage further input and discussions for possible directions of the project. Eventually we want to end up with a list of the most important goals and a possible roadmap for replication enhancements. (It is best to have something concrete as a basis of a broad community discussion/process). To start of, here are some points of interest that I collected. Everyone please chime in with your own additional points, as well as comments and further details on these one. Three areas in particular seem to be of high interest in the community currently (not excluding any other areas): - High Availability * Most seems to focus on building HA solutions on top of MySQL replication, eg. MMM and Tungsten. * For this project, seems mainly to be to implement improvements to replication that help facilitate improving these on-top HA solutions. * Tools to automate (or help automate) failover from a master. * Better facilities to do initial setup of new slave without downtime, or re-sync of an old master or slave that has been outside of the replication topology for some period of time. - Performance, especially scalability * Multi-threaded slave SQL thread. * Store the binlog inside a transactional engine (eg. InnoDB) to reduce I/O, solve problems like group commit, and simplify crash recovery. - More pluggable replication * Make the replication code and APIs be more suitable for people to build extra functionality into or on top of the stock MySQL replication. * Better documentation of C++ APIs and binlog format. * Adding extra information to binlog that can be useful for non-standard replication stuff. For example column names (for RBR), checksums. * Refactoring the server code to be more modular with APIs more suitable for external usage. * Add support for replication plugins, types to be determined. For example binlog filtering plugins? It is also very important to consider the work that the replication team at MySQL is doing (and has done). I found a good deal of interesting information about this here: http://forge.mysql.com/wiki/MySQL_Replication:_Walk-through_of_the_new_5.1_a...) This describes a number of 6.0/5.4 and preview features that we could merge and/or contribute to. Here are the highlights that I found: - Features included in 6.0/5.4 (which are cancelled I think, but presumably this will go in a milestone release): * CHANGE MASTER ... IGNORE_SERVER_IDS for better support of circular replication. * Replication heartbeat. * sync_relay_log_info, sync_master_info, sync_relay_log, relay_log_recovery for crash recovery on slave. * Binlog Performance Optimization (lock contention improvement). * Semi-synchronous Replication, with Pluggable Replication Architecture. http://forge.mysql.com/wiki/ReplicationFeatures/SemiSyncReplication - Feature previews: * Parallel slave application: WL#4648 http://forge.mysql.com/wiki/ReplicationFeatures/ParallelSlave * Time-delayed replication: WL#344 http://forge.mysql.com/wiki/ReplicationFeatures/DelayedReplication * Scriptable Replication: WL#4008 http://forge.mysql.com/wiki/ReplicationFeatures/ScriptableReplication * Synchronous Replication. Drizzle is also doing work on a new replication system. I read through the series of blog posts that Jay Pipes wrote on this subject. They mostly deal with how this is designed in terms of the Drizzle server code, and is low on detail about how the replication will actually work (the only thing I could really extract was that it is a form of row-based replication). If someone has links to information about this that I missed, it could be interesting. Let the discussion begin! - Kristian.