Development plans

We are planing to develop next extensions to standard MySQL (in random order, not in order of priority):

• Integrate patch from Ebay http://code.google.com/p/mysql-heap-dynamic-rows/
• Add replication_load statistics. The idea is to measure how much time was passed since replication thread was started (simply need to measure time when it is started) compared to current time. As well as the time SQL thread spent executing the queries. If we see replication thread was running for 500 seconds and time it spent executing queries is 200 we know replication load is 40% We should just keep it as two numbers so the can be easily incrementally measured (ie each 5 minutes)
• Query Timeouts
• Make InnoDB tablespace movable when innodb_file_per_table isused
• There are two kinds of filesorts. The MSL patch doesn't show which one was chosen for any given query. Also, we could implement a new status variable to count one of the types. This would allow to know the ratio of single-pass filesorts to dual-pass.
• SHOW MUTEX STATUS not only InnoDB mutexes, but for all (in INFORMATION_SCHEMA)
• create a counter to track the size of the largest implicit temporary table created.
• Make more of InnoDB's settings into server variables (like page fill factor, LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK)
• Look if can replace standard InnoDB zlib compression to bzip2, lzo, QuickLZ
• Make InnoDB zlib_compression_level as variable
• Include the FNV_64() hash function from maatkit into release
• Add the length of started transaction in PROCESSLIST table so long transaction can be easily caught.
• Add information about Innodb we have for slow query log in the PROCESSLIST - how long was IO wait so far, how long were row lock waits etc.
• Add CPU time for the query in PROCESSLIST table.
• Add query CPU time used in show processlist once we have it.
• Move deadlock output to end of InnoDB status so it does not displace other information
• Make InnoDB data recovery process faster - http://dammit.lt/2008/10/26/innodb-crash-recovery/
• Make InnoDB data dictionary LRU based
• Make InnoDB not single mutex per buffer_pool size, but hash of mutexes
• Implement ability to examine “raw” table/index pages (location, size, free space etc).
• Make SYS_TABLES/SYS_INDEXES tables accessible read only
• Allow to size DoubleWrite buffer, assuming it gives any benefit
• Add (optionally) queries executed with error to the slow query log with their error code.
• Fix open table scalability http://www.mysqlperformanceblog.com/2006/11/21/opening-tables-scalability/
• Make Partitioning to support different storage engines

• Integrate google-smp patch or Yasufumi rw-lock patch (done, see innodb_rw_lock )
• Add Baron's patches (http://lists.mysql.com/internals/35174) to make InnoDB status smaller and configurable (done, innodb_locks_held )
• Show pattern access to InnoDB tablespaces (done, see innodb_io_pattern.patch )
• Command to check for fragmentation/fill factor for Innodb tables and Indexes (done, see innodb_check_defrag )
• Integrate Jeremy Cole's extra plugings for having Information Schema for Innodb tables (checking what tables are in memory, what is buffer pool contents etc) (done, see innodb_show_bp )

back to Percona Lab page