MySQL Replibeertion

Kiev, Saturday, November 11, 2006 — 15:30–???

All about MySQL Replication— This session is in the style of the MySQL Replication BoF at the Users Conference. Anything goes. Lots of rumbling. Maybe a bit of fighting. Plenty of technical jargon.

What works with replication, what doesn't, what sucks, what's great, and all of your pet features. Plenty of beer to go around!

Participants

Sound good? Add your name, and if you like, your beer; we'll try to make sure some of it shows up:

• Jeremy Cole — Guinness
• Eric Bergen
• Joel Seligstein
• Baron Schwartz — Hard cider
• Monty Taylor
• Adam Ritter — A plum floating in perfume, served in a man's hat(Failing that, Shinerbock/Amberbock)
• Yoshinori Matsunobu
• Ronald Bradford (most beer work for me)
• Jim Dowling — Anything Belgian (except kriek)
• Wei Li
• Farhan "Frank Mash" Mashraqi (Corona, Miller Light, or just about anything)
• Justin Swanhart
• Joel Lin
• Mark Atwood

Notes

Reasons

• scaling
• backups
• data warehousing
• using multiple storage engines

Problems

• breaks
• lags
• hard to setup
• hard to know when the slave is broken
• its hard to fix it when it's broken
• hard to migrate to replication
• hard to make a slave read-only
• headaches with binlogs
• no row level replication in 4.X
• slaves run in serial, so may not be able to keep up with master
• the master does not keep track of the slaves
  • thus everyone has to write their own log purging scripts, (and usually get it wrong)
• no multimaster replication, a slave can not have more than one master
  • this is under active development by MySQL
• ring replication works when it works, but when it breaks, it can't be fixed
• slaves dont start a query until the master is done

Things Replication Needs

• want delayed replication, ferex keep slave on hour behind the master
• registered slaves in the master
• input binlog
• checksum table event, with running checksums
• global sequence number
• fast reconnect with exponential back off
• exclusion list, want commands that dont affect the data to log
• statement filtering on slaves
  • if the master keeps track of slaves, slaves could tell master what it's filter is, and let the master do (some of) the filtering
• smarter binlog tool
  • going backwards
  • need index
  • send subset to a server