Major compactions should only be considered when your rows tend to be split across many SSTables. i.e., you are constantly appending to or updating rows. Even then, they do put quite a load on your node, so they are less attractive if you don't have slow periods to run them. It should also be noted that tombstones were only removed during major compactions before 0.6.6.
If you do schedule them, you should stagger your compactions around the ring so that they aren't all occurring at once. Since #1812 ( https://issues.apache.org/jira/browse/CASSANDRA-1812) is done, you can also consider only compacting one CF at a time to reduce the impact with 0.7.0 or 0.6.9. - Tyler On Mon, Dec 13, 2010 at 5:37 AM, Colin Vipurs <zodiac...@gmail.com> wrote: > We've had our small cluster successfully running in production for ~6 > weeks now with everything working beautifully. Over the weekend I > noticed the load on 2 boxes shoot up to 5-6x that of the others and > after kicking off a major compaction the load has come back down to > normal. What procedure should be used for triggering major > compaction? I'm sure I read somewhere that just cron'ing it to run on > a regular schedule isn't such a good idea, but it looks like something > we're going to have to do. > > Is anyone else hitting this and what are you doing? > > -- > Maybe she awoke to see the roommate's boyfriend swinging from the > chandelier wearing a boar's head. > > Something which you, I, and everyone else would call "Tuesday", of course. >
