Hello Kevin. With CQL3 there are some important terms to define:
a. Row : means a logical row in the CQL3 semantics, logical row is what is displayed as a row in cqlsh client b. Partition: means a physical row on disk in the CQL3 semantics Even if you have tiny logical rows, if you store a lot of them under the same partition (physical row on disk) then it can add up a lot. Quick maths: 200k per logical row * 1000 logical rows = 200Mb roughtly for the partition On Mon, Jun 30, 2014 at 6:53 PM, Kevin Burton <bur...@spinn3r.com> wrote: > I'm running a full compaction now and noticed this: > > Compacting large row … incrementally > > … and the values were in the 300-500MB range. > > I'm storing NOTHING anywhere near that large. Max is about 200k... > > However, I'm storing my schema in a way so that I can do efficient > time/range scans of the data and placing things into buckets. > > So my schema looks like: > > bucket, > timestamp > > … and the partition key is bucket. Since this is a clustering row, does > that mean that EVERYTHING is in one "row" under 'bucket' ? > > So even though my INSERTs are like 200k, they're all pooling under the > same 'bucket' which is the partition key so cassandra is going to have a > hard time compacting them. > > Part of the problem here is the serious abuse of vocabulary. The > thrift/CQL impedance mismatch means that things have slightly different > names and not-so-straigtforward nomenclature. So it makes it confusing as > to what's actually happening under the hood. > > …. > > Then I saw: > > > http://mail-archives.apache.org/mod_mbox/cassandra-user/201106.mbox/%3cbanlktik0g+epq4ctw28ty+dpexprtis...@mail.gmail.com%3E > > > look for in_memory_compaction_limit_in_mb in cassandra.yaml > > > … so this seems like it will be a problem and slow me down moving forward. > Unless I figure out a workaround. > > -- > > Founder/CEO Spinn3r.com > Location: *San Francisco, CA* > blog: http://burtonator.wordpress.com > … or check out my Google+ profile > <https://plus.google.com/102718274791889610666/posts> > <http://spinn3r.com> > >