Aaron, I have not deep dived the data files in a while but this is how I understand it.
http://wiki.apache.org/cassandra/ArchitectureSSTable There is no need to store the row key each time with the column. RowKey to columns is a one to many relationship. This would be a diagram of a physical file: Hbase does it like this i guess (Or it used to I am not up on the news): rowkey1,column1,value1 rowkey1,column2,value2 First, I believe they repeat the row key for each column, which is not a huge deal because you should always use compression but it is a bit wasteful especially for a non-compressed table. I know this has some impact on very wide rows because a single rowkey must fit inside this structure of an hfile. ? (again its been a while) But to get back to your question. In cassandra: sstable1 rowkey1: numberof columns 26 (column1,value1,ts1) .... (column26,value26,ts26) sstable2 rowkey1: numberof columns 1 (column1,value1,ts2) The rowkey appears once in a given sstable if the row has 1 or more columns in the sstable. On the read path Cassandra searches all sstables find all the columns for a row (bloom filters and other criteria eliminate some sstables from read path). It then merges the row factoring in tombstones and the last update win rules for a column. On Sat, Aug 11, 2012 at 2:03 PM, Aaron Turner <synfina...@gmail.com> wrote: > So how does that work? An sstable is for a single CF, but it can and > likely will have multiple rows. There is no read to write and as I > understand it, writes are append operations. > > So if you have an sstable with say 26 different rows (A-Z) already in > it with a bunch of columns and you add a new column to row J, how does > Cassandra store the column/value pair on disk in a way to refer to row > J without re-writing the row key or some representation of it? > > Thanks, > Aaron > > On Fri, Aug 10, 2012 at 7:53 PM, Terje Marthinussen > <tmarthinus...@gmail.com> wrote: >> Rowkey is stored only once in any sstable file. >> >> That is, in the spesial case where you get sstable file per column/value, >> you are correct, but normally, I guess most of us are storing more per key. >> >> Regards, >> Terje >> >> On 11 Aug 2012, at 10:34, Aaron Turner <synfina...@gmail.com> wrote: >> >>> Curious, but does cassandra store the rowkey along with every >>> column/value pair on disk (pre-compaction) like Hbase does? If so >>> (which makes the most sense), I assume that's something that is >>> optimized during compaction? >>> >>> >>> -- >>> Aaron Turner >>> http://synfin.net/ Twitter: @synfinatic >>> http://tcpreplay.synfin.net/ - Pcap editing and replay tools for Unix & >>> Windows >>> Those who would give up essential Liberty, to purchase a little temporary >>> Safety, deserve neither Liberty nor Safety. >>> -- Benjamin Franklin >>> "carpe diem quam minimum credula postero" > > > > -- > Aaron Turner > http://synfin.net/ Twitter: @synfinatic > http://tcpreplay.synfin.net/ - Pcap editing and replay tools for Unix & > Windows > Those who would give up essential Liberty, to purchase a little temporary > Safety, deserve neither Liberty nor Safety. > -- Benjamin Franklin > "carpe diem quam minimum credula postero"
