cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apache Wiki <wikidi...@apache.org>
Subject [Cassandra Wiki] Update of "CassandraLimitations" by JonathanEllis
Date Tue, 16 Jun 2009 20:56:34 GMT
Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Cassandra Wiki" for change notification.

The following page has been changed by JonathanEllis:
http://wiki.apache.org/cassandra/CassandraLimitations

New page:
= Limitations =

>>From easiest to fix to hardest:

 * Cassandra's compaction code currently deserializes an entire row (per columnfamily) at
a time.  So all the data from a given columnfamily/key pair must fit in memory.  Fixing this
is relatively easy since columns are stored in-order on disk so there is really no reason
you have to deserialize row-at-a-time except that that is easier with the current encapsulation
of functionality.
 * Cassandra does not currently fsync the commitlog before acking a write.  Most of the time
this is Good Enough when you are writing to multiple replicas since the odds are slim of all
replicas dying before the data actually hits the disk, but the truly paranoid will want real
fsync-before-ack.  Just adding fsync would be just a few lines (to CommitLog, naturally),
but we want to do this without killing performance, so what we want is an Executor that fsyncs
after writing batches of commitlog entries (and then asynchronously notifies the write threads).
 * Cassandra has two levels of indexes: key and column.  But in super columnfamilies there
is a third level of subcolumns; these are not indexed, and any request for a subcolumn deserializes
_all_ the subcolumns in that supercolumn.  So you want to avoid a data model that requires
large numbers of subcolumns.  This can be fixed; the core classes involved are SuperColumn
and SequenceFile.
 * Cassandra's public API is based on Thrift, which offers no streaming abilities -- any value
written or fetched has to fit in memory.  This is inherent to Thrift's design; I don't see
it changing.  So (similar to traditional rdbmses) you're better off storing large blobs directly
in the filesystem with a pointer to machine:path, than storing the blobs directly in Cassandra.

Mime
View raw message