AFAIk the practical limit comes from creating a CF for each secondary index that has the same memtable settings as the containing CF. And the extra time it takes to maintain the index for each indexed column, sample the files when starting up, hold their bloom filters and index samples, keep the data pages in os cache......
10,000 secondary indexes does not sounds like a good approach. If you can provide some more background we may be able to suggest an alternative approach. Aaron On 20 Apr 2011, at 10:45, Jason Kolb wrote: > I apologize if this has been answered before, I've tried to do some pretty > exhaustive searching of the archives and haven't been able to see if this > question has been answered before. > > I was wondering if anyone knows if there is a practical upper limit on the > number of secondary indexes used, if they're sparsely populated (say, 10,000 > secondary indexes only 2 of which are populated per row). My understanding > is that Cassandra creates another column family for each secondary index in > the background, so the real limitation would appear to be the number of > column families. > > Is this correct? And if so (or even if not), does anyone know the answer to > the question about the upper limit on the number of secondary indexes? > > Thanks! > Jason
