The key is this line: Read 827 live and 6948 tombstoned cells
That means you either have a lot of deleted or TTLed columns in that row. One option to help with that is to set a lower gc_grace for the table and repair more frequently; this will help tombstones get purged more quickly. Another option is to adjust your data model so that you periodically switch to a new row. On Wed, Oct 23, 2013 at 7:20 PM, Matt Mankins <mmank...@fastcompany.com>wrote: > Hi. > > I have a table with about 300k rows in it, and am doing a query that > returns about 800 results. > > select * from fc.co WHERE thread_key = 'fastcompany:3000619'; > > The read latencies seem really high (upwards of 500ms)? Or is this > expected? Is this bad schema, or…? What's the best way to trace the > bottleneck, besides this tracing query: > > http://pastebin.com/sherFpgY > > Or, how would you interpret that? > > I'm not sure that row caches are being used, despite them being turned on > in the cassandra.yaml file. > > I'm using a 3 node cluster on amazon, using datastax community edition, > cassandra 2.0.1, in the same EC2 availability zone. > > Many thanks, > @Mankins > > -- Tyler Hobbs DataStax <http://datastax.com/>