Running Astyanax/Thrift-based client queries against a 6-node Cassandra Cluster, we want to get more insight into which queries are the top-hitters from a returning number of bytes POV. The reason for that is that Cassandra nodes occasionally dies with an OOME and memdump analysis shows large byte array allocations in e.g. TFramedTransport in the Thrift-layer, so we think certain queries are asking for too much data.
Is there some sort of trace facility, which returns the largest queries from a returned byte POV for Thrift-based (not CQL) queries?
Thanks.