We do not want to enforce log4j onto everyone using Hector just because Hector uses it to log the performance counters. We accomplish this by invoking log4j's DOMConfigurator.configure method dynamically rather than statically and by tagging the log4j dependency as runtime-scoped and optional. The slf4j-log4j and jcl-over-slf4j dependencies are only needed for testing. Also bumped slf4j version to 1.5.11. Change by: eyealike
When using SlicePredicate for range queries, Hector expects both start and finish values to be set. This is not the best behavior. For example, Cassandra allows to query for the "first N columns" simply by setting start and finish values to empty string and setting count to N. Although this query can be performed with String/StringSerializer, with UUID/UUIDSerializer it can't be done. To get "first N columns" of type UUID, one has to generate min/max UUID's and use these as range values.
SuperSliceQuery had wrong semantics. It allowed to query subcolumns slices, not super-columns as one would expect.
…t doesn't import any thrift specific classes and create a ThriftCluster implementation
…interface to the api package