I'm trying to evaluate using Riak's Post-Commit Hooks to build a distributed, incremental MapReduce-based index, but was wondering which Riak nodes the Post-Commit Hooks actually run on. Are they run on the nodes the client used to put the commits, or on the primary nodes where the data is persisted? If it's the latter, I'm thinking I can from there efficiently do a map or reduce and put additional records from the output.