In Tuning indices.memory.max_index_buffer_size for indexing throughput I'm trying to figure out why one of my test indices had only 54 merges (vs. at least 400 usually) and dramatically less total indexing and merging time (but I believe the basically the same indexed content). (I have the detailed stats - I'll put them at the end). Maybe the automatic merge policy behaved differently for that run? Each next-iteration index we create is write-only (I.e., not being searched) until it is complete and I'd like to exploit that and those merge policy settings seemed very promising. I've tried increasing translog to 1g, increased memory buffer from 10% to 25%, decreasing client indexing concurrency, and increased throttle to 100mb, but still haven't reproduced my very fast indexing run. Any tips/ideas?
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.