We are continuously getting latency in both search & indexing. We are having a cluster of 15 nodes with 3 master eligible & rest 12 are data nodes each having 30 GB RAM and we have a traffic of 300 concurrent users. Any suggestions by which I can reduce the latency?
if I see this correctly on the charts the highest reported latency for indexing is around 2.5 milliseconds and for search it spikes to roughly 12 milliseconds. Is a latency of 12 milliseconds problematic in practice or are you asking out of curiosity? Latency on this level can be caused by a lot of factors and you'd need to have a very heavily tuned system (including kernel and JVM tuning) to reduce those spikes. Some examples include:
Garbage collection pauses (in the range of tens of milliseconds but you should expect multi-second spikes occasionally as well)
Backpressure / blocking due to heavy disk I/O
CPU scheduling pressure / threads moved to different CPU cores
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.