I am using Elastic Search to index ~600M documents per day containing
several text fields. I have been trying to optimize the index building
process and am running into some snags. Using 64 shards for 8 nodes, ES
indexes ~460M documents in 90 minutes before I get an IndexMissingException
and lose shards, at which point one of the 8 nodes starts throwing
IndexMissingException's, causing the Map-Reduce job to throw enough errors
to stop the index building process.
Setup:
- Dedicated 8 node cluster (each node has 12GB RAM, quad core, 8GB heap
space for Elastic Search of that) - ES v0.19.8
- Separate cluster that runs a Map-Reduce job (I use 60 mappers) that
populate an index through the bulk API (each request contains a maximum of
10,000 requests) - Successful full index build (all documents) from running the Map-Reduce
index building and ES on same nodes. - ~600M documents (which takes up ~250GB)
- No replicas while indexing (planning to make 1 replica after indexing)
Failure (changed IP address to be localhost):
Any insights or help would be much appreciated!
Thanks!
Sushant
--