Reliability of Elasticsearch bulk indexing on client side (web server) can be improved by :
- retrying to send the failed index requests. See the BackoffPolicy of class BulkProcessor in Elasticsearch codebase.
- sizing the max number of concurrent requests according to the nb of shards in index. The implementation of
SONAR-8075currently hardcodes #maxShards=5 in the formula, whereas it depends on index. Note that nb of shards can be changed, and be greater than 5, in cluster mode with the properties sonar.search.*.shards.