We are running an 8 nodes cluster of elastic v1.5.
We are experiencing a strange behavior of statistical facets queries on one particular field (type double) of an index (4 shards, 1 replica).
It's starting to return very large values (like 1E18) for total, max, mean, sum_of_squares, variance and std_deviation. Min and count values are correct. It happens even on small doc count, we get the same result with sum aggregation.
Did anyone already experience this issue ?
How to fix it ? (close/open index, reindex, cluster restart)
I think you got hit by this bug: https://github.com/elastic/elasticsearch/issues/8688 which will be fixed in 2.0. What probably happened is that you field got dynamically mapped as a long on a shard and a double on another shard, then shard moved and you either ended up interpreting long bits as doubles or double bits as longs. The only work-around in 1.x would be to map fields explicitely prior to indexing. Unfortunately in your case there is not other fix than reindexing.
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.