I'm not sure why that would lead to different values being produced on the exact same data (assuming you didn't reindex the data, which might lead to slightly different sharding), but if you move the questing over to the elasticsearch section someone might be able to tell you.
The sequence in which the same set of documents are provided to the percentiles aggregation can lead to minor variations in results.
The insertion sequence can be affected by:
Primary and replica shards having different physical arrangements of content on disk (Lucene merge operations are non-deterministic)
The sequence of merging results from multiple shards (again, a non-deterministic order)
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.