Loading a dashboard with 5 visualizations takes 4x time with K7.9 than K7.4 (with exactly same indices and resources).
ES cluster has 3 nodes, all queries take no more than 1 seconde, but dashboard is loading on 10 secondes with Kibana 7.9.
By inspecting AJAX queries, I see K7.4 is using "msearch", whereas K7.9 is using async search. And in async search response, I see "took: 15000ms" !!! Thats really too much.
Is it normal ? Can I rollback to "msearch" strategy ?
I think that the async search is not the problem itself: I found here that the async search is called if the search takes longer than 10 seconds. So, would the call be completed in 1 second the async search would not have been called.
Do you still have the old stack as a backup? Maybe it would help to compare the Explain plans of both versions?
The last AJAX query gives: took: 24558 !!! where as the ES query it-self takes less 1 seconde. The query is just a date-histogram with sub terms aggregation, against 10G data.
To compare, here a screenshot of K7.4, same dashboard:
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.