Very interesting. I'm sure @jpountz and @danielmitterdorfer will be happy to reproduce and find the reason.
If confirmed, may be we should update our documentation to advice people using that on disk.
I guess your cluster is stable so you are not updating the cluster state frequently, right?
The fact is that the synonym file is not updated so it's probably served by the file system cache. Bonus: no needs to parse the cluster state and extract synonym list for every request or keep that data structure in heap. (I didn't look at the code so just assumptions here).
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.