ElasticSearch refresh effect on cache
A refresh causes new documents to be visible for searching. This happens by writing a new index segment. A new segment can also be made by merging old large ones.
Filter- and field caches are managed per segment, since a segment is immutable. You can use the warmer APIs to ensure caches are pre-warmed before being made available for search. If not, then parts of the cache is essentially "cleared".
A flush in Elasticsearch terms actually calls a Lucene commit. This is quite more expensive.
If you have a write heavy app, you probably want to increase the refresh interval to get better indexing throughput.
There's some more details about these things in these two articles: