There are different execution paths that can be done when executing a distributed search. The distributed search operation needs to be scattered to all the relevant shards and then all the results are gathered back. When doing scatter/gather type execution, there are several ways to do that, specifically with search engines.

One of the questions when executing a distributed search is how much results to retrieve from each shard. For example, if we have 10 shards, the 1st shard might hold the most relevant results from 0 till 10, with other shards results ranking below it. For this reason, when executing a request, we will need to get results from 0 till 10 from all shards, sort them, and then return the results if we want to insure correct results.

Another question, which relates to search engine, is the fact that each shard stands on its own. When a query is executed on a specific shard, it does not take into account term frequencies and other search engine information from the other shards. If we want to support accurate ranking, we would need to first execute the query against all shards and gather the relevant term frequencies, and then, based on it, execute the query.

ElasticSearch is very flexible and allows to control the type of search to execute on a per search request basis. The types are:

p. The most naive (and possibly fastest) implementation is to simply execute the query on all relevant shards and return the results. Each shard returns size results. Since each shard already returns size hits, this type actually returns size times number of shards results back to the caller.

p. The query is executed against all shards, but only enough information is returned (not the document content). The results are then sorted and ranked, and based on it, only the relevant shards are asked for the actual document content. The return number of hits is exactly as specified in size, since they are the only ones that are fetched. This is very handy when the index has a lot of shards (not replicas, shard id groups).

p. Same as "Query And Fetch", except for an initial scatter phase which goes and computes the distributed term frequencies for more accurate scoring.

p. Same as "Query Then Fetch", except for an initial scatter phase which goes and computes the distributed term frequencies for more accurate scoring.