performance of contextquery with many (~1000) contexts?

Hi all,

I'm trying to build a (elastic) suggester that uses context in
completionqueries to implement authorization for these suggestions.
Basically, I only want suggestions from the contexts where the user has
rights.
(not sure if this is the best way, suggestions (no pun intended) welcome)

What I observe is that if the number of contexts in the query increases,
retrieving the suggestions becomes very slow. Typically a user can have up
to 1000 access groups (yes it's complex, yes it should be reduced).

In a previous project, we used plain Lucene to build a OR query for these
access groups and that was still fast. I tried to follow the Elastic source
and at some point it leads to code that generates a ContextQuery.

So, that leads me to wonder why using these contexts in CompletionQuery
makes it so slow.

Do you have any hits?

Thanks!
-Rob