Where you decide to create indices is going to be very application/use case dependent.
Assuming that you're working on a search use case where you will be explicitly controlling mappings, you'll create the index before indexing any documents into it, including the mapping in the index creation request.
It could well be that you create the index on startup, perhaps checking if the index exists first and if not, creating it. You may go further and also check that the mapping in the index conforms to the one that you expect (I haven't typically gone this far as mappings may change over time and outside of the application itself).
If this is also a search use case with a single index, I would recommend that your application uses an alias to refer to the index. In doing so, new versions of the search index can be created at any time (perhaps with refined mappings as you iterate on your search strategy) and then once you are happy with a new index, simply swap the alias over to point to that new index, the result being that your application will now be using the new index with zero downtime.
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.