If a new dataset publish via the publisher client fails (and is rolled back) a search index update appears to be created for the failed dataset.

This can cause the search UI to fail and produce an Internal Server Error on the search results page as the corresponding dataset_id is not found.

This was observed when publishing recent CMIP5 datasets. Steps to reproduce:

1) Publish dataset via esgpublish which fails (due to duplicate file.)
2) Select search facets which would result in failed dataset in search result.
3) ISE generated when constraining facet is selected.