If a document is added to a collection with a field whose value contains an invalid UTF-8 string, creation of a text index on that field fails with the error "text contains invalid UTF-8". Creation of a normal index succeeds.

(In the scenario I tested, the invalid UTF-8 is one where only the high surrogate of a surrogate pair is included in the string.)

Given that the server allows invalid UTF-8 strings to be inserted into the database, we should consider whether the server should be more resilient to the presence of invalid UTF-8 strings when creating text indices.

It would also be reasonable to close this as Works as Designed, as I imagine it's fairly rare for this to happen in practice.