> This is actually a problem if a new TSDictionary is created, in a different > schema specified by the dumped search_path setting.
Just out of curiosity, do you have a concrete test case where it failed that way? AFAICS the emitted SQL would be like
SET search_path = schema1, pg_catalog;
CREATE TEXT SEARCH DICTIONARY somedict (...);
COMMENT ON TEXT SEARCH DICTIONARY somedict IS '...';
SET search_path = schema2, pg_catalog;
CREATE TEXT SEARCH DICTIONARY somedict (...);
COMMENT ON TEXT SEARCH DICTIONARY somedict IS '...';
so it should accidentally work anyway. It's possible that a parallel restore would get it wrong, or that a schema-selective restore would omit comments it should include, but I couldn't reproduce a failure in simple cases.
Yep, the issue is reproducible during parallel restore - I've seen this for
4 parallel processes. Sorry for having not mentioned this before.