I recently reindexed our full inventory data set at Chicago, and found that the reindex failed on the instance_contributors_idx with this error: "index row size 3256 exceeds maximum 2712 for index". Turns out that we have around 6000 instances with contributors whose aggregate length exceeds 2712 characters. As a work around, I removed the index and completed the reindex with no other problems. And the application, including searches, seemed to work just fine without it, presumably because each of these jsonb collections also have gin and full text indexes defined for them as well. It seems that trying to create a btree index on contributors, converting the whole array of contributors to text and using the text as the index key, is too brittle to accommodate the wide range of contributors that research libraries typically have. The same may be true for btree indexes on other types of jsonb collection, as well. Also, I don't actually see any need for this type of index.