The fault has been fixed.

Elasticsearch indexes have rebuilt and search is working as intended.

Issues has been discovered; one of the indexes was behaving in strange ways and causing regular errors/retries. After recreating the index, cluster is now behaving as expected. Index is currently rebuilding from current data.

Performance issues were a combination of a suspected spam attack, and some imbalance with the kubernetes nodes. Spam sources have been blocked and nodes have been rebalanced.