According to this link, https://mariadb.atlassian.net/browse/MDEV-4662 Jesper is correct and I am wrong: when InnoDB warns about not enough space in the logs, it’s overwritten its last checkpoint. Thanks to Jeremy Cole for finding out the truth.
Comment: Sphinx search performance optimization: attribute-based filters
we use sphinx to our website where we have about 1200 tables each with … described here, but the search results have gone from milliseconds to more than 43 seconds

