Since I got the TokuDB version with the background analyze code, I have enabled it and performed the necessary steps to force a row recount on existing TokuDB tables.
This has worked fine, however on a half dozen occasions across multiple replicas I have had tables lose row counts on restarts (usually when I'm patching, rebooting, upgrading, etc)
So I know this because I end up with log entries like this straight after the restart:
2016-09-22 19:19:43 140259201485568 [Note] TokuDB: Auto scheduling background analysis for <table name>, delta_activity 1 is greater than 30 percent of 0 rows. - succeeded.
And of course this appears a lot because every row update triggers the rule.
I then force another recount of this table, and it goes back to normal.
Not sure what other info can help.. let me know.
MariaDB 10.1.17 (also affected 16 I believe, which had previous TokuDB version)