It will be solved as the separate ticket (and separate PR), to limit changes of code for solving 1 problem per Jira ticket (2 issues are similar, but handled by different parts of code)
Steps to reproduce:
- create table with STATS_PERSISTENT=0 clause (or alter table used in
PS-6113to use transient statistics (I used following SQL to create table where ANALYZE TABLE takes 30 secs - not enough for crash standard server, but enough to investigate after hacking srv_fatal_semaphore_wait_threshold variable in InnoDB code)
- set global innodb_stats_transient_sample_pages=30000 (it has to be huge value)
Server does not crash. Select query works.