Affects Version/s: 8.0.17-8
Fix Version/s: None
Ubuntu 18.04 LTS Desktop, Kernel 5.0, Ryzen 3970x (32 cores with HT), 128GB RAM, Intel Optane . Database version is actually 8.0.18 but was not available as select option.
10000 tables, all TokuDB, 10 columns per table, only primary key, 64 bit long, no other indexes
Insert scenario is batch mode (2000 rows at once), once per minute for every table. During insert, the database eats more than 50% of whole CPU sitting in kernel (native_queued_spin_lock_slowpath).
The issue appears both with or without tokudb_directio set and is fully absent from InnoDB. The kernel load is not constant but comes and goes (see netdata picture attached)
Below are TokuDB parameters used (anything is default):
tokudb_row_format = TOKUDB_ZLIB
tokudb_cache_size = 32768M
tokudb_lock_timeout = 30000
tokudb_checkpointing_period = 1800
tokudb_fs_reserve_percent = 1
tokudb_directio = 1