As part of an application-upgrade, we need to dump the existing production database, then load the data into the new DB (on the same server), which will be upgraded.
The dumping takes under a minute, yielding a 1.3G .sql file. The loading of it back takes about 30 hours...
We use these settings to dump:
[mysqldump]
extended-insert=false
lock-tables=false
single-transaction=false
I tried raising the innodb_buffer_pool_size from 1G to 7G, and the innodb_buffer_pool_instances from 1 to 2 -- to no perceptible effect.
When the loading begins, mysqld-process' CPU-consumption jumps to about 100% briefly, but then drops down to about 2-4%. According to iostat, the disk I/O is under 1M/s after that. If it is neither CPU, nor disk I/O, nor network (same host, communicating via Unix socket), what else can it be?
The host runs CentOS Linux release 7.9.2009, and has 32G of RAM (of which only 3G are actually used right now). The server is installed via MariaDB-server-10.3.28-1.el7.centos.x86_64.
Thank you!