My specific scenario is: 20TB mysql instance with about 400 innodb tables. From that many tables there are only a few very large (>2TB). The tables sizes are varaying 1-100G.
The time it takes to do full xtrabackup is scaling by using the parallel parameter up to 10-12. But at the last 1-2 hours of the backup there is only one table left running and the speed is low. Practically its single threaded.
Currently I dont see any logic of how the tables are ordered.
So my suggestion here would be that if the tables are ordered from largest to smallest than the whole time needed to complete will be much lower because most of the time the xtrabackup will run multithreaded.
Hello, we are aware of the issue and the JIRA: PXB-3502 Jira . I hope it will be fixed in future releases, but I cannot promise an exact date.
Thank you for your quick reply. Fingers crossed it will be completed 