[MDEV-9930] Allow to choose an aggressive InnoDB buffer pool load mode Created: 2016-04-15 Updated: 2021-04-14 Resolved: 2021-03-03 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Storage Engine - InnoDB |
| Fix Version/s: | N/A |
| Type: | Task | Priority: | Major |
| Reporter: | volans | Assignee: | Unassigned |
| Resolution: | Incomplete | Votes: | 1 |
| Labels: | innodb, upstream | ||
| Issue Links: |
|
||||||||||||
| Description |
|
When innodb_buffer_pool_load_at_startup is ON, with a large innodb_buffer_pool_size the time to load the buffer pool is very long. For example with innodb_buffer_pool_size = 378G it took 6.5 hours to load the buffer pool. I understand that this feature is intended to load it in the background without affecting as much as possible the server performances, but for many high traffic production environments the server cannot be re-pooled into production without a reloaded buffer pool anyway. It could be very useful in my opinion to have an option, or to modify innodb_buffer_pool_load_at_startup from boolean to multivalue, to force an aggressive load of the buffer pool using a good part of the available resources. The most flexible implementation of this feature should allow to modify at runtime this value so that if buffer pool load is in progress it will allow to modify it's behaviour between the current one and the aggressive one. This feature will allow to reduce a lot the time a server has to be taken out of production for maintenance on any heavy traffic environments where modern hardware with large memories is in use. |
| Comments |
| Comment by Daniel Black [ 2018-01-01 ] |
|
is setting innodb_buffer_pool_dump_pct since 10.1.10 sufficient? Which resource limitation are you hitting? Maybe some multi threaded (reflecting innodb_buffer_pool_instances ) load could be done? I'm also curious of your innodb_buffer_pool_instances/innodb_buffer_pool_chunk_size setting? |