Performance: Reduce the amount of redo copied by PXB
General
Escalation
General
Escalation
Description
Environment
None
Smart Checklist
Activity
Show:
Julia Vural February 19, 2025 at 11:55 AM
It appears that this issue is no longer being worked on, so we are closing it for housekeeping purposes. If you believe the issue still exists, please open a new ticket after confirming it's present in the latest release.
PXB when it copies redo, it does in two parts
Copy all the uncheckpointed redo (till the last block)
start copying new redo (in background thread)
start copying datafiles
ensure the copied redo is upto the log_status.innodb.lsn
stop copying redo
1) was done primarily to handle backups that start when the redo is close to full. the start lsn of PXB (current checkpoint of server) would be soon overwritten. Backup is more likely to fail.
This (1) though causes a problem if there lot of active transactions. the redo will keep on growing and PXB will never find an end of the redo.. (there is new redo that is being generated). It is like trying to catch a moving bus/train/plane
So how bad is 1) It depends on how much uncheckpointed redo is there at start of backup and also how fast the new redo is being generated.
The idea of this ticket is to completely get rid of 1). 8.0 now has redo log archiving feature to prevent redo being overwritten. Specially designed for backups.
We should utilize this feature for 8.0 PXB