This is obviously a rhetorical question. I see no reason why this MySQL System Variable is not (yet) dynamic. Obviously, changing the value of this variable could need replication to be stopped (like for slave_parallel_type, slave_parallel_workers and slave_preserve_commit_order), but at least it would not need a MySQL restart.
I have been faced many times with situations where changing relay_log_space_limit would have been the right solution, but not being able to do it without restarting MySQL made my life harder. I faced this again recently, its annoyance grew even more, so I am writing this rant !
I also opened Bug#92891 - Please make relay_log_space_limit dynamic - for tracing this. I was not able to find a feature request specifically referring to this but it is mentioned by some bugs. In one of these - Bug#35611 - it is said (on March 13, 2012) that relay_log_space_limit should be made dynamic, but this comment seems to have been somehow forgotten (this bug from 2008 is still open...). In another one - Bug#68236 from 2013 - the following is mentioned: "that setting is not dynamic and so it loses all it's (sic) use".
The situation where I would have liked this variable to be dynamic is as follows:
- I am keeping 1 day of binlogs on a slave in a log partition (binlog + relay logs),
- I want to delay this slave by 24 hours with delayed replication,
- Adding 1 day of relay logs would fill the log partition,
- There is no use of keeping 24 hours of relay logs on that delayed slave as the master is already keeping all the needed binlogs (7 days in my case),
- The right solution would have been to set relay_log_space_limit to a few GB,
- But changing relay_log_space_limit would have needed to restart mysqld,
- So I ended-up throwing more disk space at the log partition (I am not proud of this, but lucky GCP: you are selling a little more disk space now).
No comments:
Post a Comment