MySQL 5.7: Sharp checkpointing due to binlogs?

[Thomas Bätzler]

Hi,

we're running some moderately busy replicated database servers. We've upgraded to MySQL 5.7 recently (running mysql-server 5.7.34-1debian10) and we've been seeing occurrences of what might be described as "hard checkpointing" - i.e. all tables locked, no new connections accepted - several times a day. We suspected this was happening on ib_logfile rotation. The same effect could be manually triggered by running mysqladmin flush-logs. By looking at the monitoring info we had on the systems we saw read i/o peak during those checkpointing events. Now apparently purging old binary logs has solved this problem for us - after purging the binlogs to the current file (instead of keeping 3 days' worth around), mysqladmin flush-logs runs in milliseconds. Can somebody please explain what's going on here?

TIA!