Ensure db1, db3 and db4 use the same disk setup
Right now db1 (the primary) uses a different and less efficient disk/LVM setup compared to the secondaries. In turn (probably due to this) db1 is having very high IO wait timings (often in the order of seconds), compared to the secondaries which have next to no IO wait. I suspect this is also the cause of writes on the primary sometimes being fast, and sometimes taking seconds (for the same query when you're already connected).
Changing the disks either requires a failover, or downtime since I don't think we can do this in-place.