Indeed, have become painfully aware. Until now that cluster was rock solid. Installed, and only needed to take it into maintenance once in 4 years to upgrade the disks. Will look at upgrading, but that will take some time as migrating from 2.1 directly to 4.x is not possible. If I recall well, I have to go through 2 intermediate versions, plus forced OS upgrades, and we might need another disk upgrade, as 3.x+ has shown to be having an approximately 3 times higher I/O load. And it all has to go through QA, and it all must be done without downtime, under load/live, and without data loss. In the meantime, I will try to debug this situation, will give feedback, and stick to 2.7.0 as client.