Migrating to Debian 12 should have been a routine update but we were met with a silent IO storm that slammed our disks with over 500 MB/s of writes. MariaDB stalled and Redis began complaining about slow fsync as our storage ground to a halt.
- The kernel changed how it divides dirty page budgets across containers.
- Moving from cgroup v1 to v2 altered the writeback path in unexpected ways.
- A tiny 0.6 MB limit in the budget became a total production catastrophe.
- We had to dig into how dirty page writeback actually works to find the fix.
It was a stressful way to learn about cgroup v2 and kernel defaults. If you are planning a migration or seeing weird IO behavior, this might save your production environment.
Read the full article here:
Top comments (0)