Fwiw we discovered the k8s worker this was running on was using root disk on slow ceph (Ceph with HDD behind it). We switched it to using a ceph pool with SSD and the problems disappeared. This confirms the suspicion that it's not an inherent problem with juju/pebble, but only one that manifests under certain poor infrastructure performance conditions.
Fwiw we discovered the k8s worker this was running on was using root disk on slow ceph (Ceph with HDD behind it). We switched it to using a ceph pool with SSD and the problems disappeared. This confirms the suspicion that it's not an inherent problem with juju/pebble, but only one that manifests under certain poor infrastructure performance conditions.