Setting such insane limits is a work around to let the cluster recover. A production environment
should use more conservative limits, actually the kernel defaults are OK. Having > 1000 threads
per an OSD is abnormal, we should find out what is (was) causing such a misbehavior.
Roman,
Setting such insane limits is a work around to let the cluster recover. A production environment
should use more conservative limits, actually the kernel defaults are OK. Having > 1000 threads
per an OSD is abnormal, we should find out what is (was) causing such a misbehavior.