Comment 4 for bug 1578827

> I somewhat doubt it. A typical OSD spawns ~ 100 threads

It depends much on the cluster size and the current cluster status. See:

http://docs.ceph.com/docs/master/rados/troubleshooting/troubleshooting-osd/#osd-not-running

We had a customer who claimed that increasing kernel.pid_max in their case helped to recover from this error.

We also have a blueprint for this:

https://blueprints.launchpad.net/fuel/+spec/ceph-osd-nodes-kernel-pid-max

"Not enough ram" is also possible reason, though in production usually segfaults observed in this case (with corresponding messages in dmesg), not create thread assertion failure.