Comment 0 for bug 1536271

Revision history for this message
Miroslav Anashkin (manashkin) wrote :

Fuel deploys the nodes with the default kernel.pid_max=32768 value.

However, normally Ceph OSD node may require more than 1000 PIDs per OSD in idle state, running on the node. This number of required PIDs increases with the number of placement groups configured and with the load to OSDs.

Lack of allowed PIDs leads to all running OSD restart on the node. Such restart may sometimes may trigger cascade OSD node failures.

Related Ceph issue: http://tracker.ceph.com/issues/10988 (please go directly to the several last posts after reading the description)

I propose to change Fuel Library to set kernel.pid_max=4194303 out of the box for each deployed node.

Workaround is to increase the allowed PID number manually, it is possible on the fly with:

`sysctl -w kernel.pid_max=4194303`
`echo 4194303 > /proc/sys/kernel/pid_max`