> > Any interest in using csi with ceph for stateful tasks? > > I would! We already use ceph in mesos but via rexray. I was thinking to > take a deeper look into to csi driver therefore I can remove rexray from > our stack. > >
I found rexray to be a bit buggy with releasing rbd's on time, making it difficult for tasks to move to different hosts. But I tested this maybe more than a year ago. Because I want to limit linux capabilities on tasks I have decided to configure the external/unmanaged csidriver. [@c04 mesos-csi]# tail /etc/rc.local # mesos csi nfs plugin /usr/libexec/csi/csinfs-start.sh # mesos csi ceph plugin /usr/libexec/csi/csiceph-start.sh [@c04 mesos-csi]# cat /usr/libexec/csi/csiceph-start.sh #!/bin/bash # umask 0077 export CSI_DEFUSERID="app.mesos.test" export CSI_DEFUSERKEY="KKKKKEEEEEYYYY==" mkdir /tmp/mesos-csi-uuh6Ra 2>/dev/null cd /tmp/mesos-csi-uuh6Ra /usr/libexec/csi/csiceph -endpoint 'unix:///tmp/mesos-csi-uuh6Ra/endpoint.sock' -nodeid $HOSTNAME -type rbd -drivername ceph-csi-rbd -v 10 -log_file=./csiceph-plugin.log -logtostderr=false >stdout 2>stderr & You can give it a try. Be carefull using the original csi driver, it overwrites your /etc/ceph/ceph.conf https://github.com/f1-outsourcing/csiceph/tree/master