Hello,
Yes, this is Debian Jessie with sysv-init, not systemd.
I prefer my servers to be deterministic.
Firstly an issue with /var/run/ceph.
The init.d/ceph script has these lines:
---
if [ ! -d $run_dir ]; then
# assume /var/run exists
install -d -m0770 -o ceph -g ceph /var/run/ceph
fi
---
Which should do the right thing.
However with 10.2.6 (and probably before) under Debian Jessie something
else seems to create the directory with root ownership before the startup
script runs.
So the above check finds the directory and never installs it.
Which consequently leads to a failure to start any services.
Removing the check and making the "install" unconditional fixes this.
http://tracker.ceph.com/issues/19242
The second issue is related in the sense that it's also caused by the
desired ceph:ceph ownership of things.
My OSDs are all manually deployed since at the very least the SSD based
ones share the disk with OS/swap partitions.
And mounted via fstab, no ceph-deploy, GUID and udev magic.
All this works fine with Hammer and Jewel when running as root with
"setuser match path = /var/lib/ceph/$type/$cluster-$id" in ceph.conf.
Of course this fails predictably when trying to do things the Jewel way as
ceph:ceph when it comes to external journals, as ownership is not
preserved with udev between reboots.
Last time I checked, ceph-deploy still doesn't understand dealing with
just a partition (for the data) instead of a full disk.
So from where I'm standing the alternatives seem to be:
a) drink the ceph-deploy cool-aid, buy more SSDs to house the OS.
b) run everything as root, until kingdom come.
c) spend loads of time to construct something fragile that will do the
right thing via udev rules.
Correct?
Christian
--
Christian Balzer Network/Systems Engineer
chibi@xxxxxxx Global OnLine Japan/Rakuten Communications
http://www.gol.com/
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
