drbd storage driver¶
Introduction¶
Drbd is a Linux kernel driver presenting shared block devices from pairs of local block devices. Replication data goes over the network and can be synchronous ou near-synchronous. It supports the active-active or active-passive operation modes. The former being a prerequisite for cluster filesystem.
OpenSVC does not configure the drbd framework. It just handles the actions needed upon service 'start' and 'stop'.
Node configuration¶
The drbd resource configuration is expected to reside in /etc/drbd.d/$res.res.
If a vg is layered above drbd, a specific LVM configuration is recommended (link)
Set write_cache_state = 0 in /etc/lvm/lvm.conf
Remove /etc/lvm/cache/.cache
Files contributed to sync#i0¶
/etc/drbd.d/${res}.res
This file is thus replicated from the node running the service to the other nodes and drpnodes.
Command set¶
The drbd driver is a member of the 'disk group' class.
start
Promote to 'primary' state
stop
Demote to 'secondary' state
Tags¶
prevg
Upon service 'start', drbd 'start' is scheduled before volume group 'start'. To use when the volume group is layered over the drbd.
postvg
Upon service 'start', drbd 'start' is scheduled after volume group 'start'. To use when the the drbd is layered over the volume group.
Syntax¶
##############################################################################
#
# 'drbd' resource
# Linux only. Set up the attachment, connection and role of a defined drbd
# resource. Depending on weather this resource is stacked over or under
# other disk group resources we need to start it respectively late or early.
# This is controlled by a set of tags: 'prevg', 'postvg'.
#
[drbd#0]
#
# 'res'
# mandatory. String. The name of the drbd resource associated with this
# service resource. OpenSVC expect the resource configuration file to
# reside in '/etc/drbd.d/resname.res'. The 'sync#i0' resource will take
# care of replicating this file to remote nodes.
#
res = data
#
# 'always_on'
# state the expected status on nodes specified as value is 'up'. With drbd
# the 'up' status is granted when the drbd driver reports UpToDate/UpToDate,
# so always_on should point to all nodes participating in the drbd resource.
#
always_on = drpnodes nodes
#
# 'tags'
# prevg: upon service 'start', drbd 'start' is scheduled before volume
# group 'start'. To use when the volume group is layered over the
# drbd.
# postvg: upon service 'start', drbd 'start' is scheduled after volume
# group 'start'. To use when the the drbd is layered over the
# volume group.
#
tags = prevg
Examples¶
In this example the service drives an OpenVZ container installed in a filesystem (fs#1) build on a drbd (drbd#1, resource name 'data'). OpenSVC Also handles the svc/etc/init.d/foo start-up script inside the vz container.
Start¶
root@deb1:/# svcmgr -s drbdsvc.opensvc.com start
DRBD#0 INFO drbd resource data is already up
DRBD#0 INFO drbdadm primary data
FS#1 INFO e2fsck -p /dev/drbd0
FS#1 INFO output:
/dev/drbd0: clean, 27663/131072 files, 161431/524263 blocks
FS#1 INFO mount -t ext3 -o defaults /dev/drbd0 /drbdsvc
VZ INFO vzctl start 101
VZ INFO output:
Starting container ...
Container is mounted
Adding IP address(es): 192.168.0.204
Setting CPU units: 1000
Set hostname: drbdsvc.opensvc.com
Container start in progress...
VZ INFO wait for container up status
VZ INFO wait for container operational
APP INFO spawn: vzctl exec 101 /svc/etc/init.d/S10foo start
APP INFO start done in 0:00:00.068714 - ret 0 - logs in /var/tmp/opensvc/svc_drbdsvc.opensvc.com_S10foo.log
Stop¶
APP INFO spawn: vzctl exec 101 /svc/etc/init.d/K90foo stop
APP INFO stop done in 0:00:00.039901 - ret 0 - logs in /var/tmp/opensvc/svc_drbdsvc.opensvc.com_K90foo.log
VZ INFO vzctl stop 101
VZ INFO output:
Stopping container ...
Container was stopped
Container is unmounted
VZ INFO wait for container down status
FS#1 INFO umount /drbdsvc
DRBD#0 INFO drbd resource data is already up
DRBD#0 INFO drbdadm secondary data