View source for Db20 to NFS1-2 migration plan

NFS home DRBD migration plan.

== Hosts ==
* db20 - old server
* nfs1 - fresh host
* nfs2 - host replicating from db20

== Differences ==
* Move replication from db20->nfs2 to nfs2->nfs1
* Use protocol C, not A

== Plan ==

=== Preparation ===
* Check DRAC access
* Prepare nfs1 for drbd
* Add DNS entries for new internal IPs
* Set up cross link networking between nfs1/nfs2
** Protocol C, not A
** replication rate 110M
* Write final configuration files for nfs1/nfs2
* Install <tt>/etc/exports</tt> on nfs1/nfs2
* Install NFS kernel server on nfs1/nfs2

==== IP plan ====
 ; Cross link between nfs1/nfs2: 10.3.0.0/30
 
 1       1H      IN PTR  nfs1-xlink0.pmtpa.wmnet.
 2       1H      IN PTR  nfs2-xlink0.pmtpa.wmnet.

 nfs1-xlink0     1H      IN A            10.3.0.1
 nfs2-xlink0     1H      IN A            10.3.0.2

==== New drbd.conf ====
<pre>
# drbd.conf for replication between nfs1 and nfs2
# Written on 2009/11/23 by Mark Bergsma <mark@wikimedia.org>

global {
	usage-count no;
}

common {
	protocol C;
}

resource nfshome {
	on nfs1 {
		device /dev/drbd1;
		disk /dev/mapper/mirror-nfshome;
		address 10.3.0.1:7789;
		meta-disk internal;
	}

	on nfs2 {
		device /dev/drbd1;
		disk /dev/mapper/mirror-nfshome;
		address 10.3.0.2:7789;
		meta-disk internal;
	}
}
</pre>

=== Migration ===

'''On nfs1''':
* Install new <tt>/etc/drbd.conf</tt>
* Create device meta-data:
 drbdadm create-md nfshome
* Attach and setup new resource, but do not connect yet:
 drbdadm attach nfshome
 drbdadm syncer nfshome

'''On db20''':
* Stop NFS kernel server
* Remove NFS service IP <tt>10.0.5.8</tt>:
 ip addr del 10.0.5.8/32 dev eth0
* Stop userland processes using <tt>/home</tt> (<tt>fuser -m</tt>)
* Unmount <tt>/home</tt>
* Comment <tt>/home</tt> mount in <tt>/etc/fstab</tt>
* Check consistency of nfshome replication (<tt>cat /proc/drbdadm</tt>)
* Disable resource:
 drbdadm down nfshome
* Save the meta-data in a text file:
 drbdadm dump-md nfshome > /root/db20-nfs2-drbd.metadata

nfs2 should now have an exact block level replica of the /home filesystem

'''On nfs2''':
* Check consistency of nfshome replication (<tt>cat /proc/drbdadm</tt>)
* Disable resource:
 drbdadm down nfshome
* Save the meta-data in a text file:
 drbdadm dump-md nfshome > /root/db20-nfs2-drbd.metadata
* Install new <tt>/etc/drbd.conf</tt>
* Create device meta-data:
 drbdadm create-md nfshome
* Enable the new resource:
 drbdadm attach nfshome
 drbdadm syncer nfshome
 drbdadm connect nfshome

'''On nfs1''':
* Connect the new resource:
 drbdadm connect nfshome

'''On nfs2''':
* Start initial device synchronisation (ON NFS2 ONLY!):
 drbdadm -- --overwrite-data-of-peer primary nfshome
* Check replication status (cat /proc/drbd)
* Add <tt>/etc/fstab</tt> entry:
 /dev/drbd1	/home           jfs     defaults        0       2
* Mount <tt>/home</tt>:
 mount /home
* Migrate NFS service IP <tt>10.0.5.8</tt> to nfs2:
 ip addr add 10.0.5.8/32 dev eth0
* Configure service IP on nfs2 on bootup. Add entry to <tt>/etc/networking/interfaces</tt> eth0:
 up ip addr add 10.0.5.8/32 dev $IFACE
* Start NFS kernel server

=== Wrap up ===
* Migrate udp2log, /usr/local/bin/demux.py from db20 to nfs1 and nfs2
* Write documentation on wikitech about the DRBD setup

=== Rollback ===
If anything goes wrong, the filesystem on <tt>/dev/drbd1</tt>, or alternatively <tt>/dev/sda6</tt> should be in tact and can be mounted. NFS service can then be restored quickly.

[[Category:Migration plans]]