Re: Is it possible to start geo-replication between two volumes with data already present in the slave?

[Aravinda <avishwan@xxxxxxxxxx>]

When you directly sync files using rsync, rsync creates a file in slave if not exists. Due to this, rsync creates a file in slave with different GFID than the one available in master. This is a problem for geo-rep to continue syncing to slave.

Before starting geo-replication as a prerequisite steps you can do the following to fix the GFID changes,

Run this in a master node,
If you downloaded glusterfs source directory,

    cd $GLUSTER_SRC/extras/geo-rep
    sh generate-gfid-file.sh localhost:<MASTER VOL NAME> $PWD/get-gfid.sh  /tmp/master-gfid-values.txt

Copy the generated file to slave
    scp /tmp/master-gfid-values.txt root@slavehost:/tmp/

Run this script in slave,
    cd $GLUSTER_SRC/extras/geo-rep
    sh slave-upgrade.sh localhost:<SLAVE VOL NAME> /tmp/master-gfid-values.txt $PWD/gsync-sync-gfid

Once all these steps complete, GFID in master volume matches with GFID in slave.

Now update the stime xattr in each brick root in master volume. Enclosed a Python script to update stime of each brick root to current time, run it in each master node for each brick.

    sudo python set_stime.py <MASTER VOLUME ID> <SLAVE VOLUME ID> <BRICK PATH>

For example,

    sudo python set_stime.py f8c6276f-7ab5-4098-b41d-c82909940799 563681d7-a8fd-4cea-bf97-eca74203a0fe /exports/brick1

You can get master volume ID and slave volume ID using gluster volume info command,

    gluster volume info <MASTER VOL> | grep -i "volume id"
    gluster volume info <SLAVE VOL> | grep -i "volume id"

Once this is done, create the geo-rep session using force option,

    gluster volume geo-replication <MASTERVOL> <SLAVEHOST>::<SLAVEVOL> create push-pem force

Start geo-replication,

    gluster volume geo-replication <MASTERVOL> <SLAVEHOST>::<SLAVEVOL> start force

Now onwards, geo-rep picks only new changes and syncs to slave.

Let me know if you face any issues.

--
regards
Aravinda
http://aravindavk.in


On 12/14/2014 12:32 AM, Nathan Aldridge wrote:

Hi,

 

I have a large volume that I want to geo-replicate using Gluster (> 1Tb). I have the data rsynced on both servers and up to date. Can I start a geo-replication session without having to send the whole contents over the wire to the slave, since it’s already there? I’m running Gluster 3.6.1.

 

I’ve read through all the various on-line documents I can find but nothing pops out that describes this scenario.

 

Thanks in advance,

 

Nathan Aldridge

_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://supercolony.gluster.org/mailman/listinfo/gluster-users

#!/usr/bin/env python

import struct
import xattr
import sys
import time


def get_stime(master_uuid, slave_uuid, brick):
    stime_key = "trusted.glusterfs.%s.%s.stime" % (master_uuid, slave_uuid)
    try:
        return struct.unpack('!II', xattr.get(brick, stime_key))
    except (OSError, IOError):
        return "N/A"


def set_stime(master_uuid, slave_uuid, brick):
    stime_key = "trusted.glusterfs.%s.%s.stime" % (master_uuid, slave_uuid)
    t = time.time()
    sec = int(t)
    nsec = int((t - sec) * 1000000)
    xattr.set(brick, stime_key, struct.pack("!II", sec, nsec))


if __name__ == "__main__":
    """
    Usage:

    sudo python set_stime.py f8c6276f-7ab5-4098-b41d-c82909940799 \
    563681d7-a8fd-4cea-bf97-eca74203a0fe /exports/brick1
    """
    print "BEFORE: %s" % repr(get_stime(sys.argv[1], sys.argv[2], sys.argv[3]))
    set_stime(sys.argv[1], sys.argv[2], sys.argv[3])
    print "AFTER: %s" % repr(get_stime(sys.argv[1], sys.argv[2], sys.argv[3]))

_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://supercolony.gluster.org/mailman/listinfo/gluster-users