[Gluster-devel] Async replication

Gordan Bobic gordan at bobich.net
Thu May 20 20:38:01 UTC 2010

phil cryer wrote:
> I'm testing with lscynd now to sync remote clusters (ours running
> gluster, others may be, may not be) and it's working well.

Just make sure you are running the latest trunk version. The exclude 
file handling bugs were fixed about 5 hours ago. :-)

> Also look
> at http://code.google.com/p/pylsyncd/ which boasts:
> "The main advantage of pylsyncd against lsyncd is that it uses message
> queues in order to synchronize in a parallel way several destination
> servers, saving up time when it is required to have more than one
> destination. It has been tested in heavy loaded environments."

I saw it, but lsyncd looked better to me for the following reasons:

1) Speed - it's written in C.

2) lsyncd's rsync hooks are fully configurable - both the binary and the 
parameters. You don't have to use rsync, that's just the default. I am 
currently experimenting with using csync2, as that provides the very 
feature you mentioned above - it is specifically designed for 1->many 
cluster synchronisation.

3) I haven't managed to find any documentation on pylsyncd's handling 
for excluding parts of the watched subtree, so I have to assume it 
cannot do this. lsyncd understands a subset of rsync's exclude pattern 
syntax so it both omits the excluded files from the watch and passes the 
same exclude file to rsync. This is essential in my use case as I want 
to mirror root directories across servers in a cluster.


More information about the Gluster-devel mailing list