[Gluster-users] architecture suggestions
vladkopy at gmail.com
Thu Oct 18 21:18:37 UTC 2018
You can experiment with FUSE mount options: halo-latency & halo-min-replicas
halo-latency: The threshold below which all other clients will
consider children (bricks) connected.
halo-min-replicas: The minimum number of replicas which are to
be enforced regardless of latency specified in the above 3 options.
If the number of children falls below this threshold the next
best (chosen by latency) shall be swapped in.
If you have one brick per site set halo-min-replicas to 1 and halo-latency
very low so it will write to the local brick (watch that fuse connects to
local gluster - it tends to mount to remote on boot);
then files written will "heal" to the other bricks on a background. Heal
performance is pretty solid.
I wouldn't do it for VMs
With my clusters around 10ms max latency I dont mind slow writes to cluster.
Real problem are reads from cluster as it does all attr queries 20 times
and if you have maildir 100k emails in a folder ...
I had to write overlay fuse file system to direct reads directly to the
brick through the kernel, bypassing all gluster fuse stuff:
all reads fly, and writes go regular way
You can also look in to Syncthing - surprisingly made on Go it is pretty
solid. Especially for 2 replicas you want.
Shoot own discovery server on each node or on some cloud and have your
sites sync with file change monitoring 0 delay.
On Wed, Oct 17, 2018 at 1:42 AM Oğuz Yarımtepe <oguzyarimtepe at gmail.com>
> Any idea about the Halo configuration? Didn't find any documentation about
> On Wed, Oct 17, 2018 at 1:10 AM Vlad Kopylov <vladkopy at gmail.com> wrote:
>> if you going for redundancy go for 3 full nodes, arbiter setup seen bugs
>> doubt anything good will come out of using NFS
>> if you doing websites use VMs as hypervisors are using libgfapi
>> or implement libgfapi in your app directly
>> On Sun, Oct 14, 2018 at 2:13 PM Oğuz Yarımtepe <oguzyarimtepe at gmail.com>
>>> I am trying to replicate nfs mount areas between to sites that will work
>>> active active. I was checking the methods that i can use via GlusterFS and
>>> it seems replication should be the way. I am not sure how many nodes should
>>> i use fo replication. I may need to make maintenance on the GlusterFS
>>> servers and i don't want to cause server 500 failures because of
>>> inexistance of mount areas. Any tips? Should i use arbiter?
>>> Oğuz Yarımtepe
>>> Gluster-users mailing list
>>> Gluster-users at gluster.org
> Oğuz Yarımtepe
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Gluster-users