On 25/10/2016 7:35 AM, Steven Vacaroaia wrote:
I am using bricks
Here is the info
Ok, thanks for the info. Could you post a zpool status as well? Also how
much RAM do your nodes have?
with zfs the following are essential for gluster usage:
- sync=standard
- xattr=sa
- acltype=posixacl
sync = disabled is very dangerous! you run a severe risk of gluster
corruption if the node crashes or power goes out.
Recommended:
- compression=lz4
- atime=off
Compression can actually speed up io as less data is written/read to the
disk.
Gluster:
I really recommend a replication factor of 3 for any replicated
filesystem. You run a much reduced chance of split-brain issues. If you
can't afford a 3rd disk set then gluster has "arbiter" nodes
(http://staged-gluster-docs.readthedocs.io/en/release3.7.0beta1/Features/afr-arbiter-volumes/)
Also for VM Hosting the following are *required* settings for gluster
volumes:
cluster.quorum-type: auto
cluster.server-quorum-type: server
performance.quick-read: off
performance.read-ahead: off
performance.io-cache: off
cluster.eager-lock: enable
network.remote-dio: enable
And I'd recommend the following:
performance.strict-write-ordering: off
performance.stat-prefetch: on
cluster.locking-scheme: granular
cluster.granular-entry-heal: yes
performance.readdir-ahead: on
cluster.data-self-heal: on
Additionally I'd look at using sharding, it makes a huge difference to
heal times:
features.shard: on
features.shard-block-size: 64MB
It breaks the VM's up into shards which can be healed sperately, rather
than the entire VM image.
--
Lindsay Mathieson
_______________________________________________
pve-user mailing list
[email protected]
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
