2

I'm on AWS with my clients for a long time now but I'm going to need cut costs now to keep providing my services. On AWS I'm using RSync to keep some folders synchronized and DRDB to provide high availability with transparent failover having always an operacional and ready to use a mirror for each client machine.

Now I can't keep using DRBD because the much more cheaper cloud solution I'm migrating just provide for each machine an Ubuntu 14.04 LTS with just one partition and without LVM, this cloud platform also become a requirement for some clients of me.

The solution I'm thinking about is schedule shell scripts to daily BKP on one side, transfer it by SSH to another side and restore the BKP, it will become complex, error-prone and a will need a lot of work to develop and manage.

Many of my clients are just Wordpress+Mysql and accept a day of delay, I'm looking for alternatives to provide "high availability" even if it comes with a day of delay that don't force me to develop and manage scripts for each case with the restricted context.

Adam Miklosi
  • 159
  • 1
  • 2
  • 11
Mark
  • 47
  • 6
  • What provider is this? I'd like to be able to see what kind of options you have to work with. – Spooler Sep 11 '18 at 22:45
  • The site is in portuguese language its a brazilian provider: https://configr.com/ – Mark Sep 11 '18 at 22:53
  • Translated to english: https://translate.google.com/translate?hl=pt-BR&sl=pt&tl=en&u=https://configr.com/planos/index.html – Mark Sep 11 '18 at 22:54
  • DRBD would probably work OK over a loopback device, for suitably lax definitions of "work". – womble Sep 12 '18 at 07:19

1 Answers1

1

If you truly cannot viably use a block device (DRBD would probably be better here and you already have experience with it), GlusterFS can provide you with the replication features you seek at the file level.

Gluster "bricks", while ideally a singular storage device with its own thin LVM stack ending in XFS, can really be any POSIX compliant filesystem (or even just a directory rather than a dedicated FS) on a node.

These bricks are aggregated into a unified "volume" with a "replica" policy that defines now many bricks will be written with any given file - in this case probably replica 2 or 3. These replicas will strive to be located on different nodes if at all possible.

The failure semantics with Gluster are not yet as coherent as DRBD. Split-brain conditions are easier to accomplish as data replication is the responsibility of the connecting client (it sends N copies of all writes to each Gluster node rather than writing to a master that then replicates data). However, it can be potentially easier to resolve split-brains with divergent data since each brick is an intact filesystem with completely readable data when using replication.

It won't be as fast as DRBD, but maybe you don't need it to be?

Spooler
  • 7,016
  • 16
  • 29
  • Thanks for the help, I've never heard of Gluster and was searching a lot, for most clients I will use it and once a day by late night then dont need to be so fast. – Mark Sep 11 '18 at 23:08
  • There are a couple other options in the area of distributed file system, such as Ceph. There exist clustered LVM too, but as established you may not be able to use LVM. – John Mahowald Sep 12 '18 at 12:23
  • Unfortunately, clustered LVM does not consider data replication beyond RAID on local machines. It's for accessing already replicated read-write many block devices coherently with a distributed locking system. – Spooler Sep 12 '18 at 13:27
  • For someone with the same problem i discovered latest gluster cant be installed on ubuntu 14.04 LTS, see details here: https://serverfault.com/a/935521/472111 – Mark Oct 15 '18 at 04:22