On Tue, 12 Jan 2016 10:44:16 PM Marco d'Itri wrote:
> On Jan 12, Piotr Robert Konopelko <piotr.konopelko@moosefs.com> wrote:
> > If you loose Master Server, user action is needed: he can run another
> > Master Server e.g. basing on medatada collected on Metalogger (sometadata
> > is *not* lost) or "repair" broken Master Server.
>
> In other words, the system is not highly available.
In LizardFS master (i.e. metadata) server is not shipped with built-in HA.
However master daemons are ready for failover -- in case of master server
failure one just needs to take over master's IP address and reload shadow
master configuration to promote it to authoritative master. Failover takes
only few seconds at most. It can be implemented with ucarp, corosync, etc.
using trivial (or not so trivial) scripts. Proper built-in HA is on the
LizardFS roadmap.
I suppose MooseFS might be using similar failover mechanism which leaves
particularities of HA implementation to sysadmin.
--
Cheers,
Dmitry Smirnov
GPG key : 4096R/53968D1B
---
Truth — Something somehow discreditable to someone.
-- H. L. Mencken, 1949
Attachment:
signature.asc
Description: This is a digitally signed message part.