I have set up a 3-node PVE+Ceph node and recolated it without issues.

It is true that when powering on the cluster care must be taken so that all nodes start within about a minute; I tuned grub waiting seconds for this, so that fast nodes wait for slow nodes.

El 16/09/16 a las 14:02, Adam Thompson escribió:
We've observed that if any of the nodes boot much faster or slower than the 
other nodes, this causes big problems with both CEPH and PVE, particularly with 
quorum issues.
I've just finished switching a 9-node cluster to NFS because CEPH was too 
unreliable after repeated power failure crashes.
Turns out powering off the last few nodes is hard because they've lost quorum 
by that point and hang during shutdown for longer than the UPSes last.
Unless you have redundant power (I.e. generator) I'm not sure I would ever 
recommend a large PVE+CEPH cluster again.

On September 16, 2016 4:36:41 AM CDT, Marco Gaiarin <g...@sv.lnf.it> wrote:
Mandi! Fabian Grünbichler
  In chel di` si favelave...

two ceph nodes, two mons and two osds are all way too few for a
(production) ceph setup.
I know, this is my 'test' ceph cluster as stated... ;-)

at least three nodes/mons (for quorum reasons),
and multiple osds per storage node (for performance and failure
are required.
Production, as planned, will have 3 nodes/mon, and 2 OSD per node.

I'm simply curious if starting a ceph cluster from cold iron could be a
common failure condition, or is a consequence of my little setup...

dott. Marco Gaiarin                                     GNUPG Key ID: 240A3D66
Associazione ``La Nostra Famiglia''
Polo FVG   -   Via della Bontà, 7 - 33078   -   San Vito al Tagliamento
marco.gaiarin(at)lanostrafamiglia.it   t +39-0434-842711   f

                Dona il 5 PER MILLE a LA NOSTRA FAMIGLIA!
        (cf 00307430132, categoria ONLUS oppure RICERCA SANITARIA)
pve-user mailing list

Zuzendari Teknikoa / Director Técnico
Binovo IT Human Project, S.L.
Telf. 943493611
Astigarraga bidea 2, planta 6 dcha., ofi. 3-2; 20180 Oiartzun (Gipuzkoa)

pve-user mailing list

Reply via email to