Bug 1895008 - [RFE] Support for the deployment of Ganesha standalone, without any access to the Ceph orchestrator
Summary: [RFE] Support for the deployment of Ganesha standalone, without any access to...
Keywords:
Status: CLOSED WONTFIX
Alias: None
Product: Red Hat Ceph Storage
Classification: Red Hat Storage
Component: Cephadm
Version: 5.0
Hardware: Unspecified
OS: Unspecified
unspecified
high
Target Milestone: ---
: 5.1
Assignee: Juan Miguel Olmo
QA Contact: Vasishta
Karen Norteman
URL:
Whiteboard:
Depends On:
Blocks: 1820257 1839169
TreeView+ depends on / blocked
 
Reported: 2020-11-05 14:55 UTC by Giulio Fidente
Modified: 2021-05-31 10:04 UTC (History)
8 users (show)

Fixed In Version:
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
Environment:
Last Closed: 2021-05-31 10:04:26 UTC
Embargoed:


Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Ceph Project Bug Tracker 48339 0 None None None 2020-11-24 10:32:22 UTC

Description Giulio Fidente 2020-11-05 14:55:58 UTC
With ceph-ansible it was possible to deploy Ganesha as a standalone service, that is, without deploying the Ceph cluster as well.

This was useful, for example, in TripleO when it didn't manage the Ceph cluster deployment but would still deploy Ganesha (via ceph-ansible) for Manila to serve CephFS via NFS.

Currently cephadm *can* deploy Ganesha but *only* via the orchestrator, not as a standalone service.

Comment 1 RHEL Program Management 2020-11-05 14:56:06 UTC
Please specify the severity of this bug. Severity is defined here:
https://bugzilla.redhat.com/page.cgi?id=fields.html#bug_severity.

Comment 7 Sebastian Wagner 2021-05-04 11:37:35 UTC
We have an architectural problem here: Cephadm is in control of the MON deployment. A requested feature for cephadm is to failover daemons to different hosts, in case a host fails. In this case the MON quorum changes and cephadm deploys a new /etc/ceph/ceph.conf to all hosts in the cluster in order to avoid connection timeouts to MONs. 

By deploying ganesha independent of the cluster orchestration, it will be affected by connection timeouts. Thus I don't recommend this path. 

In fact, upstring is going into the opposite direction by also managing the ceph.conf on kernel-client hosts. 

Is there a way for us to avoid this architecture?

Comment 11 Sebastian Wagner 2021-05-05 08:30:27 UTC
Hi Francesco, hi John and hi Tom, 

Haha, I'm sensing some desire to keep the existing Manila architecture :-) 

Regarding VIPs for MONs, I don't think they'd give any benefit: (1) MONs need to have unique and a single dedicated IP addresses, which would by itself mandate a VIP per MON. (2) when migrating MONs to a different host, we're creating a new MON with it's own name. 


Regarding standalone ganesha, very well then. But it won't be pretty.


Note You need to log in before you can comment on or make changes to this bug.