[Deepsea-users] Antw: Re: Integration of DeepSea in a ceph-deploy based installation

[Martin Weiss]

> On Thursday, January 19, 2017 07:40:28 AM Martin Weiss wrote:
>> Hi *,
>> 
>> in case I want to add DeepSea to an existing SES deployment ‑ how can I do
>> that without "destroying" or "changing" or "re‑deploying" anything in the
>> existing cluster? (in case that is already possible)
>> 
>> Thanks,
>> Martin
> Hi Martin,
>   I have not gone down this path yet.  Technically, you can run Stage 1 
> without causing any harm to your existing cluster.  It's running a discovery 
> 
> and only writing files to the master.  The next step would be the issue.
>   How comfortable do you feel in creating a policy.cfg that matches the 
> existing environment?  

What about enhancing deepsea with creating the policy.cfg automatically?
In case I want to integrated deepsea into an existing environment - I believe this is a must have..

> This also implies verifying/creating/selecting the 
> hardware profile for the various storage nodes.  For a simple cluster of 
> stand‑
> alone OSDs and a ceph cluster of only monitors and mds, you could quite 
> literally be done in minutes.

Not enough knowledge and experience so far - and I am not sure what might happen in case there is a mistake in the policy.cfg..

>   For a cluster of separate journals and other additional services such as 
> iSCSI or RGW, you would need to roll up your sleeves.  At this point, if you 
> 
> have such a cluster, I do not think it would be worth your time.  This is 
> the 
> harder task of accurately recording all the data and journal devices as well 
> 
> as centralizing configurations back onto the Salt master.

I have heard that we want to deprecate ceph-deploy in one of the next releases and replace it with deepsea.
With that in mind I believe we have to deliver a migration path from "ceph-deploy" to "deepsea"..

Ad yes - I agree - at the moment where deepsea is not yet on paar with ceph-deploy this might not be worth the effort...
But we need to start building the proper solution before deprecating ceph-deploy ;-).

Oh - and btw. - this would also be an added value for customers migrating from NON SLES bases Ceph deployments to SES..

> 
>   What happens if you try and something goes wrong?  I don't know.  My 
> personal paranoia level would be high enough that I would skip using any of 
> the stage orchestration commands (i.e. salt‑run state.orch ceph.stage.3) and 
> run the individual steps on the individual nodes.  It would be a bit tedious 
> 
> for a migration, but much easier to recover.

Agreed - this adds to my wish of an automatically generated policy.cfg for an existing cluster.

> 
>   In the longer term, we do have an existing card in 
> https://github.com/SUSE/DeepSea/projects/2, but have not pursued it yet.  I 
> think the above would need to be automated for creating a policy.cfg with 
> accurate hardware profiles and somehow verifiable.  Also, I do not know if 
> this 
> could be a generic solution for any Ceph cluster.

So for the moment I understand this status:

1. Recommendation: Do not use DeepSea for existing ceph-deploy based clusters.
2. Inventory can be done with DeepSea
3. policy.cfg can be created manually
4. ?

-> In case I build the policy.cfg manually and correct - what would happen if I go through the next steps of DeepSea, then? Will this kill / overwrite anything already existing?

Thanks,
Martin

> 
> Eric