On 27/11/12 10:53, Alon Bar-Lev wrote:
----- Original Message -----
From: "Livnat Peer" lpeer@redhat.com To: "Adam Litke" agl@us.ibm.com Cc: "Alon Bar-Lev" abarlev@redhat.com, "VDSM Project Development" vdsm-devel@lists.fedorahosted.org Sent: Tuesday, November 27, 2012 10:42:00 AM Subject: Re: [vdsm] Future of Vdsm network configuration - Thread mid-summary
On 26/11/12 16:59, Adam Litke wrote:
On Mon, Nov 26, 2012 at 02:57:19PM +0200, Livnat Peer wrote:
On 26/11/12 03:15, Shu Ming wrote:
Livnat,
Thanks for your summary. I got comments below.
2012-11-25 18:53, Livnat Peer:
Hi All, We have been discussing $subject for a while and I'd like to summarized what we agreed and disagreed on thus far.
The way I see it there are two related discussions:
- Getting VDSM networking stack to be distribution agnostic.
- We are all in agreement that VDSM API should be generic enough
to incorporate multiple implementation. (discussed on this thread: Alon's suggestion, Mark's patch for adding support for netcf etc.)
- We would like to maintain at least one implementation as the
working/up-to-date implementation for our users, this implementation should be distribution agnostic (as we all acknowledge this is an important goal for VDSM). I also think that with the agreement of this community we can choose to change our focus, from time to time, from one implementation to another as we see fit (today it can be OVS+netcf and in a few months we'll use the quantum based implementation if we agree it is better)
- The second discussion is about persisting the network
configuration on the host vs. dynamically retrieving it from a centralized location like the engine. Danken raised a concern that even if going with the dynamic approach the host should persist the management network configuration.
About dynamical retrieving from a centralized location, when will the retrieving start? Just in the very early stage of host booting before network functions? Or after the host startup and in the normal running state of the host? Before retrieving the configuration, how does the host network connecting to the engine? I think we need a basic well known network between hosts and the engine first. Then after the retrieving, hosts should reconfigure the network for later management. However, the timing to retrieve and reconfigure are challenging.
We did not discuss the dynamic approach in details on the list so far and I think this is a good opportunity to start this discussion...
From what was discussed previously I can say that the need for a well known network was raised by danken, it was referred to as the management network, this network would be used for pulling the full host network configuration from the centralized location, at this point the engine.
About the timing for retrieving the configuration, there are several approaches. One of them was described by Alon, and I think he'll join this discussion and maybe put it in his own words, but the idea was to 'keep' the network synchronized at all times. When the host have communication channel to the engine and the engine detects there is a mismatch in the host configuration, the engine initiates 'apply network configuration' action on the host.
Using this approach we'll have a single path of code to maintain and that would reduce code complexity and bugs - That's quoting Alon Bar Lev (Alon I hope I did not twisted your words/idea).
On the other hand the above approach makes local tweaks on the host (done manually by the administrator) much harder.
I worry a lot about the above if we take the dynamic approach. It seems we'd need to introduce before/after 'apply network configuration' hooks where the admin could add custom config commands that aren't yet modeled by engine.
yes, and I'm not sure the administrators would like the fact that we are 'forcing' them to write everything in a script and getting familiar with VDSM hooking mechanism (which in some cases require the use of custom properties on the engine level) instead of running a simple command line.
In which case will we force? Please be more specific. If we can pass most of the iproute2, brctl, bond parameters via key/value pairs via the API, what in your view that is common or even seldom should be used? This hook mechanism is only as fallback, provided to calm people down.
I understand, I'm saying it can irritate the administrators that needs to use it, it does not help that we are calmed down ;)
Just to make it clear I'm not against the stateless approach, I'm trying to understand it better and make sure we are all aware of the drawbacks this approach has. Complicating local tweaks to the admin is one of them.
I'll reply on your original mail with the questions I have on your proposal.
Any other approaches ?
Static configuration has the advantage of allowing a host to bring itself back online independent of the engine. This is also useful for anyone who may want to deploy a vdsm node in standalone mode.
I think it would be possible to easily support a quasi-static configuration mode simply by extending the design of the dynamic approach slightly. In dynamic mode, the network configuration is passed down as a well-defined data structure. When a particular configuration has been committed, vdsm could write a copy of that configuration data structure to /var/run/vdsm/network-config.json. During a subsequent boot, if the engine cannot be contacted after activating the management network, the cached configuration can be applied using the same code as for dynamic mode. We'd have to flesh out the circumstances under which this would happen.
I like this approach a lot but we need to consider that network configuration is an accumulated state, for example -
- The engine sends a setup-network command with the full host
network configuration 2. The user configures new network on the host, the engine sends a new setup-network request to VDSM which includes only the delta requested by the user (adding the required network) 3. VDSM adds the new network
THIS IS COMPLEX!!!!!!! Almost AI. As you need to complete the network setting with what you know.
I think we should clear this first - We have a running hypervisor with running VMs on it, we would like to configure an additional network on the host. You don't want to apply the network-configuration from scratch and mess with the running VMs networks or the storage network or anything else that is running and was not change by the administrator ==> you need to calculate the delta of the changes to perform as less intrusive operation as possible.
and this can go on and on, for dealing with this issue:
We can either hold network-config.json per setup-network command and then for recovering the network configuration state we need to execute chain of set-up networks commands.
Or we can move the logic of calculating the delta from engine to VDSM and on each setup network have the engine pass the full configuration. The problem with that approach is that the analysis logic of the delta has to be done on the engine anyway to give a quick feedback to the user on the validity of his action. Maintaining this logic/code twice is not something we want (it's bad enough to do it once....)
I don't understand how the two algorithm are the same... UI is much more/less verbose at different aspects, while taking the full configuration and convert to actual setting is a completely different sequence. What the feedback of the user? as far as I understand the user is only interested in the end-result... building his own network and expect it to be applied.
A third option is to extend the current API of setup network to include the full configuration in addition to the delta that is sent today. The full configuration would be used for creating network-config.json and for that alone, VDSM would change network configuration according to the delta sent as it does today.
Always pass full configuration, why deal with two cases?
The problem with that approach is that I'm sure someone on the list would say it is a contamination to the API, and we should 'never' pass 'duplicate' information. Personally I find this option the easiest one to deal with the above issue.
Livnat, I don't see any argument of persistence vs non persistence as the above is common to any approach taken.
Only this "manual configuration" argument keeps poping, which as I wrote is irrelevant in large scale and we do want to go into large scale.
Alon
Livnat
I'd like to add a more general question to the discussion what are the advantages of taking the dynamic approach? So far I collected two reasons:
-It is a 'cleaner' design, removes complexity on VDSM code, easier to maintain going forward, and less bug prone (I agree with that one, as long as we keep the retrieving configuration mechanism/algorithm simple).
-It adheres to the idea of having a stateless hypervisor - some more input on this point would be appreciated
Any other advantages?
discussing the benefits of having the persisted
As I mentioned above, the main benefit I see of having some sort of persistent configuration is:
- To allow the host to operate independently of the engine in
either a failure scenario or in a standalone configuration.
vdsm-devel mailing list vdsm-devel@lists.fedorahosted.org https://lists.fedorahosted.org/mailman/listinfo/vdsm-devel