Content here is by Michael Still mikal@stillhq.com. All opinions are my own.
See recent comments. RSS feed of all comments.


Mon, 13 Oct 2014



One week of Nova Kilo specifications

posted at: 03:27 | path: /openstack/kilo | permanent link to this entry


Sun, 12 Oct 2014



Compute Kilo specs are open

    From my email last week on the topic:
    I am pleased to announce that the specs process for nova in kilo is
    now open. There are some tweaks to the previous process, so please
    read this entire email before uploading your spec!
    
    Blueprints approved in Juno
    ===========================
    
    For specs approved in Juno, there is a fast track approval process for
    Kilo. The steps to get your spec re-approved are:
    
     - Copy your spec from the specs/juno/approved directory to the
    specs/kilo/approved directory. Note that if we declared your spec to
    be a "partial" implementation in Juno, it might be in the implemented
    directory. This was rare however.
     - Update the spec to match the new template
     - Commit, with the "Previously-approved: juno" commit message tag
     - Upload using git review as normal
    
    Reviewers will still do a full review of the spec, we are not offering
    a rubber stamp of previously approved specs. However, we are requiring
    only one +2 to merge these previously approved specs, so the process
    should be a lot faster.
    
    A note for core reviewers here -- please include a short note on why
    you're doing a single +2 approval on the spec so future generations
    remember why.
    
    Trivial blueprints
    ==================
    
    We are not requiring specs for trivial blueprints in Kilo. Instead,
    create a blueprint in Launchpad
    at https://blueprints.launchpad.net/nova/+addspec and target the
    specification to Kilo. New, targeted, unapproved specs will be
    reviewed in weekly nova meetings. If it is agreed they are indeed
    trivial in the meeting, they will be approved.
    
    Other proposals
    ===============
    
    For other proposals, the process is the same as Juno... Propose a spec
    review against the specs/kilo/approved directory and we'll review it
    from there.
    


    After a week I'm seeing something interesting. In Juno the specs process was new, and we saw a pause in the development cycle while people actually wrote down their designs before sending the code. This time around people know what to expect, and there are left over specs from Juno lying around. We're therefore seeing specs approved much faster than in Kilo. This should reduce the effect of the "pipeline flush" that we saw in Juno.

    So far we have five approved specs after only a week.

    Tags for this post: openstack kilo blueprints spec
    Related posts: One week of Nova Kilo specifications; On layers; Blueprints to land in Nova during Juno; My candidacy for Kilo Compute PTL; OpenStack at linux.conf.au 2013; Wow, qemu-img is fast

posted at: 16:39 | path: /openstack/kilo | permanent link to this entry


Wed, 08 Oct 2014



Lock In

posted at: 02:43 | path: /book/John_Scalzi | permanent link to this entry


Tue, 30 Sep 2014



On layers

    There's been a lot of talk recently about what we should include in OpenStack and what is out of scope. This is interesting, in that many of us used to believe that we should do ''everything''. I think what's changed is that we're learning that solving all the problems in the world is hard, and that we need to re-focus on our core products. In this post I want to talk through the various "layers" proposals that have been made in the last month or so. Layers don't directly address what we should include in OpenStack or not, but they are a useful mechanism for trying to break up OpenStack into simpler to examine chunks, and I think that makes them useful in their own right.

    I would address what I believe the scope of the OpenStack project should be, but I feel that it makes this post so long that no one will ever actually read it. Instead, I'll cover that in a later post in this series. For now, let's explore what people are proposing as a layering model for OpenStack.

    What are layers?

    Dean Troyer did a good job of describing a layers model for the OpenStack project on his blog quite a while ago. He proposed the following layers (this is a summary, you should really read his post):

    • layer 0: operating system and Oslo
    • layer 1: basic services -- Keystone, Glance, Nova
    • layer 2: extended basics -- Neutron, Cinder, Swift, Ironic
    • layer 3: optional services -- Horizon and Ceilometer
    • layer 4: turtles all the way up -- Heat, Trove, Moniker / Designate, Marconi / Zaqar


    Dean notes that Neutron would move to layer 1 when nova-network goes away and Neutron becomes required for all compute deployments. Dean's post was also over a year ago, so it misses services like Barbican that have appeared since then. Services are only allowed to require services from lower numbered layers, but can use services from higher number layers as optional add ins. So Nova for example can use Neutron, but cannot require it until it moves into layer 1. Similarly, there have been proposals to add Ceilometer as a dependency to schedule instances in Nova, and if we were to do that then we would need to move Ceilometer down to layer 1 as well. (I think doing that would be a mistake by the way, and have argued against it during at least two summits).

    Sean Dague re-ignited this discussion with his own blog post relatively recently. Sean proposes new names for most of the layers, but the intent remains the same -- a compute-centric view of the services that are required to build a working OpenStack deployment. Sean and Dean's layer definitions are otherwise strongly aligned, and Sean notes that the probability of seeing something deployed at a given installation reduces as the layer count increases -- so for example Trove is way less commonly deployed than Nova, because the set of people who want a managed database as a service is smaller than the set of of people who just want to be able to boot instances.

    Now, I'm not sure I agree with the compute centric nature of the two layers proposals mentioned so far. I see people installing just Swift to solve a storage problem, and I think that's a completely valid use of OpenStack and should be supported as a first class citizen. On the other hand, resolving my concern with the layers model there is trivial -- we just move Swift to layer 1.

    What do layers give us?

    Sean makes a good point about the complexity of OpenStack installs and how we scare away new users. I agree completely -- we show people our architecture diagrams which are deliberately confusing, and then we wonder why they're not impressed. I think we do it because we're proud of the scope of the thing we've built, but I think our audiences walk away thinking that we don't really know what problem we're trying to solve. Do I really need to deploy Horizon to have working compute? No of course not, but our architecture diagrams don't make that obvious. I gave a talk along these lines at pyconau, and I think as a community we need to be better at explaining to people what we're trying to do, while remembering that not everyone is as excited about writing a whole heap of cloud infrastructure code as we are. This is also why the OpenStack miniconf at linux.conf.au 2015 has pivoted from being a generic OpenStack chatfest to being something more solidly focussed on issues of interest to deployers -- we're just not great at talking to our users and we need to reboot the conversation at community conferences until its something which meets their needs.


    We intend this diagram to amaze and confuse our victims


    Agreeing on a set of layers gives us a framework within which to describe OpenStack to our users. It lets us communicate the services we think are basic and always required, versus those which are icing on the cake. It also let's us explain the dependency between projects better, and that helps deployers work out what order to deploy things in.

    Do layers help us work out what OpenStack should focus on?

    Sean's blog post then pivots and starts talking about the size of the OpenStack ecosystem -- or the "size of our tent" as he phrases it. While I agree that we need to shrink the number of projects we're working on at the moment, I feel that the blog post is missing a logical link between the previous layers discussion and the tent size conundrum. It feels to me that Sean wanted to propose that OpenStack focus on a specific set of layers, but didn't quite get there for whatever reason.

    Next Monty Taylor had a go at furthering this conversation with his own blog post on the topic. Monty starts by making a very important point -- he (like all involved) both want the OpenStack community to be as inclusive as possible. I want lots of interesting people at the design summits, even if they don't work directly on projects that OpenStack ships. You can be a part of the OpenStack community without having our logo on your product.

    A concrete example of including non-OpenStack projects in our wider community was visible at the Atlanta summit -- I know for a fact that there were software engineers at the summit who work on Google Compute Engine. I know this because I used to work with them at Google when I was a SRE there. I have no problem with people working on competing products being at our summits, as long as they are there to contribute meaningfully in the sessions, and not just take from us. It needs to be a two way street. Another concrete example is Ceph. I think Ceph is cool, and I'm completely fine with people using it as part of their OpenStack deploy. What upsets me is when people conflate Ceph with OpenStack. They are different. They're separate. And that is fine. Let's just not confuse people by saying Ceph is part of the OpenStack project -- it simply isn't because it doesn't fall under our governance model. Ceph is still a valued member of our community and more than welcome at our summits.

    Do layers help us work our what to focus OpenStack on for now? I think they do. Should we simply say that we're only going to work on a single layer? Absolutely not. What we've tried to do up until now is have OpenStack be a single big thing, what we call "the integrated release". I think layers gives us a tool to find logical ways to break that thing up. Perhaps we need a smaller integrated release, but then continue with the other projects but on their own release cycles? Or perhaps they release at the same time, but we don't block the release of a layer 1 service on the basis of release critical bugs in a layer 4 service?

    Is there consensus on what sits in each layer?

    Looking at the posts I can find on this topic so far, I'd have to say the answer is no. We're close, but we're not aligned yet. For example, one proposal has a tweak to the previously proposed layer model that adds Cinder, Designate and Neutron down into layer 1 (basic services). The author argues that this is because stateless cloud isn't particularly useful to users of OpenStack. However, I think this is wrong to be honest. I can see that stateless cloud isn't super useful by itself, but we are assuming that OpenStack is the only piece of infrastructure that a given organization has. Perhaps that's true for the public cloud case, but the vast majority of OpenStack deployments at this point are private clouds. So, you're an existing IT organization and you're deploying OpenStack to increase the level of flexibility in compute resources. You don't need to deploy Cinder or Designate to do that. Let's take the storage case for a second -- our hypothetical IT organization probably already has some form of storage -- a SAN, or NFS appliances, or something like that. So stateful cloud is easy for them -- they just have their instances mount resources from those existing storage pools like they would any other machine. Eventually they'll decide that hand managing that is horrible and move to Cinder, but that's probably later once they've gotten through the initial baby step of deploying Nova, Glance and Keystone.

    The first step to using layers to decide what we should focus on is to decide what is in each layer. I think the conversation needs to revolve around that for now, because it we drift off into whether existing in a given layer means you're voted off the OpenStack island, when we'll never even come up with a set of agreed layers.

    Let's ignore tents for now

    The size of the OpenStack "tent" is the metaphor being used at the moment for working out what to include in OpenStack. As I say above, I think we need to reach agreement on what is in each layer before we can move on to that very important conversation.

    Conclusion

    Given the focus of this post is the layers model, I want to stop introducing new concepts here for now. Instead let me summarize where I stand so far -- I think the layers model is useful. I also think the layers should be an inverted pyramid -- layer 1 should be as small as possible for example. This is because of the dependency model that the layers model proposes -- it is important to keep the list of things that a layer 2 service must use as small and coherent as possible. Another reason to keep the lower layers as small as possible is because each layer represents the smallest possible increment of an OpenStack deployment that we think is reasonable. We believe it is currently reasonable to deploy Nova without Cinder or Neutron for example.

    Most importantly of all, having those incremental stages of OpenStack deployment gives us a framework we have been missing in talking to our deployers and users. It makes OpenStack less confusing to outsiders, as it gives them bite sized morsels to consume one at a time.

    So here are the layers as I see them for now:

    • layer 0: operating system, and Oslo
    • layer 1: basic services -- Keystone, Glance, Nova, and Swift
    • layer 2: extended basics -- Neutron, Cinder, and Ironic
    • layer 3: optional services -- Horizon, and Ceilometer
    • layer 4: application services -- Heat, Trove, Designate, and Zaqar


    I am not saying that everything inside a single layer is required to be deployed simultaneously, but I do think its reasonable for Ceilometer to assume that Swift is installed and functioning. The big difference here between my view of layers and that of Dean, Sean and Monty is that I think that Swift is a layer 1 service -- it provides basic functionality that may be assumed to exist by services above it in the model.

    I believe that when projects come to the Technical Committee requesting incubation or integration, they should specify what layer they see their project sitting at, and the justification for a lower layer number should be harder than that for a higher layer. So for example, we should be reasonably willing to accept proposals at layer 4, whilst we should be super concerned about the implications of adding another project at layer 1.

    In the next post in this series I'll try to address the size of the OpenStack "tent", and what projects we should be focussing on.

    Tags for this post: openstack kilo technical committee tc layers
    Related posts: One week of Nova Kilo specifications; Juno TC Candidacy; Compute Kilo specs are open; My candidacy for Kilo Compute PTL; OpenStack at linux.conf.au 2013; Wow, qemu-img is fast

posted at: 18:57 | path: /openstack/kilo | permanent link to this entry


Blueprints implemented in Nova during Juno

posted at: 13:56 | path: /openstack/juno | permanent link to this entry


Mon, 29 Sep 2014



Chronological list of Juno Nova mid-cycle meetup posts

posted at: 23:10 | path: /openstack/juno | permanent link to this entry


My candidacy for Kilo Compute PTL

    This is mostly historical at this point, but I forgot to post it here when I emailed it a week or so ago. So, for future reference:

    I'd like another term as Compute PTL, if you'll have me.
    
    We live in interesting times. openstack has clearly gained a large
    amount of mind share in the open cloud marketplace, with Nova being a
    very commonly deployed component. Yet, we don't have a fantastic
    container solution, which is our biggest feature gap at this point.
    Worse -- we have a code base with a huge number of bugs filed against
    it, an unreliable gate because of subtle bugs in our code and
    interactions with other openstack code, and have a continued need to
    add features to stay relevant. These are hard problems to solve.
    
    Interestingly, I think the solution to these problems calls for a
    social approach, much like I argued for in my Juno PTL candidacy
    email. The problems we face aren't purely technical -- we need to work
    out how to pay down our technical debt without blocking all new
    features. We also need to ask for understanding and patience from
    those feature authors as we try and improve the foundation they are
    building on.
    
    The specifications process we used in Juno helped with these problems,
    but one of the things we've learned from the experiment is that we
    don't require specifications for all changes. Let's take an approach
    where trivial changes (no API changes, only one review to implement)
    don't require a specification. There will of course sometimes be
    variations on that rule if we discover something, but it means that
    many micro-features will be unblocked.
    
    In terms of technical debt, I don't personally believe that pulling
    all hypervisor drivers out of Nova fixes the problems we face, it just
    moves the technical debt to a different repository. However, we
    clearly need to discuss the way forward at the summit, and come up
    with some sort of plan. If we do something like this, then I am not
    sure that the hypervisor driver interface is the right place to do
    that work -- I'd rather see something closer to the hypervisor itself
    so that the Nova business logic stays with Nova.
    
    Kilo is also the release where we need to get the v2.1 API work done
    now that we finally have a shared vision for how to progress. It took
    us a long time to get to a good shared vision there, so we need to
    ensure that we see that work through to the end.
    
    We live in interesting times, but they're also exciting as well.
    


    I have since been elected unopposed, so thanks for that!

    Tags for this post: openstack kilo compute ptl
    Related posts: One week of Nova Kilo specifications; Thoughts from the PTL; On layers; Expectations of core reviewers; Juno Nova PTL Candidacy; Review priorities as we approach juno-3

posted at: 18:34 | path: /openstack/kilo | permanent link to this entry


Fri, 26 Sep 2014



The Decline and Fall of IBM: End of an American Icon?

posted at: 00:39 | path: /book/Robert_Cringely | permanent link to this entry


Thu, 21 Aug 2014



Juno nova mid-cycle meetup summary: conclusion

posted at: 23:47 | path: /openstack/juno | permanent link to this entry


Juno nova mid-cycle meetup summary: the next generation Nova API

    This is the final post in my series covering the highlights from the Juno Nova mid-cycle meetup. In this post I will cover our next generation API, which used to be called the v3 API but is largely now referred to as the v2.1 API. Getting to this point has been one of the more painful processes I think I've ever seen in Nova's development history, and I think we've learnt some important things about how large distributed projects operate along the way. My hope is that we remember these lessons next time we hit something as contentious as our API re-write has been.

    Now on to the API itself. It started out as an attempt to improve our current API to be more maintainable and less confusing to our users. We deliberately decided that we would not focus on adding features, but instead attempt to reduce as much technical debt as possible. This development effort went on for about a year before we realized we'd made a mistake. The mistake we made is that we assumed that our users would agree it was trivial to move to a new API, and that they'd do that even if there weren't compelling new features, which it turned out was entirely incorrect.

    I want to make it clear that this wasn't a mistake on the part of the v3 API team. They implemented what the technical leadership of Nova at the time asked for, and were very surprised when we discovered our mistake. We've now spent over a release cycle trying to recover from that mistake as gracefully as possible, but the upside is that the API we will be delivering is significantly more future proof than what we have in the current v2 API.

    At the Atlanta Juno summit, it was agreed that the v3 API would never ship in its current form, and that what we would instead do is provide a v2.1 API. This API would be 99% compatible with the current v2 API, with the incompatible things being stuff like if you pass a malformed parameter to the API we will now tell you instead of silently ignoring it, which we call 'input validation'. The other thing we are going to add in the v2.1 API is a system of 'micro-versions', which allow a client to specify what version of the API it understands, and for the server to gracefully degrade to older versions if required.

    This micro-version system is important, because the next step is to then start adding the v3 cleanups and fixes into the v2.1 API, but as a series of micro-versions. That way we can drag the majority of our users with us into a better future, without abandoning users of older API versions. I should note at this point that the mechanics for deciding what the minimum micro-version a version of Nova will support are largely undefined at the moment. My instinct is that we will tie to stable release versions in some way; if your client dates back to a release of Nova that we no longer support, then we might expect you to upgrade. However, that hasn't been debated yet, so don't take my thoughts on that as rigid truth.

    Frustratingly, the intent of the v2.1 API has been agreed and unchanged since the Atlanta summit, yet we're late in the Juno release and most of the work isn't done yet. This is because we got bogged down in the mechanics of how micro-versions will work, and how the translation for older API versions will work inside the Nova code later on. We finally unblocked this at the mid-cycle meetup, which means this work can finally progress again.

    The main concern that we needed to resolve at the mid-cycle was the belief that if the v2.1 API was implemented as a series of translations on top of the v3 code, then the translation layer would be quite thick and complicated. This raises issues of maintainability, as well as the amount of code we need to understand. The API team has now agreed to produce an API implementation that is just the v2.1 functionality, and will then layer things on top of that. This is actually invisible to users of the API, but it leaves us with an implementation where changes after v2.1 are additive, which should be easier to maintain.

    One of the other changes in the original v3 code is that we stopped proxying functionality for Neutron, Cinder and Glance. With the decision to implement a v2.1 API instead, we will need to rebuild that proxying implementation. To unblock v2.1, and based on advice from the HP and Rackspace public cloud teams, we have decided to delay implementing these proxies. So, the first version of the v2.1 API we ship will not have proxies, but later versions will add them in. The current v2 API implementation will not be removed until all the proxies have been added to v2.1. This is prompted by the belief that many advanced API users don't use the Nova API proxies, and therefore could move to v2.1 without them being implemented.

    Finally, I want to thank the Nova API team, especially Chris Yeoh and Kenichi Oomichi for their patience with us while we have worked through these complicated issues. It's much appreciated, and I find them a consistent pleasure to work with.

    That brings us to the end of my summary of the Nova Juno mid-cycle meetup. I'll write up a quick summary post that ties all of the posts together, but apart from that this series is now finished. Thanks for following along.

    Tags for this post: openstack juno nova mid-cycle summary api v3 v2.1
    Related posts: Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Chronological list of Juno Nova mid-cycle meetup posts; Juno nova mid-cycle meetup summary: conclusion; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: containers

posted at: 16:52 | path: /openstack/juno | permanent link to this entry


Don't Tell Mum I Work On The Rigs

posted at: 13:45 | path: /book/Paul_Carter | permanent link to this entry


Tue, 19 Aug 2014



Juno nova mid-cycle meetup summary: nova-network to Neutron migration

    This will be my second last post about the Juno Nova mid-cycle meetup, which covers the state of play for work on the nova-network to Neutron upgrade.

    First off, some background information. Neutron (formerly Quantum) was developed over a long period of time to replace nova-network, and added to the OpenStack Folsom release. The development of new features for nova-network was frozen in the Nova code base, so that users would transition to Neutron. Unfortunately the transition period took longer than expected. We ended up having to unfreeze development of nova-network, in order to fix reliability problems that were affecting our CI gating and the reliability of deployments for existing nova-network users. Also, at least two OpenStack companies were carrying significant feature patches for nova-network, which we wanted to merge into the main code base.

    You can see the announcement at http://lists.openstack.org/pipermail/openstack-dev/2014-January/025824.html. The main enhancements post-freeze were a conversion to use our new objects infrastructure (and therefore conductor), as well as features that were being developed by Nebula. I can't find any contributions from the other OpenStack company in the code base at this time, so I assume they haven't been proposed.

    The nova-network to Neutron migration path has come to the attention of the OpenStack Technical Committee, who have asked for a more formal plan to address Neutron feature gaps and deprecate nova-network. That plan is tracked at https://wiki.openstack.org/wiki/Governance/TechnicalCommittee/Neutron_Gap_Coverage. As you can see, there are still some things to be merged which are targeted for juno-3. At the time of writing this includes grenade testing; Neutron being the devstack default; a replacement for nova-network multi-host; a migration plan; and some documentation. They are all making good progress, but until these action items are completed, Nova can't start the process of deprecating nova-network.

    The discussion at the Nova mid-cycle meetup was around the migration planning item in the plan. There is a Nova specification that outlines one possible plan for live upgrading instances (i.e, no instance downtime) at https://review.openstack.org/#/c/101921/, but this will probably now be replaced with a simpler migration path involving cold migrations. This is prompted by not being able to find a user that absolutely has to have live upgrade. There was some confusion, because of a belief that the TC was requiring a live upgrade plan. But as Russell Bryant says in the meetup etherpad:

    "Note that the TC has made no such statement on migration expectations other than a migration path must exist, both projects must agree on the plan, and that plan must be submitted to the TC as a part of the project's graduation review (or project gap review in this case). I wouldn't expect the TC to make much of a fuss about the plan if both Nova and Neutron teams are in agreement."


    The current plan is to go forward with a cold upgrade path, unless a user comes forward with an absolute hard requirement for a live upgrade, and a plan to fund developers to work on it.

    At this point, it looks like we are on track to get all of the functionality we need from Neutron in the Juno release. If that happens, we will start the nova-network deprecation timer in Kilo, with my expectation being that nova-network would be removed in the "M" release. There is also an option to change the default networking implementation to Neutron before the deprecation of nova-network is complete, which will mean that new deployments are defaulting to the long term supported option.

    In the next (and probably final) post in this series, I'll talk about the API formerly known as Nova API v3.

    Tags for this post: openstack juno nova mid-cycle summary nova-network neutron migration
    Related posts: Chronological list of Juno Nova mid-cycle meetup posts; Juno nova mid-cycle meetup summary: conclusion; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: containers; Juno nova mid-cycle meetup summary: ironic

posted at: 20:37 | path: /openstack/juno | permanent link to this entry


Juno nova mid-cycle meetup summary: slots

    If I had to guess what would be a controversial topic from the mid-cycle meetup, it would have to be this slots proposal. I was actually in a Technical Committee meeting when this proposal was first made, but I'm told there were plenty of people in the room keen to give this idea a try. Since the mid-cycle Joe Gordon has written up a more formal proposal, which can be found at https://review.openstack.org/#/c/112733.

    If you look at the last few Nova releases, core reviewers have been drowning under code reviews, so we need to control the review workload. What is currently happening is that everyone throws up their thing into Gerrit, and then each core tries to identify the important things and review them. There is a list of prioritized blueprints in Launchpad, but it is not used much as a way of determining what to review. The result of this is that there are hundreds of reviews outstanding for Nova (500 when I wrote this post). Many of these will get a review, but it is hard for authors to get two cores to pay attention to a review long enough for it to be approved and merged.

    If we could rate limit the number of proposed reviews in Gerrit, then cores would be able to focus their attention on the smaller number of outstanding reviews, and land more code. Because each review would merge faster, we believe this rate limiting would help us land more code rather than less, as our workload would be better managed. You could argue that this will mean we just say 'no' more often, but that's not the intent, it's more about bringing focus to what we're reviewing, so that we can get patches through the process completely. There's nothing more frustrating to a code author than having one +2 on their code and then hitting some merge freeze deadline.

    The proposal is therefore to designate a number of blueprints that can be under review at any one time. The initial proposal was for ten, and the term 'slot' was coined to describe the available review capacity. If your blueprint was not allocated a slot, then it would either not be proposed in Gerrit yet, or if it was it would have a procedural -2 on it (much like code reviews associated with unapproved specifications do now).

    The number of slots is arbitrary at this point. Ten is our best guess of how much we can dilute core's focus without losing efficiency. We would tweak the number as we gained experience if we went ahead with this proposal. Remember, too, that a slot isn't always a single code review. If the VMWare refactor was in a slot for example, we might find that there were also ten code reviews associated with that single slot.

    How do you determine what occupies a review slot? The proposal is to groom the list of approved specifications more carefully. We would collaboratively produce a ranked list of blueprints in the order of their importance to Nova and OpenStack overall. As slots become available, the next highest ranked blueprint with code ready for review would be moved into one of the review slots. A blueprint would be considered 'ready for review' once the specification is merged, and the code is complete and ready for intensive code review.

    What happens if code is in a slot and something goes wrong? Imagine if a proposer goes on vacation and stops responding to review comments. If that happened we would bump the code out of the slot, but would put it back on the backlog in the location dictated by its priority. In other words there is no penalty for being bumped, you just need to wait for a slot to reappear when you're available again.

    We also talked about whether we were requiring specifications for changes which are too simple. If something is relatively uncontroversial and simple (a better tag for internationalization for example), but not a bug, it falls through the cracks of our process at the moment and ends up needing to have a specification written. There was talk of finding another way to track this work. I'm not sure I agree with this part, because a trivial specification is a relatively cheap thing to do. However, it's something I'm happy to talk about.

    We also know that Nova needs to spend more time paying down its accrued technical debt, which you can see in the huge amount of bugs we have outstanding at the moment. There is no shortage of people willing to write code for Nova, but there is a shortage of people fixing bugs and working on strategic things instead of new features. If we could reserve slots for technical debt, then it would help us to get people to work on those aspects, because they wouldn't spend time on a less interesting problem and then discover they can't even get their code reviewed. We even talked about having an alternating focus for Nova releases; we could have a release focused on paying down technical debt and stability, and then the next release focused on new features. The Linux kernel does something quite similar to this and it seems to work well for them.

    Using slots would allow us to land more valuable code faster. Of course, it also means that some patches will get dropped on the floor, but if the system is working properly, those features will be ones that aren't important to OpenStack. Considering that right now we're not landing many features at all, this would be an improvement.

    This proposal is obviously complicated, and everyone will have an opinion. We haven't really thought through all the mechanics fully, yet, and it's certainly not a done deal at this point. The ranking process seems to be the most contentious point. We could encourage the community to help us rank things by priority, but it's not clear how that process would work. Regardless, I feel like we need to be more systematic about what code we're trying to land. It's embarrassing how little has landed in Juno for Nova, and we need to be working on that. I would like to continue discussing this as a community to make sure that we end up with something that works well and that everyone is happy with.

    This series is nearly done, but in the next post I'll cover the current status of the nova-network to neutron upgrade path.

    Tags for this post: openstack juno nova mid-cycle summary review slots blueprint priority project management
    Related posts: Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: ironic; Chronological list of Juno Nova mid-cycle meetup posts; Juno nova mid-cycle meetup summary: conclusion

posted at: 00:34 | path: /openstack/juno | permanent link to this entry


Sun, 17 Aug 2014



Juno nova mid-cycle meetup summary: scheduler

    This post is in a series covering the discussions at the Juno Nova mid-cycle meetup. This post will cover the current state of play of our scheduler refactoring efforts. The scheduler refactor has been running for a fair while now, dating back to at least the Hong Kong summit (so about 1.5 release cycles ago).

    The original intent of the scheduler sub-team's effort was to pull the scheduling code out of Nova so that it could be rapidly iterated on its own, with the eventual goal being to support a single scheduler across the various OpenStack services. For example, the scheduler that makes placement decisions about your instances could also be making decisions about the placement of your storage resources and could therefore ensure that they are co-located as much as possible.

    During this process we realized that a big bang replacement is actually much harder than we thought, and the plan has morphed into being a multi-phase effort. The first step is to make the interface for the scheduler more clearly defined inside the Nova code base. For example, in previous releases, it was the scheduler that launched instances: the API would ask the scheduler to find available hypervisor nodes, and then the scheduler would instruct those nodes to boot the instances. We need to refactor this so that the scheduler picks a set of nodes, but then the API is the one which actually does the instance launch. That way, when the scheduler does move out it's not trusted to perform actions that change hypervisor state, and the Nova code does that for it. This refactoring work is under way, along with work to isolate the SQL database accesses inside the scheduler.

    I would like to set expectations that this work is what will land in Juno. It has little visible impact for users, but positions us to better solve these problems in Kilo.

    We discussed the need to ensure that any new scheduler is at least as fast and accurate as the current one. Jay Pipes has volunteered to work with the scheduler sub-team to build a testing framework to validate this work. Jay also has some concerns about the resource tracker work that is being done at the moment that he is going to discuss with the scheduler sub-team. Since the mid-cycle meetup there has been a thread on the openstack-dev mailing list about similar resource tracker concerns (here), which might be of interest to people interested in scheduler work.

    We also need to test our assumption at some point that other OpenStack services such as Neutron and Cinder would be even willing to share a scheduler service if a central one was implemented. We believe that Neutron is interested, but we shouldn't be surprising our fellow OpenStack projects by just appearing with a complete solution. There is a plan to propose a cross-project session at the Paris summit to cover this work.

    In the next post in this series we'll discuss possibly the most controversial part of the mid-cycle meetup. The proposal for "slots" for landing blueprints during Kilo.

    Tags for this post: openstack juno nova mid-cycle summary scheduler
    Related posts: Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: ironic; Juno nova mid-cycle meetup summary: slots; Chronological list of Juno Nova mid-cycle meetup posts; Juno nova mid-cycle meetup summary: conclusion

posted at: 20:06 | path: /openstack/juno | permanent link to this entry


Juno nova mid-cycle meetup summary: bug management

    Welcome to the next exciting installment of the Nova Juno mid-cycle meetup summary. In the previous chapter, our hero battled a partially complete cells implementation, by using his +2 smile of good intentions. In this next exciting chapter, watch him battle our seemingly never ending pile of bugs! Sorry, now that I'm on to my sixth post in this series I feel like it's time to get more adventurous in the introductions.

    For at least the last cycle, and probably longer, Nova has been struggling with the number of bugs filed in Launchpad. I don't think the problem is that Nova has terrible code, it is instead that we have a lot of users filing bugs, and the team working on triaging and closing bugs is small. The complexity of the deployment options with Nova make this problem worse, and that complexity increases as we allow new drivers for things like different storage engines to land in the code base.

    The increasing number of permutations possible with Nova configurations is a problem for our CI systems as well, as we don't cover all of these options and this sometimes leads us to discover that they don't work as expected in the field. CI is a tangent from the main intent of this post though, so I will reserve further discussion of our CI system until a later post.

    Tracy Jones and Joe Gordon have been doing good work in this cycle trying to get a grip on the state of the bugs filed against Nova. For example, a very large number of bugs (hundreds) were for problems we'd fixed, but where the bug bot had failed to close the bug when the fix merged. Many other bugs were waiting for feedback from users, but had been waiting for longer than six months. In both those cases the response was to close the bug, with the understanding that the user can always reopen it if they come back to talk to us again. Doing "quick hit" things like this has reduced our open bug count to about one thousand bugs. You can see a dashboard that Tracy has produced that shows the state of our bugs at http://54.201.139.117/nova-bugs.html. I believe that Joe has been moving towards moving this onto OpenStack hosted infrastructure, but this hasn't happened yet.

    At the mid-cycle meetup, the goal of the conversation was to try and find other ways to get our bug queue further under control. Some of the suggestions were largely mechanical, like tightening up our definitions of the confirmed (we agree this is a bug) and triaged (and we know how to fix it) bug states. Others were things like auto-abandoning bugs which are marked incomplete for more than 60 days without a reply from the person who filed the bug, or unassigning bugs when the review that proposed a fix is abandoned in Gerrit.

    Unfortunately, we have more ideas for how to automate dealing with bugs than we have people writing automation. If there's someone out there who wants to have a big impact on Nova, but isn't sure where to get started, helping us out with this automation would be a super helpful way to get started. Let Tracy or I know if you're interested.

    We also talked about having more targeted bug days. This was prompted by our last bug day being largely unsuccessful. Instead we're proposing that the next bug day have a really well defined theme, such as moving things from the "undecided" to the "confirmed" state, or similar. I believe the current plan is to run a bug day like this after J-3 when we're winding down from feature development and starting to focus on stabilization.

    Finally, I would encourage people fixing bugs in Nova to do a quick search for duplicate bugs when they are closing a bug. I wouldn't be at all surprised to discover that there are many bugs where you can close duplicates at the same time with minimal effort.

    In the next post I'll cover our discussions of the state of the current scheduler work in Nova.

    Tags for this post: openstack juno nova mi-cycle summary bugs
    Related posts: Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: ironic; Juno nova mid-cycle meetup summary: slots; Chronological list of Juno Nova mid-cycle meetup posts

posted at: 19:38 | path: /openstack/juno | permanent link to this entry


Thu, 14 Aug 2014



Juno nova mid-cycle meetup summary: cells

    This is the next post summarizing the Juno Nova mid-cycle meetup. This post covers the cells functionality used by some deployments to scale Nova.

    For those unfamiliar with cells, it's a way of combining smaller Nova installations into a thing which feels like a single large Nova install. So for example, Rackspace deploys Nova in cells of hundreds of machines, and these cells form a Nova availability zone which might contain thousands of machines. The cells in one of these deployments form a tree: users talk to the top level of the tree, which might only contain API services. That cell then routes requests to child cells which can actually perform the operation requested.

    There are a few reasons why Rackspace does this. Firstly, it keeps the MySQL databases smaller, which can improve the performance of database operations and backups. Additionally, cells can contain different types of hardware, which are then partitioned logically. For example, OnMetal (Rackspace's Ironic-based baremetal product) instances come from a cell which contains OnMetal machines and only publishes OnMetal flavors to the parent cell.

    Cells was originally written by Rackspace to meet its deployment needs, but is now used by other sites as well. However, I think it would be a stretch to say that cells is commonly used, and it is certainly not the deployment default. In fact, most deployments don't run any of the cells code, so you can't really call them even a "single cell install". One of the reasons cells isn't more widely deployed is that it doesn't implement the entire Nova API, which means some features are missing. As a simple example, you can't live-migrate an instance between two child cells.

    At the meetup, the first thing we discussed regarding cells was a general desire to see cells finished and become the default deployment method for Nova. Perhaps most people end up running a single cell, but in that case at least the cells code paths are well used. The first step to get there is improving the Tempest coverage for cells. There was a recent openstack-dev mailing list thread on this topic, which was discussed at the meetup. There was commitment from several Nova developers to work on this, and notably not all of them are from Rackspace.

    It's important that we improve the Tempest coverage for cells, because it positions us for the next step in the process, which is bringing feature parity to cells compared with a non-cells deployment. There is some level of frustration that the work on cells hasn't really progressed in Juno, and that it is currently incomplete. At the meetup, we made a commitment to bringing a well-researched plan to the Kilo summit for implementing feature parity for a single cell deployment compared with a current default deployment. We also made a commitment to make cells the default deployment model when this work is complete. If this doesn't happen in time for Kilo, then we will be forced to seriously consider removing cells from Nova. A half-done cells deployment has so far stopped other development teams from trying to solve the problems that cells addresses, so we either need to finish cells, or get out of the way so that someone else can have a go. I am confident that the cells team will take this feedback on board and come to the summit with a good plan. Once we have a plan we can ask the whole community to rally around and help finish this effort, which I think will benefit all of us.

    In the next blog post I will cover something we've been struggling with for the last few releases: how we get our bug count down to a reasonable level.

    Tags for this post: openstack juno nova mid-cycle summary cells
    Related posts: Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: ironic; Juno nova mid-cycle meetup summary: slots; Chronological list of Juno Nova mid-cycle meetup posts

posted at: 21:20 | path: /openstack/juno | permanent link to this entry


More bowls and pens

posted at: 19:35 | path: /wood/turning/20140805-woodturning | permanent link to this entry


Juno nova mid-cycle meetup summary: DB2 support

    This post is one part of a series discussing the OpenStack Nova Juno mid-cycle meetup. It's a bit shorter than most of the others, because the next thing on my list to talk about is DB2, and that's relatively contained.

    IBM is interested in adding DB2 support as a SQL database for Nova. Theoretically, this is a relatively simple thing to do because we use SQLAlchemy to abstract away the specifics of the SQL engine. However, in reality, the abstraction is leaky. The obvious example in this case is that DB2 has different rules for foreign keys than other SQL engines we've used. So, in order to be able to make this change, we need to tighten up our schema for the database.

    The change that was discussed is the requirement that the UUID column on the instances table be not null. This seems like a relatively obvious thing to allow, given that UUID is the official way to identify an instance, and has been for a really long time. However, there are a few things which make this complicated: we need to understand the state of databases that might have been through a long chain of upgrades from previous Nova releases, and we need to ensure that the schema alterations don't cause significant performance problems for existing large deployments.

    As an aside, people sometimes complain that Nova development is too slow these days, and they're probably right, because things like this slow us down. A relatively simple change to our database schema requires a whole bunch of performance testing and negotiation with operators to ensure that its not going to be a problem for people. It's good that we do these things, but sometimes it's hard to explain to people why forward progress is slow in these situations.

    Matt Riedemann from IBM has been doing a good job of handling this change. He's written a tool that operators can run before the change lands in Juno that checks if they have instance rows with null UUIDs. Additionally, the upgrade process has been well planned, and is documented in the specification available on the fancy pants new specs website.

    We had a long discussion about this change at the meetup, and how it would impact on large deployments. Both Rackspace and HP were asked if they could run performance tests to see if the schema change would be a problem for them. Unfortunately HP's testing hardware was tied up with another project, so we only got numbers from Rackspace. For them, the schema change took 42 minutes for a large database. Almost all of that was altering the column to be non-nullable; creating the new index was only 29 seconds of runtime. However, the Rackspace database is large because they don't currently purge deleted rows, if they can get that done before running this schema upgrade then the impact will be much smaller.

    So the recommendation here for operators is that it is best practice to purge deleted rows from your databases before an upgrade, especially when schema migrations need to occur at the same time. There are some other takeaways for operators as well: if we know that operators have a large deployment, then we can ask if an upgrade will be a problem. This is why being active on the openstack-operators mailing list is important. Additionally, if operators are willing to donate a dataset to Turbo-Hipster for DB CI testing, then we can use that in our automation to try and make sure these upgrades don't cause you pain in the future.

    In the next post in this series I'll talk about the future of cells, and the work that needs to be done there to make it a first class citizen.

    Tags for this post: openstack juno nova mid-cycle summary sql database sqlalchemy db2
    Related posts: Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: ironic; Juno nova mid-cycle meetup summary: slots; Chronological list of Juno Nova mid-cycle meetup posts

posted at: 19:20 | path: /openstack/juno | permanent link to this entry


Review priorities as we approach juno-3

    I just send this email out to openstack-dev, but I am posting it here in case it makes it more discoverable to people drowning in email:

    To: openstack-dev
    Subject: [nova] Review priorities as we approach juno-3
    
    Hi.
    
    We're rapidly approaching j-3, so I want to remind people of the
    current reviews that are high priority. The definition of high
    priority I am using here is blueprints that are marked high priority
    in launchpad that have outstanding code for review -- I am sure there
    are other reviews that are important as well, but I want us to try to
    land more blueprints than we have so far. These are listed in the
    order they appear in launchpad.
    
    == Compute Manager uses Objects (Juno Work) ==
    
    https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bp/compute-manager-objects-juno,n,z
    
    This is ongoing work, but if you're after some quick code review
    points they're very easy to review and help push the project forward
    in an important manner.
    
    == Move Virt Drivers to use Objects (Juno Work) ==
    
    I couldn't actually find any code out for review for this one apart
    from https://review.openstack.org/#/c/94477/, is there more out there?
    
    == Add a virt driver for Ironic ==
    
    This one is in progress, but we need to keep going at it or we wont
    get it merged in time.
    
    * https://review.openstack.org/#/c/111223/ was approved, but a rebased
    ate it. Should be quick to re-approve.
    * https://review.openstack.org/#/c/111423/
    * https://review.openstack.org/#/c/111425/
    * ...there are more reviews in this series, but I'd be super happy to
    see even a few reviewed
    
    == Create Scheduler Python Library ==
    
    * https://review.openstack.org/#/c/82778/
    * https://review.openstack.org/#/c/104556/
    
    (There are a few abandoned patches in this series, I think those two
    are the active ones but please correct me if I am wrong).
    
    == VMware: spawn refactor ==
    
    * https://review.openstack.org/#/c/104145/
    * https://review.openstack.org/#/c/104147/ (Dan Smith's -2 on this one
    seems procedural to me)
    * https://review.openstack.org/#/c/105738/
    * ...another chain with many more patches to review
    
    Thanks,
    Michael
    


    The actual email thread is at http://lists.openstack.org/pipermail/openstack-dev/2014-August/043098.html.

    Tags for this post: openstack juno review nova ptl
    Related posts: Expectations of core reviewers; Juno nova mid-cycle meetup summary: social issues; Juno Nova PTL Candidacy; Juno nova mid-cycle meetup summary: slots; Thoughts from the PTL; Juno nova mid-cycle meetup summary: nova-network to Neutron migration

posted at: 13:01 | path: /openstack/juno | permanent link to this entry


Juno nova mid-cycle meetup summary: ironic

    Welcome to the third in my set of posts covering discussion topics at the nova juno mid-cycle meetup. The series might never end to be honest.

    This post will cover the progress of the ironic nova driver. This driver is interesting as an example of a large contribution to the nova code base for a couple of reasons -- its an official OpenStack project instead of a vendor driver, which means we should already have well aligned goals. The driver has been written entirely using our development process, so its already been reviewed to OpenStack standards, instead of being a large code dump from a separate development process. Finally, its forced us to think through what merging a non-trivial code contribution should look like, and I think that formula will be useful for later similar efforts, the Docker driver for example.

    One of the sticking points with getting the ironic driver landed is exactly how upgrade for baremetal driver users will work. The nova team has been unwilling to just remove the baremetal driver, as we know that it has been deployed by at least a few OpenStack users -- the largest deployment I am aware of is over 1,000 machines. Now, this unfortunate because the baremetal driver was always intended to be experimental. I think what we've learnt from this is that any driver which merges into the nova code base has to be supported for a reasonable period of time -- nova isn't the right place for experiments. Now that we have the stackforge driver model I don't think that's too terrible, because people can iterate quickly in stackforge, and when they have something stable and supportable they can merge it into nova. This gives us the best of both worlds, while providing a strong signal to deployers about what the nova team is willing to support for long periods of time.

    The solution we came up with for upgrades from baremetal to ironic is that the deployer will upgrade to juno, and then run a script which converts their baremetal nodes to ironic nodes. This script is "off line" in the sense that we do not expect new baremetal nodes to be launchable during this process, nor after it is completed. All further launches would be via the ironic driver.

    These nodes that are upgraded to ironic will exist in a degraded state. We are not requiring ironic to support their full set of functionality on these nodes, just the bare minimum that baremetal did, which is listing instances, rebooting them, and deleting them. Launch is excluded for the reasoning described above.

    We have also asked the ironic team to help us provide a baremetal API extension which knows how to talk to ironic, but this was identified as a need fairly late in the cycle and I expect it to be a request for a feature freeze exception when the time comes.

    The current plan is to remove the baremetal driver in the Kilo release.

    Previously in this post I alluded to the review mechanism we're using for the ironic driver. What does that actually look like? Well, what we've done is ask the ironic team to propose the driver as a series of smallish (500 line) changes. These changes are broken up by functionality, for example the code to boot an instance might be in one of these changes. However, because of the complexity of splitting existing code up, we're not requiring a tempest pass on each step in the chain of reviews. We're instead only requiring this for the final member in the chain. This means that we're not compromising our CI requirements, while maximizing the readability of what would otherwise be a very large review. To stop the reviews from merging before we're comfortable with them, there's a marker review at the beginning of the chain which is currently -2'ed. When all the code is ready to go, I remove the -2 and approve that first review and they should all merge together.

    In the next post I'll cover the state of adding DB2 support to nova.

    Tags for this post: openstack juno nova mid-cycle summary ironic
    Related posts: Juno nova mid-cycle meetup summary: nova-network to Neutron migration; Juno nova mid-cycle meetup summary: social issues; Juno nova mid-cycle meetup summary: scheduler; Juno nova mid-cycle meetup summary: slots; Chronological list of Juno Nova mid-cycle meetup posts; Juno nova mid-cycle meetup summary: conclusion

posted at: 01:49 | path: /openstack/juno | permanent link to this entry