On 06/05/2014 15:31, Lars Kurth wrote:
Hi all, I summarized actions on http://wiki.centos.org/SpecialInterestGroup/Virtualization/Status tagged with 06/05 The meeting LOG is below Regards Lars
Didn't realize the log would turn out that bad. Here we go again ...
lars_kurth: How do you want to run this? We have a set of loose ends: the roadmap via http://lists.centos.org/pipermail/centos-virt/2014-April/003763.html kbsingh: if someone (lars_kurth) wants to just do it point by point we can run through those. lars_kurth: And some open actions: http://wiki.centos.org/SpecialInterestGroup/Virtualization/Status kbsingh: there is a 60 min hard stop at the end jonludlam: hi all lars_kurth: How about the following: Actions first, then George can do the roadmap? kbsingh: ok, works for me lars_kurth: Do we all have http://wiki.centos.org/SpecialInterestGroup/Virtualization/Status up? kbsingh: I do gwd:yep pasik:yep lars_kurth:kbsingh: there were 3 technical items on you. I know you and hughesjr and gwd had a conversation last week. Is there anything that can be ticked off in the technical category? kbsingh:I believe gwd is setup with the basic workflow, and has git access we only imported the main xen repo at this point, but if things are looking ok and if the process is something we can work with - i can go ahead and import the rest of the repos gwd:So it's imported into git.centos.org? kbsingh:humm lars_kurth: kbsingh: what would the URL be? kbsingh:https://git.centos.org/project/sig-virt is where it should come up on lars_kurth: definitely there kbsingh:right, so the blocker was how are we going to organise the git repos on github - are we going to setup some teams at the project level or the repos level kbsingh:http://lists.centos.org/pipermail/centos-devel/2014-April/010175.html is the conversation kbsingh:i dont believe we all got to a result there. lars_kurth: Do we need to reply to this thread? Or is this more general? gwd:Well you had asked about having a different org for each sig, and Karsten said that sounded reasonable. Is there any reason not to give that a try for now? lars_kurth: Can we close this now. Or do we just have an action to engage with the discussion? gwd:I can reply to the thread. kbsingh: There are a couple of threads that fall out from this, eg. where is the kernel going to be maintained - and is every sig that needs a kernel then going to need to maintain the entire thing or can we just have a single git repo, with sig's maintaining their own branches hughesjr:I see a meeting in progress ...cool lars_kurth: hughesjr: hi. A little painful on IRC, but welcome lars_kurth: kbsingh: does sounds like a centos-wide decision that needs to be made. I propose to take an action for gwd and me to replay to the respective threads. gwd:Is there really a difference? Isn't that the point of DVCS? kbsingh:gwd: for the sake of convenience, I'd say maybe we just trial the model of having everything under /CentOS/ and if or when we run into a problem, we can try to change things around gwd:That's certainly a lot easier to begin with. kbsingh:ok lars_kurth: Cool: I made a note kbsingh:lets take that away then as a todo lars_kurth: Added as new action. Gwd may need to tidy up if I misunderstood gwd:kbsingh: So you're going to clone all the repos into git.centos.org and github.com/CentOS/ ? kbsingh:I've replied to the thread as well kbsingh:gwd: yeah, I can go ahead and do that as well - not online right now, but it can be done today lars_kurth: Is “KaranbirSingh to put together list of repository names in Xen4CentOS such that we can use it as a baseline” - still open? kbsingh:yes. that should get resolved with the move lars_kurth: Alright. Move to Community? lars_kurth: My list policy item is still open gwd:What was the list policy question? I forget. lars_kurth: gwd: just send a reminder to people that posting to the list while not subscribed = mail discarded Do we want to keep this, or change it? gwd:discard> got it. lars_kurth: And we were discussing whether kbsingh wanted to attend the Hackathon. I will need to know pretty soon, as we are running out of space kbsingh:I do want to come to the Hackathon. i believe there is some libvirt people as well ? lars_kurth: Yes. Daniel Berrage. As well as some other Xen folks working on libvirt OK. In that case, I will reserve a space and add you to the wiki gwd:jonludlam: Do you know who from the XenServer team is coming? jonludlam: dave scott, me, not sure about others jonludlam:euanh, do you know? euanh:I'm hoping to come lars_kurth: Let me check – see http://wiki.xenproject.org/wiki/Hackathon/May2014#Confirmed_attendees gwdjonludlam: For this meeting, knowing that you & dave are coming is sufficient I think. jonludlam: David Vrabel and Andrew Cooper on that list gwd:lars_kurth: You have an outstanding item to e-mail the -virt mailing list. Are you planning on doing that? Does it make sense to do so if there are only a handful of places left? lars_kurth: not doing it, as we don’t have space now lars_kurth: kbsingh: OK, I will add a few more people as provisional to the list. kbsingh: will send instructions to sign you up fully kbsingh:there was also some word from ovirt guys. i dont know if they have someone local - but they ~ dont have xen support lars_kurth: kbsingh: I need to know ASAP who that would be, if they want to get a space kbsingh:ok, i can ping around lars_kurth: kbsingh: great pasik:iirc ovirt had some sort of Xen support in the beginning and then later they switched to "kvm only" gwd:Hopefully with the improved libvirt support, getting xen support back in should be straightforward. lars_kurth: pasik: I was copied on a thread which says that integrating ovirt with xen+libvirt may be non-trivial as ovirt makes assumptions underneath libvirt pasik:lars_kurth: yep gwd:Shall we put that as another agenda item and finish going through the actions? lars_kurth: there are no quick fixes. I would say let's get libvirt support better first and then look at next steps pasik:lars_kurth: agreed lars_kurth: gwd: How about adding it to longer term roadmap goals lars_kurth: gwd: I will add an action on you to create a roadmap wiki page. Is that OK? gwd:lars_kurth: Ack. gwd:lars_kurth: We can certainly have a wishlist -- the question is who's going to do the work. lars_kurth: gwd: good question. I think wishlist = looking for a volunteer lars_kurth: OK. The last set of open actions was Publicty lars_kurth: The state of play was that we got the centos board feedback, made fixes and reported back to Karsten Karsten still has to ACK lars_kurth: I asked for Advisory Board approval on the assumption that he would do so lars_kurth: So the blocking issue is for the centos board to ACK lars_kurth: kbsingh: can you take this on or chase Karsten? kbsingh:lars_kurth: i can ping him lars_kurth: thanks. Tracked everything in the minutes lars_kurth: gwd: over to you lars_kurth: gwd: I think we are all looking at http://lists.centos.org/pipermail/centos-virt/2014- and responses lars_kurth: Do we have any disagreements / open questions in this thread? gwd:Roadmap> I sent out a mail with a basic discussion of what we talked about two weeks ago; I don't think there were any additions or objections gwd:So I'll make a concrete to-do list and post it on a wiki page. kbsingh:cool gwd:I think one thing kbsingh suggested is that we want to have most of the basic stuff sorted out before RHEL 7 comes out in... what, June / July? jonludlam: Not sure if this is the right time to bring this up, but I was wondering about bumping the version of ocaml used when building xen pasik:I think the current xen4.2 rpms are built against ocaml3 jonludlam: yep pasik:and there was/is the dev repo against ocaml4 aswell jonludlam: I think at some point in xen4centos's history, there was ocaml4 in there pasik:yep pasik:so first we need to get the xen 4.4 rpms built, probably against ocaml4 gwd:What happened with that? I don't know any of the history pasik:gwd: what happened is, we just decided to go with the "distro stock ocaml3" because xapi stuff wasn't done yet hughesjr:this issue is it breaks everything else ocaml in the distro jonludlam: yes, ocaml is incredibly picky about versions gwd:Not used to being enterprise yet, I take it... hughesjr:that means, we have to take on also fixing other distro things that are ocaml pasik:so do we want to first build xen 4.4 again ocaml3 ? jonludlam: right, unless there's a way for things to co-exist pasik:+st hughesjr:jonludlam: I suppose we can do an SCL for xen stuff gwd:and there's no way to have both versions somehow, and have each package use the library it needs? jonludlam : recent version of ocaml are much happier about coexisting with other versions hughesjr:gwd: if we do ocaml4 as an SCL that would work .. now if we can is another issue jonludlam: hughesjr, would the SCL _contain_ xen then? gwd:What does SCL stand for? hughesjr:software collections jonludlam: == coexistence of different versions of things gwd:Software CoLlection. Nice. pasik:xen 4.4.0 src.rpm (for fedora 21) here: ftp://ftp.funet.fi/pub/mirrors/fedora.redhat.com/pub/fedora/linux/development/rawhide/source/SRPMS/x/xen-4.4.0-3.fc21.src.rpm gwd:pasik: kbsingh showed me how the basic CentOS build system works, so I'm going to try doing the 4.4 update once the repos are sorted out. gwd:(Maybe after doing a more minor update as a test first.) pasik:gwd: that sounds good gwd:OK, so can we update to 4.4, then update libvirt, and look into doing ocaml for Xen as an SCL? hughesjr:gwd: that sounds like a good plan jonludlam: How would an SCL work if xen depends upon it? gwd:Are there any other package updates we need to do? pasik:so the first iteration of xen 4.4 rpms would be built against ocaml3, so there's an upgrade path from xen 4.2 rpms ? DV:missed the libvirt/xen discussion earlier lars_kurth: DV: I don't think we had a specific discussion before DV:lars_kurth: just a few exchange on IRC ^^^ lars_kurth: gwd: I guess the question is which libvirt version gwd:lars_kurth: On the list we decided 1.2.3, the same as Ubuntu 14.04. (Or maybe it was 1.2.2 -- whichever one Ubuntu is using). DV: lars_kurth: what is in Fedora 20 get backports, but it's a bit 'old' already gwd:DV: We seem to be talking about libvirt now -- did you have a comment / question / concern? hughesjr:lars_kurth: I think whatever version of libvirt is in ubuntu 14.04 LTS is a good starting point (seriously :D) gwd:libvirt 1.2.2 it was, I think. DV:gwd: mostly watching, and seeing if there is pit to avoid lars_kurth: We could agree on this for now and decide to upgrade later (after the Hackathon discussion) gwd:DV: Well we're planning on updating to libvirt 1.2.2, and then maybe backporting important functionality (such as live migration). pasik:too bad el7 looks to have libvirt 1.1.1 DV:pasik: usually in RHEL there is a lot of backports, so base version not always a good indicator DV:notes other SIGs may want libvirt updates without becoming over zealous might be worth checking around to minimize builds gwd:DV: Noted. Know of any that we should coordinate with? lars_kurth: DV, kbsingh: do you have any visibility to what other SIG's need? If not, we should maybe poll on the centos-devel list kbsingh:lars_kurth: yes, and no. i think we should deliver something, anything, just ship it gwd:I think going forward, all changes / pull requests will be posted to the centos-virt list, so any specific change can be objected to when it actually gets submitted. kbsingh:that gives people a target to point at otherwise we might get into massive wish lists DV:kbsingh, if other sigs are not more advanced, ACk gwd:kbsingh: That sounds like a plan. lars_kurth: kbsingh: works for me pasik:DV: yep kbsingh:yeah, i think the virtsig is perhaps the best place at the moment gwd:If kbsingh / hughesjr have their fingers in other sigs, they can bring us in when we need to start coordinating about something. lars_kurth: Works for me. So we agreed to start with 1.2.2? gwd:That's what we talked about on the -virt list, and nobody has objected. pasik:let's do what ubuntu 14.04 lts has; so libvirt 1.2.2 lars_kurth: Just confirming for the minutes. Agreed then. kbsingh:1.2.2 works for me, if - (1) we have qemu rbd supported for both ceph and gluster ( libfgfacl ) DV:there have been a lot of libxl updates in 1.2.3, but you know better what you need pasik:from earlier.. did we agree to build the first xen 4.4 rpms against ocaml3, so the current users of el6 xen 4.2 rpms can upgrade smoothly ? kbsingh:pasik: if we didnt, then we should kbsingh:hughesjr: ^ ? ocaml3 ? pasik:DV: those libxl updates need to be backported jonludlam: pasik, does the ocaml version have much bearing on the upgrade? gwd:Should we move the really technical discussion to the mailing list? lalatenduM: kbsingh, I think libvirt 1.2.3 has few imp bug fixes for gluster DV:pasik: and lot of code churn in 1.2.3, this may not be that easy pasik:hmm kbsingh:lalatenduM: in that case, maybe we should consider 1.2.3 lars_kurth: How about, we m we propose 1.2.2 on the list and ask for any backports necessary (pointing to patches) gwd:Well if backporting is too hard, we can think about moving forward instead. kbsingh:how about we decide on a 1.2.x and take the conversation to the lists to decide pasik:so then let's move to 1.2.4 directly? hughesjr:well .. if we maintain the same major version as ubuntu, things will work better in both places and we can more easily reuse code gwd:libvirt said they would do maintenance updates for any downstream actively using it; I think CentOS would be big enough to be worth their consideration. kbsingh:given DV has an eye on libvirt, what do you think ? DV:Indent top-level labels by one space in ... Use K&R style for curly braces in ... that kind of patches goes in the way of backports DV:but hard to tell a priori if it really will make it hard hughesjr:we can do whatever you guys want ... but centos is not intended to be fedora gwd:Going with 1.2.2 was just an attempt to reduce the burden on upstream. DV:gwd: Well if backporting is too hard, we can think about moving forward instead. : that's reasonable gwd:Well let's look at 1.2.2 and try a few backports and see how it works. DV:ack lars_kurth: Agreed. Will add an action on gwd DV: used to assemble RHEL libvirt builds for too long, some reflexes persists gwd: DV: I'd much rather learn the easy way, from your experience, than the hard way, from my own. DV:gwd: TBH the libvirt team use to rebase frequently, 6.1, 6.2, 6.3 ... because code changes fast and want to minimize backport loads DV:gwd: that said being 0.0.2 behind means only 2 months, should be okay pasik:jonludlam: about ocaml version: like said the current xen 4.2 rpms are built against ocaml3, and if xen 4.4 rpms added ocaml4 as a requirement, then all the distro ocaml stuff would break in the xen 4.2 -> xen 4.4 upgrade.. I don't think we want to do that. pasik:jonludlam: so I'd say let's first build xen 4.4 against ocaml3; and then let's figure out the SCL stuff so we can properly do xen 4.4 with ocaml4 jonludlam: pasik, they shouldn't be run-time requirements jonludlam: pasik, btw, no problem starting with ocaml3 with a view to seeing how ocaml4 might turn out pasik:jonludlam: ok gwd:One thing we might want to talk about (maybe on the -devel list) is changing the "binary blob" thing in the CentOS build system to make it earier to collaborate with people who don't have access to the "blob repo" (sorry I don't know the proper terms for these things) pasik:gwd: lookasides or something kbsingh:gwd: ok, if you start that off, we can hash it out ( the lookaside thing ) gwd:Most of the blobs in Xen4Centos seem to be upstream tarballs. One option would be to have a URL of the upstream, with a local cache to make builds reliable. kbsingh:gwd: i guess people can still checkout and do local things, and the get_sources.sh could be injected into the git repo itself pasik:gwd: I was able to do custom local rebuilds of centos src.rpms; I used the get_sources.sh kbsingh:i am guessing there isnt a LTS sort of version - so we'd need a plan to keep things patched without needing a system update every 2 months gwd:pasik: The hard part would be pull requests. pasik:gwd: true kbsingh:gwd: pasik: pull requests on binary data will always be against ( or well, used to be against ) a new srpm so we just run the import kbsingh:but we can also have remote urls in the .package.metadata file - and people can request merges against that kbsingh:do we have any other major points on the agenda gwd:I don't think so -- I'll kick off a discussion on -virt about packages to update gwd:And make a roadmap wiki lars_kurth: Anything else we need to discuss? pasik:lars_kurth: I think we should write something about ocaml version lars_kurth: pasik, jonludlam: no - can I get you to make a proposal on the list jonludlam: I believe there's been some discussion before about the structure of the xen rpms - is that minuted anywhere? jcpunk:^^+1 DV:gwd: my suggestions, use git, cherry pick from upstream into a set of commit bringing what you want on top of your branch, use that as you patch list kbsingh:jcpunk: +1 to what jcpunk:looking for documented structure of rpms DV:gwd: see the libvirt rpms from Fedora or RHEL they should be largely explicit kbsingh:this was on the virt list ages ago. we essentially went with the carry-in-spec for the leaf nodes needed at build time. gwd:DV: That seems reasonable. jcpunk:kbsingh: ah, got it kbsingh:i suspect jonludlam is talking about the entire wider scope pasik:lars_kurth: yeah sure, jonludlam is probably the expert with ocaml stuff lars_kurth: OK: will add action jonludlam: right, interested in splitting dom0 rpms from domU, etc kbsingh:jonludlam: i dont think we've had that conversation at this point. but we should jonludlam: ok, perhaps next meeting then? kbsingh:if we dont have any other items, i just want to bring one in - and this is past the scheduled hour, but worth talking about lars_kurth: Maybe next time or on list? jonludlam: gotta run to another meeting now gwd:DV: I'm fairly new to RPM building: I take it there's a step where you explicitly export base.tip as a series of patches? gwd:kbsingh: I've got time. kbsingh:we have a bit of a problem with el7rc kernel not doing pv out of the blocks gwd:That's not as a guest, right? pasik:kbsingh: only redhat kernel team can do something about that.. kbsingh:when i say a bit of a problem - i mean this is a major train crash kind of problem, given how much of hosting and cloud is pv only pasik:gwd: redhat disabled xen pv support for el7 kernel DV:gwd: yes you can ask git to give them a name based on commit text, and in the rpm spec you list them in order pasik:gwd: (xen hvm is still supported) kbsingh:pasik: right, i dont care about the rhel7 state of play, but we might need to do something in the centos ecosystem and own it locally pasik:kbsingh: yep, definitely we should "fix" that for centos7 kbsingh:gwd: it is.. as a guest, cant run rhel7 as a pv guest. only hvm-pv and hvm ( of course ) kbsingh:I will get the git repos in place and the acl's setup this week for the xen stuff to happen gwd:kbsingh: Oh, that's right. But does it have PVHVM support? kbsingh:and we can maybe target a 4.4 build as a testing/qa process. kbsingh:gwd: yes. pvhvm works gwd:kbsingh: So are you thinking of having a CentOS guest kernel that supports PV? pasik:yep pasik:so probably just a small .config tweak hughesjr:gwd: well, we will have a kernel that supports dom0 too at some point DV:gwd: git log --pretty=%f -1 your-commit and use that as a base for the filename of the associated patch kbsingh:gwd: i dont know - dont want to predecide and havent really spent any time thinking this though - but i do know that we need to ( spend the time, think it through, do something ) kbsingh:lars_kurth are we calling this meeting closed ? ( i am already on the phone waiting for the next one to open ) lars_kurth: yes: let's call it closed pasik:kbsingh: yes, i think we should have some kernel variant with "fixed" .config gwd:Would the PV kernel be basically the same as the Xen4CentOS dom0? Or a stripped-down version? Or were you thinking of taking the upstream RH kernel and re-enabling PV (meaning another separate kernel branch to maintain)? pasik:gwd: i think it should upstream-el7-kernel with just the .config tweak hughesjr:gwd: we can do more than one if we want (or need to) pasik:gwd: and any xen pv specific patches, if there's a need for those ever pasik:gwd: so maintenance-wise it would be quite low effort.. hopefully gwd:pasik: Well, one would hope... gwd:And would that be a change to the main CentOS kernel? Or a separate package that's included for cloud targets? kbsingh:change to main kernel would need quite a major reason, and a lot of shouting and requests - I suspect we dont/wont get that initially. kbsingh:we could carry it on the install media though... so its available to use right off the bat, post install ( specially if its the same version as the install / main kernel ) gwd:That sounds pretty do-able. pasik:yep gwd:OK -- lars_kurth, are you going to send out a meeting report? lars_kurth: gwd: pointing to the actions and appending this log for now