)]}'
{"specs/mitaka/approved/add-maxcpus-support-for-qemu-kvm.rst":[{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"291d1c9f54ad877a7d192e375b20c6692b615cb2","unresolved":false,"context_lines":[{"line_number":14,"context_line":"libvirt(since version 1.2). This feature is userful for some users to scale"},{"line_number":15,"context_line":"VM without reboot. In order to implement the feature, the maxcpus must be"},{"line_number":16,"context_line":"specified by qemu command line. You can get more information via the below"},{"line_number":17,"context_line":"link."},{"line_number":18,"context_line":""},{"line_number":19,"context_line":"http://wiki.qemu.org/Features/CPUHotplug"},{"line_number":20,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_65cd4584","line":17,"updated":"2015-11-27 18:20:35.000000000","message":"In lieu of an implementation of live resize is this actually useful for something - or is the intent that the deployer would resize the instance manually up to the max vcpus using e.g. virsh without Nova being aware of it?","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"e4b09c49e6a8ac0bc8a08bab1c1273985781b9f1","unresolved":false,"context_lines":[{"line_number":14,"context_line":"libvirt(since version 1.2). This feature is userful for some users to scale"},{"line_number":15,"context_line":"VM without reboot. In order to implement the feature, the maxcpus must be"},{"line_number":16,"context_line":"specified by qemu command line. You can get more information via the below"},{"line_number":17,"context_line":"link."},{"line_number":18,"context_line":""},{"line_number":19,"context_line":"http://wiki.qemu.org/Features/CPUHotplug"},{"line_number":20,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_9695b889","line":17,"in_reply_to":"9a8ffd7b_1e259f5a","updated":"2015-11-30 19:20:57.000000000","message":"Right, but I\u0027m trying to understand why that is actually useful as a thing to implement as a standalone \"phase\". It doesn\u0027t seem useful as a standalone thing and no rationale for it being one is in the spec.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"8a20a7751998db9e31bd68fe7ba8f0589ea3d018","unresolved":false,"context_lines":[{"line_number":14,"context_line":"libvirt(since version 1.2). This feature is userful for some users to scale"},{"line_number":15,"context_line":"VM without reboot. In order to implement the feature, the maxcpus must be"},{"line_number":16,"context_line":"specified by qemu command line. You can get more information via the below"},{"line_number":17,"context_line":"link."},{"line_number":18,"context_line":""},{"line_number":19,"context_line":"http://wiki.qemu.org/Features/CPUHotplug"},{"line_number":20,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_1e259f5a","line":17,"in_reply_to":"9a8ffd7b_65cd4584","updated":"2015-11-28 09:19:44.000000000","message":"Yes, My finial purpose is to live add vcpu via nova, I had already implemented it, but I found that lieu also is implementing similar functionality, and he pushed a spec for it, so I can only push the spec to add maxcpus to libvirt\u0027s xml if we want to support cpu hotplug for qemu/kvm, Lieu should also need it.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"291d1c9f54ad877a7d192e375b20c6692b615cb2","unresolved":false,"context_lines":[{"line_number":29,"context_line":""},{"line_number":30,"context_line":"The core goal is to help other developer to further implement"},{"line_number":31,"context_line":"CPU Hotplug in qemu, they are able to implement CPU Hotplug feature"},{"line_number":32,"context_line":"via setting \"hw:max_vcpus\u003dxx\" to flavor\u0027s extra specs"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Proposed change"},{"line_number":35,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_e5d93558","line":32,"updated":"2015-11-27 18:20:35.000000000","message":"Per comment on https://review.openstack.org/#/c/141219 how does this intersect with non-QEMU/KVM hypervisors? Is this a QEMU/KVM specific implementation detail that we are leaking to the user or is a similar construct required for the other drivers as well?","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"e4b09c49e6a8ac0bc8a08bab1c1273985781b9f1","unresolved":false,"context_lines":[{"line_number":29,"context_line":""},{"line_number":30,"context_line":"The core goal is to help other developer to further implement"},{"line_number":31,"context_line":"CPU Hotplug in qemu, they are able to implement CPU Hotplug feature"},{"line_number":32,"context_line":"via setting \"hw:max_vcpus\u003dxx\" to flavor\u0027s extra specs"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Proposed change"},{"line_number":35,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_d63fc085","line":32,"in_reply_to":"9a8ffd7b_48084c28","updated":"2015-11-30 19:20:57.000000000","message":"That VMware page is simply speaking to the generally applicable rules for the total maximum number of VCPUs that can be supported (with or without hotplugging). If you review the rest of the documentation (e.g. https://pubs.vmware.com/vsphere-50/index.jsp#com.vmware.vsphere.vm_admin.doc_50/GUID-76FC7E9F-8037-4C8E-BEB9-91C266C1EA9A.html ) I believe you will find VMware does not actually require you to set the maximum number of vCPUs that can be plugged up front in the same way as the other hypervisors.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"3b71c37d043dbeae173f7251a31aed4b3e7c8739","unresolved":false,"context_lines":[{"line_number":29,"context_line":""},{"line_number":30,"context_line":"The core goal is to help other developer to further implement"},{"line_number":31,"context_line":"CPU Hotplug in qemu, they are able to implement CPU Hotplug feature"},{"line_number":32,"context_line":"via setting \"hw:max_vcpus\u003dxx\" to flavor\u0027s extra specs"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Proposed change"},{"line_number":35,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_ed3b0037","line":32,"in_reply_to":"9a8ffd7b_d63fc085","updated":"2015-12-02 03:49:24.000000000","message":"Okay, I would look a deep dive to VMware about it.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"8a20a7751998db9e31bd68fe7ba8f0589ea3d018","unresolved":false,"context_lines":[{"line_number":29,"context_line":""},{"line_number":30,"context_line":"The core goal is to help other developer to further implement"},{"line_number":31,"context_line":"CPU Hotplug in qemu, they are able to implement CPU Hotplug feature"},{"line_number":32,"context_line":"via setting \"hw:max_vcpus\u003dxx\" to flavor\u0027s extra specs"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Proposed change"},{"line_number":35,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_48084c28","line":32,"in_reply_to":"9a8ffd7b_e5d93558","updated":"2015-11-28 09:19:44.000000000","message":"Both Xen and VMware also need similiar option.\n\nhttp://backdrift.org/how-to-hot-addremove-vcpus-from-a-xen-domain\nhttps://pubs.vmware.com/vsphere-50/index.jsp#com.vmware.vsphere.vm_admin.doc_50/GUID-13AD347E-3B77-4A67-B3F4-4AC2230E4509.html","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"e4b09c49e6a8ac0bc8a08bab1c1273985781b9f1","unresolved":false,"context_lines":[{"line_number":37,"context_line":"Add \"hw:max_vcpus\u003dxx\" to flavor\u0027s extra specs. The vcpu"},{"line_number":38,"context_line":"element of instance\u0027s xml would look like this:"},{"line_number":39,"context_line":""},{"line_number":40,"context_line":"\u003cvcpu current\u003d2\u003e4\u003c/vcpu\u003e"},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"Alternatives"},{"line_number":43,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_566bd085","line":40,"updated":"2015-11-30 19:20:57.000000000","message":"How is this factored into scheduling decisions? Obviously only 2 are used at boot, but it seems like your intent is that the operator would change this external to nova using virsh until live-resize is implemented?","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"3b71c37d043dbeae173f7251a31aed4b3e7c8739","unresolved":false,"context_lines":[{"line_number":37,"context_line":"Add \"hw:max_vcpus\u003dxx\" to flavor\u0027s extra specs. The vcpu"},{"line_number":38,"context_line":"element of instance\u0027s xml would look like this:"},{"line_number":39,"context_line":""},{"line_number":40,"context_line":"\u003cvcpu current\u003d2\u003e4\u003c/vcpu\u003e"},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"Alternatives"},{"line_number":43,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_6d087092","line":40,"in_reply_to":"9a8ffd7b_566bd085","updated":"2015-12-02 03:49:24.000000000","message":"Yes, We would use live resize cpu via \"virsh qemu-monitor-command\" util lieu\u0027s spec and code is merged.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"c5eaae4a01334bfaf3882493e0bd2189d548e713","unresolved":false,"context_lines":[{"line_number":79,"context_line":""},{"line_number":80,"context_line":"AFAIK, some guys are trying to integrating CPU Hotplug feature to nova."},{"line_number":81,"context_line":"They can use the specs to add vcpu for qemu via setting \"hw:max_vcpu\u003dxx\""},{"line_number":82,"context_line":"to extra specs of flavor"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"Developer impact"},{"line_number":85,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_12fb2a49","line":82,"updated":"2015-11-30 13:31:23.000000000","message":"Possibly, feels like this should be part of that spec, rather than separate.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"d4b42d0808e9959128799333362ac3f30e0dc055","unresolved":false,"context_lines":[{"line_number":79,"context_line":""},{"line_number":80,"context_line":"AFAIK, some guys are trying to integrating CPU Hotplug feature to nova."},{"line_number":81,"context_line":"They can use the specs to add vcpu for qemu via setting \"hw:max_vcpu\u003dxx\""},{"line_number":82,"context_line":"to extra specs of flavor"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"Developer impact"},{"line_number":85,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a740942_2d60ed00","line":82,"in_reply_to":"9a8ffd7b_055034d3","updated":"2015-12-06 01:29:04.000000000","message":"I understand what you said, Hmm, so far the purpose of adding the spec is not enable a workaround to add dynamically current cpu resource via virsh, but provide a  method to add \"hw:max_vcpus\" to make qemu or qemu-kvm be able to support cpu hotplug. If I do not add the spec, lieu also must be add it, to implement live resize vcpu on qemu or qemu-kvm. so far, lieu\u0027s spec does not contain it. his spec is consist of :\n1. live resize the same host\n2. live migration and then live resize\n3. modify nova resize api\nI also agree that if not nova interface,  my spec is not useful to deployer, but it should be useful to lieu, he only simply set \"hw:max_vcpu\u003dxxx\" to support live resize vcpu on qemu or qemu-kvm.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"5cccb0dfac752543981e5fffa5d6c4017785f885","unresolved":false,"context_lines":[{"line_number":79,"context_line":""},{"line_number":80,"context_line":"AFAIK, some guys are trying to integrating CPU Hotplug feature to nova."},{"line_number":81,"context_line":"They can use the specs to add vcpu for qemu via setting \"hw:max_vcpu\u003dxx\""},{"line_number":82,"context_line":"to extra specs of flavor"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"Developer impact"},{"line_number":85,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_825d7f4a","line":82,"in_reply_to":"9a8ffd7b_12fb2a49","updated":"2015-11-30 16:30:05.000000000","message":"The reason of pushing my spec is that I want to make the implementation of that spec simpler, and that spec consists of three phrases, it really contains lots of thing,\nso  just like Andrea Rosa\u0027s suggestion, I also suggest split that spec, and make that spec simpler.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"397bc0a4798390698eaa78ed2591fb728513445c","unresolved":false,"context_lines":[{"line_number":79,"context_line":""},{"line_number":80,"context_line":"AFAIK, some guys are trying to integrating CPU Hotplug feature to nova."},{"line_number":81,"context_line":"They can use the specs to add vcpu for qemu via setting \"hw:max_vcpu\u003dxx\""},{"line_number":82,"context_line":"to extra specs of flavor"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"Developer impact"},{"line_number":85,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_055034d3","line":82,"in_reply_to":"9a8ffd7b_16a411a4","updated":"2015-12-02 16:19:06.000000000","message":"No, I specifically mean modifying the number of current CPUs up to the assigned maximum. As you note in another comment even with this spec implemented there is no Nova interface to do this so the deployer has to do it using virsh, thus bypassing Nova. Yet this spec makes no attempt to document the ramifications of this in terms of the inherent issues with the admin assigning what are effectively finite resources (even if overcommitting) without the scheduler etc. knowing about it.\n\nMy opinion is that this needs to be documented in any spec that implements this functionality or part thereof, not simply pushed off as someone elses problem. I agree that the other spec needs to be split up but this does not appear to address a logical subset of the problem space, rather it\u0027s just a quick hack to enable a workaround without considering any of the design issues.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":18832,"name":"Wei Yang","email":"wei.yang@easystack.cn","username":"wyang"},"change_message_id":"3b71c37d043dbeae173f7251a31aed4b3e7c8739","unresolved":false,"context_lines":[{"line_number":79,"context_line":""},{"line_number":80,"context_line":"AFAIK, some guys are trying to integrating CPU Hotplug feature to nova."},{"line_number":81,"context_line":"They can use the specs to add vcpu for qemu via setting \"hw:max_vcpu\u003dxx\""},{"line_number":82,"context_line":"to extra specs of flavor"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"Developer impact"},{"line_number":85,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_16a411a4","line":82,"in_reply_to":"9a8ffd7b_36be4cd5","updated":"2015-12-02 03:49:24.000000000","message":"For quota or scheduling, IMO, it should be considered by lieu\u0027s spec, and not in this spec. This spec only implements hypervisor specific of live resize. If all of thing are implemented by lieu\u0027s spec, lieu must consider all of hypervisor(qemu/hyper-v/vmware/xen) to how to live resize cpu/memory/disk, and also must handle some implementations mentioned by lieu.  Yes, the spec is only impact on deployer, and it only can add cpu via virsh command until lieu\u0027s spec is implemented.\n\nDo you mean that it cannot modify maxcpus? If yes, maxcpus is not able to be modified dynamically, if you want  to modify it,  you can only set new value to extra spec of flavor again. If you mean that you want to downstream resources(cpu/memory), by now, qemu/kvm does not support to unhotplug  cpu/memory.","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"e4b09c49e6a8ac0bc8a08bab1c1273985781b9f1","unresolved":false,"context_lines":[{"line_number":79,"context_line":""},{"line_number":80,"context_line":"AFAIK, some guys are trying to integrating CPU Hotplug feature to nova."},{"line_number":81,"context_line":"They can use the specs to add vcpu for qemu via setting \"hw:max_vcpu\u003dxx\""},{"line_number":82,"context_line":"to extra specs of flavor"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"Developer impact"},{"line_number":85,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"9a8ffd7b_36be4cd5","line":82,"in_reply_to":"9a8ffd7b_825d7f4a","updated":"2015-11-30 19:20:57.000000000","message":"The problem is that this phase doesn\u0027t seem to make sense as a standalone item nor does it really seem like it simplifies the other one in any significant way. As specified it seems like setting a relatively simple config option - which yes would be required for live-resize - without addressing any of the issues that need to be handled (e.g. how does this impact my quota, scheduling, etc.?) before we can deliver something that is actually useful.\n\nCase in point, the only impact specified is for deployers - the impact being that they can set this value, but aren\u0027t given an interface to modify current. So what is the benefit to the deployer?","commit_id":"cf430a72a5e74b36106dbb30144010b83b2466ef"}]}