)]}'
{"specs/newton/compute-abstraction.rst":[{"author":{"_account_id":9237,"name":"Kevin Fox","email":"kevin@efox.cc","username":"kfox1111"},"change_message_id":"a833860c333312a5981d59a55c676181dc3c7f38","unresolved":false,"context_lines":[{"line_number":48,"context_line":"  docker. Unfortunately, this is not a good enough solution. The exposed API is"},{"line_number":49,"context_line":"  limiting, the driver itself is in support mode only and it does not support"},{"line_number":50,"context_line":"  [docker swarm]_, which is ideal for deploying database clusters."},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"* [nova-ironic]_: This is the most common way to deploy bare metal nodes in"},{"line_number":53,"context_line":"  OpenStack. It\u0027s also possible to use [ironic]_ in standalone mode but it\u0027s"},{"line_number":54,"context_line":"  less common."}],"source_content_type":"text/x-rst","patch_set":4,"id":"dab17558_062a637e","line":51,"updated":"2016-05-08 21:21:35.000000000","message":"I\u0027ve had issues with non k8s orchestration with lack of co\u0027scheduling. k8s has a notion of pods which allows the database container and corresponding trove guest agent container to be co\u0027scheduled together. docker swarm doesn\u0027t allow this, and I don\u0027t think Mesos does either. So they are less then ideal for deploying clusters.","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"},{"author":{"_account_id":9664,"name":"amrith","email":"amrith.kumar@gmail.com","username":"amrith"},"change_message_id":"885684787772cc1eefa6f9a39288679f63a2d67d","unresolved":false,"context_lines":[{"line_number":39,"context_line":"currently imposed by Trove on this area have been highlighted on some threads"},{"line_number":40,"context_line":"and they\u0027ve also been brought up on other mediums."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"It\u0027s currently possible to deploy database instances on bare metal using"},{"line_number":43,"context_line":"[ironic]_ and on containers (docker specifically) using [nova-docker]_ but these"},{"line_number":44,"context_line":"two drivers are not enough. Please, find some extra info below about these 2"},{"line_number":45,"context_line":"drivers."},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"* [nova-docker]_: This driver can be used to deploy a database instance on"},{"line_number":48,"context_line":"  docker. Unfortunately, this is not a good enough solution. The exposed API is"},{"line_number":49,"context_line":"  limiting, the driver itself is in support mode only and it does not support"},{"line_number":50,"context_line":"  [docker swarm]_, which is ideal for deploying database clusters."},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"* [nova-ironic]_: This is the most common way to deploy bare metal nodes in"},{"line_number":53,"context_line":"  OpenStack. It\u0027s also possible to use [ironic]_ in standalone mode but it\u0027s"},{"line_number":54,"context_line":"  less common."},{"line_number":55,"context_line":""},{"line_number":56,"context_line":"Taking advantage of different deployment models is a key aspect of scaling"},{"line_number":57,"context_line":"services like Trove and providing a better cloud service. By allowing deployers"}],"source_content_type":"text/x-rst","patch_set":4,"id":"1a122d0e_9264044a","line":54,"range":{"start_line":42,"start_character":0,"end_line":54,"end_character":14},"updated":"2016-04-22 11:08:09.000000000","message":"It would be worthwhile, either as part of this specification, or as part of some discussions we have on this to highlight and contrast the specific issues raised in [1] and [2]. I\u0027m posting some of that here but the comments will get lost to history and it would therefore be worthwhile if these are discussed as alternatives in the spec itself.\n\nFor Trove, in the short term, the prize is being able to deploy compute into containers. The long term goal may well be to support other compute engines and Magnum is certainly not the only one. \n\nThere are some ways to accomplish the short term goal; nova-docker is one, libvirt-lxc is another.\n\nAs described in [1] by Adrian, and I agree with this assertion, for Trove using nova-docker, libvirt-lxc, (or nova-compute-lxd) and drivers like that provide us with what we want in the short term.\n\nMagnum is a different thing altogether. And I recognize that users of Magnum would ask why it is that Trove doesn\u0027t leverage Magnum, and why it is that they are being asked to operate Magnum and also deal with Nova drivers for containers.\n\nAnd Trove is by no means in a unique position here and it appears from [3] that the Magnum team (for example) is not entirely decided on the subject of whether Magnum should provide a single unifying container API, and it appears from other opinions expressed in this same set of email threads that there is more that remains to be learned here.\n\nIn [1], Adrian proposes three alternatives and he recommends one of them (the last paragraph). Personally, I think the first recommendation he makes is the best for Trove at this stage and this spec is going in what appears, at least in the short term, to be the wrong path.\n\nI propose that we go down the first of the three alternatives above, and in [1], using nova as the only compute service in the short term but properly supporting the many kinds of computing that Nova can offer today, and do the many things that are going to required in the rest of Trove to make that happen, things that are not yet recognized as being required in this spec.\n\nFor example, this spec indicates that there will be no guest changes. Unfortunately, that\u0027s not going to be the case.\n\nOnce we do those things, we would be in a better position to make a transition to Magnum, but I believe that the wind is in the direction of Nova offering an LCD Magnum driver.\n\nTo be continued in Austin.\n\n[1] http://lists.openstack.org/pipermail/openstack-dev/2016-April/091982.html\n[2] http://lists.openstack.org/pipermail/openstack-dev/2016-April/093055.html\n[3] http://lists.openstack.org/pipermail/openstack-dev/2016-April/091997.html","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"},{"author":{"_account_id":9664,"name":"amrith","email":"amrith.kumar@gmail.com","username":"amrith"},"change_message_id":"84d981430b65f204bcd905cdc706cab0e35d7331","unresolved":false,"context_lines":[{"line_number":200,"context_line":"Guest Agent"},{"line_number":201,"context_line":"-----------"},{"line_number":202,"context_line":""},{"line_number":203,"context_line":"No changes to the guest agent are planned."},{"line_number":204,"context_line":""},{"line_number":205,"context_line":"Alternatives"},{"line_number":206,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"1a122d0e_0b0c3a41","line":203,"range":{"start_line":203,"start_character":0,"end_line":203,"end_character":42},"updated":"2016-04-22 13:55:03.000000000","message":"I\u0027m quite sure that this will not work. Been there, done that, have the t-shirt.","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"},{"author":{"_account_id":6159,"name":"Flavio Percoco Premoli","display_name":"flaper87","email":"flavio.percoco@flyrlabs.com","username":"flaper87"},"change_message_id":"fc19a877f85e70bea9779af0a6b271e899141da2","unresolved":false,"context_lines":[{"line_number":200,"context_line":"Guest Agent"},{"line_number":201,"context_line":"-----------"},{"line_number":202,"context_line":""},{"line_number":203,"context_line":"No changes to the guest agent are planned."},{"line_number":204,"context_line":""},{"line_number":205,"context_line":"Alternatives"},{"line_number":206,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"1a122d0e_32bc6cdb","line":203,"range":{"start_line":203,"start_character":0,"end_line":203,"end_character":42},"in_reply_to":"1a122d0e_0b0c3a41","updated":"2016-04-28 17:17:48.000000000","message":"Yup. FWIW, the spec is not complete. I should\u0027ve probably added a note in this section mentioning that it needed more content.","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"},{"author":{"_account_id":9237,"name":"Kevin Fox","email":"kevin@efox.cc","username":"kfox1111"},"change_message_id":"a833860c333312a5981d59a55c676181dc3c7f38","unresolved":false,"context_lines":[{"line_number":200,"context_line":"Guest Agent"},{"line_number":201,"context_line":"-----------"},{"line_number":202,"context_line":""},{"line_number":203,"context_line":"No changes to the guest agent are planned."},{"line_number":204,"context_line":""},{"line_number":205,"context_line":"Alternatives"},{"line_number":206,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"dab17558_c647bb49","line":203,"range":{"start_line":203,"start_character":0,"end_line":203,"end_character":42},"in_reply_to":"1a122d0e_32bc6cdb","updated":"2016-05-08 21:21:35.000000000","message":"Yeah, I think the guest agent does two things?\n 1. Installs/Initializes a fresh db\n 2. Controls the db once started, doing things like creating the users, db\u0027s, listing things, etc.\n\nMaybe the guest agent should be split into those two parts?\n\nPart 1 could have 3 modes:\n 1. native. as now.\n 2. docker start/init things using the local docker engine to start/restart containers locally.\n 3. uses kubernetes api\n\nPart 2 would just talk to the local mysql socket.\n\n-----\n\nIn a phased approach towards containerization,\nPhase 1, nova vm is launched, guest agent is local, db is containerized. Uses guest agent part 1 (mode 1) and part 2.\n\nPhase 2, nova vm is launched guest agent is containerized, db is containerized, uses guest agent part 1 (mode 2) and part 2. This is much easier for operators to deploy/upgrade. (If there is time in the schedule, phase 2 could be skipped and we go right to phase 3)\n\nPhase 3, nova vm is launched, single node kubernetes cluster is deployed, guest agent and db are launched in it. uses guest agent part 1 (mode 3) and part 2.\n\nPhase 4. Trove guest agent part 1 in mode 3 is run on trove controller node instead of in a vm. It launches pods to Magnum launched k8s clusters and a pod only contains the db container and trove guest agent container in part 2 mode. Hardly any code changes for this should be needed.\n\nAfter some time....\n\nPhase 5, write migration scripts to allow all existing instances to be converted to phase 4 instances.\n\nPhase 6, remove part 1 mode 1 and 2, and all the nova launching code from Trove. Trove can rely on advanced features like rolling upgrades always being available at this point.\n\nWith this phased approach, each phase is a small work unit can can be done and benefits can be felt by users even if the total work takes multiple Trove releases.","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"},{"author":{"_account_id":9237,"name":"Kevin Fox","email":"kevin@efox.cc","username":"kfox1111"},"change_message_id":"a833860c333312a5981d59a55c676181dc3c7f38","unresolved":false,"context_lines":[{"line_number":204,"context_line":""},{"line_number":205,"context_line":"Alternatives"},{"line_number":206,"context_line":"------------"},{"line_number":207,"context_line":""},{"line_number":208,"context_line":"* As an alternative to the proposed package structure, this work could add a"},{"line_number":209,"context_line":"  single package for [Magnum]_ and introduce an abstraction in that package for"},{"line_number":210,"context_line":"  each container technology. This structure would make it easier to migrate to a"}],"source_content_type":"text/x-rst","patch_set":4,"id":"dab17558_a669d7c0","line":207,"updated":"2016-05-08 21:21:35.000000000","message":"An unlisted option. Rely on just one COE. Support for multiple is a lot more work, and at this point, I\u0027m pretty sure standardizing on k8s would be a pretty reasonable thing to do.","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"},{"author":{"_account_id":9237,"name":"Kevin Fox","email":"kevin@efox.cc","username":"kfox1111"},"change_message_id":"a833860c333312a5981d59a55c676181dc3c7f38","unresolved":false,"context_lines":[{"line_number":221,"context_line":"  very unlike to happen any time soon. Given some of the other proposals for the"},{"line_number":222,"context_line":"  Newton cycle, it could be argued this is the right moment to implement this"},{"line_number":223,"context_line":"  spec and use future cycles to stabilize the code."},{"line_number":224,"context_line":""},{"line_number":225,"context_line":"Dashboard Impact (UX)"},{"line_number":226,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":227,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"dab17558_0678a38e","line":224,"updated":"2016-05-08 21:21:35.000000000","message":"There is a project starting up for this. Though I think direct k8s support will be added much sooner then an abstract api will be.\n\nhttps://github.com/kubernetes/kubernetes/issues/24982","commit_id":"6524e11913638cce328ed149e3e3a37a962045d3"}]}
