)]}'
{"/COMMIT_MSG":[{"author":{"_account_id":13997,"name":"SHIGEMATSU Mitsuhiro","email":"shigematsu.mitsuhiro@lab.ntt.co.jp","username":"pshige"},"change_message_id":"0600c718726287dc382b3485fa2dc0f1cae232c6","unresolved":false,"context_lines":[{"line_number":8,"context_line":""},{"line_number":9,"context_line":"This blueprint/spec proposes replacing the server groups API extension"},{"line_number":10,"context_line":"with an entirely different, simpler, and more flexible system for"},{"line_number":11,"context_line":"infoming the Nova scheduler about placement preferences at the time"},{"line_number":12,"context_line":"of launch, including \"soft\" scheduling and proximate-distance placement."},{"line_number":13,"context_line":""},{"line_number":14,"context_line":"Change-Id: I9a35eb9d13123bb7f020c382f6b402e91a08de31"}],"source_content_type":"text/x-gerrit-commit-message","patch_set":4,"id":"da20952f_cbad559c","line":11,"updated":"2015-08-30 03:18:16.000000000","message":"s/infoming/informing/","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"}],"specs/liberty/approved/generic-scheduling-policies.rst":[{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"a56a9f9dff43e8797d94a89e4b369a8af83cc82c","unresolved":false,"context_lines":[{"line_number":20,"context_line":"The current \"server group\" API extension is awkward, incomplete, and confusing"},{"line_number":21,"context_line":"to cloud users."},{"line_number":22,"context_line":""},{"line_number":23,"context_line":"The use case here is very simple: "},{"line_number":24,"context_line":""},{"line_number":25,"context_line":"Alice wants to launch an instance and make sure that the instance does"},{"line_number":26,"context_line":"not land on a compute host that contains other instances of that type."}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_ef113c05","line":23,"updated":"2015-05-21 08:09:55.000000000","message":"Nit... ending space.  :-)","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":8768,"name":"Chris Friesen","email":"chris.friesen@windriver.com","username":"cbf123"},"change_message_id":"49bf39f589c858ad1c559ceb453e178c5ea49695","unresolved":false,"context_lines":[{"line_number":88,"context_line":"describes generic placement preferences:"},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"`POST /servers` -- The main API call to launch an instance will add a new"},{"line_number":91,"context_line":"`placement` key to the resource body::"},{"line_number":92,"context_line":""},{"line_number":93,"context_line":"    \u0027placement\u0027: {"},{"line_number":94,"context_line":"        ..."}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_ddccfd6b","line":91,"updated":"2015-05-21 07:12:39.000000000","message":"Would it make sense to allow multiple \u0027placement\u0027 keys?  I\u0027m envisioning a scenario where we want to put A/B/C \u0027near\u0027 each other, and X/Y/Z \u0027near\u0027 each other, but ensure that the two groupings are \"not near\" each other.  Or is that getting unnecessarily general?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"0773a8870d21e98848951b5cf139a6ddcc54df6e","unresolved":false,"context_lines":[{"line_number":88,"context_line":"describes generic placement preferences:"},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"`POST /servers` -- The main API call to launch an instance will add a new"},{"line_number":91,"context_line":"`placement` key to the resource body::"},{"line_number":92,"context_line":""},{"line_number":93,"context_line":"    \u0027placement\u0027: {"},{"line_number":94,"context_line":"        ..."}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_5b7b64c1","line":91,"in_reply_to":"7a016987_ddccfd6b","updated":"2015-05-21 22:27:48.000000000","message":"I think unnecessarily complex at the moment :)","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"a56a9f9dff43e8797d94a89e4b369a8af83cc82c","unresolved":false,"context_lines":[{"line_number":115,"context_line":"   the scheduler should attempt a best-effort placement using less-rigid"},{"line_number":116,"context_line":"   placement policies. This is sometimes called \"soft (anti-)affinity\"."},{"line_number":117,"context_line":""},{"line_number":118,"context_line":" * `distance`: (defaults to 0) An unsigned integer that represents the relative"},{"line_number":119,"context_line":"   distance that the scheduler should attempt to place the instances."},{"line_number":120,"context_line":"   Different deployments are free to interpret this value in different ways."},{"line_number":121,"context_line":"   For example, let\u0027s say a large public cloud deployer wishes to allow users"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_cf98c0c2","line":118,"updated":"2015-05-21 08:09:55.000000000","message":"Great idea... but, if this is open to some level of interpretation, should we perhaps consider a more general user defined \"kwargs\" style parameter that lets various implementations pass other things they\u0027d like.\n\nI think if we don\u0027t have very explicit interpretations of all parameters, it might be worth consider this approach at least... as complete uniformity would be one reason to not do so, but we kind of lose that with this definition anyway?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"2b3c2d25eeab7d327912e66a0a8e6c856aa39ef3","unresolved":false,"context_lines":[{"line_number":115,"context_line":"   the scheduler should attempt a best-effort placement using less-rigid"},{"line_number":116,"context_line":"   placement policies. This is sometimes called \"soft (anti-)affinity\"."},{"line_number":117,"context_line":""},{"line_number":118,"context_line":" * `distance`: (defaults to 0) An unsigned integer that represents the relative"},{"line_number":119,"context_line":"   distance that the scheduler should attempt to place the instances."},{"line_number":120,"context_line":"   Different deployments are free to interpret this value in different ways."},{"line_number":121,"context_line":"   For example, let\u0027s say a large public cloud deployer wishes to allow users"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_99f564ee","line":118,"in_reply_to":"7a016987_1b2f2ca2","updated":"2015-05-22 00:29:01.000000000","message":"Good idea -- thoughts on the \"kwargs\" for other intelligence to be added by custom schedulers?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"0773a8870d21e98848951b5cf139a6ddcc54df6e","unresolved":false,"context_lines":[{"line_number":115,"context_line":"   the scheduler should attempt a best-effort placement using less-rigid"},{"line_number":116,"context_line":"   placement policies. This is sometimes called \"soft (anti-)affinity\"."},{"line_number":117,"context_line":""},{"line_number":118,"context_line":" * `distance`: (defaults to 0) An unsigned integer that represents the relative"},{"line_number":119,"context_line":"   distance that the scheduler should attempt to place the instances."},{"line_number":120,"context_line":"   Different deployments are free to interpret this value in different ways."},{"line_number":121,"context_line":"   For example, let\u0027s say a large public cloud deployer wishes to allow users"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_1b2f2ca2","line":118,"in_reply_to":"7a016987_cf98c0c2","updated":"2015-05-21 22:27:48.000000000","message":"Right, it kinda needs to be open to interpretation since different deployments may have very different ways of indicating distances to the user. I am also thinking about some kind of API like:\n\n GET /scheduling-info\n 200 OK\n {\n     \"distances\": {\n         \"0\": \"A compute node\",\n         \"1\": \"A single rack of compute nodes\",\n         \"2\": \"A pair of racks connected to a single power-supply\",\n         \"3\": \"A row of compute racks\",\n         \"4\": \"A datacenter\"\n }\n\nor something like that, to allow the operator/deployer to define what levels (if any) of proximity/distance that they support.","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":8768,"name":"Chris Friesen","email":"chris.friesen@windriver.com","username":"cbf123"},"change_message_id":"49bf39f589c858ad1c559ceb453e178c5ea49695","unresolved":false,"context_lines":[{"line_number":138,"context_line":"2. Since there is no more need to maintain a separate server group object, if a"},{"line_number":139,"context_line":"   user launched 3 instances and then wanted to make sure that 3 new instances"},{"line_number":140,"context_line":"   don\u0027t end up on the same hosts, all the user needs to do is tag the existing"},{"line_number":141,"context_line":"   instances with a tag, and issue a call to:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":"    nova boot --not-near-tag $TAG ..."},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_5dbc8ddb","line":141,"updated":"2015-05-21 07:12:39.000000000","message":"what would be the REST API for tagging/untagging existing instances?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"0773a8870d21e98848951b5cf139a6ddcc54df6e","unresolved":false,"context_lines":[{"line_number":138,"context_line":"2. Since there is no more need to maintain a separate server group object, if a"},{"line_number":139,"context_line":"   user launched 3 instances and then wanted to make sure that 3 new instances"},{"line_number":140,"context_line":"   don\u0027t end up on the same hosts, all the user needs to do is tag the existing"},{"line_number":141,"context_line":"   instances with a tag, and issue a call to:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":"    nova boot --not-near-tag $TAG ..."},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_de02a209","line":141,"in_reply_to":"7a016987_5dbc8ddb","updated":"2015-05-21 22:27:48.000000000","message":"Are you asking what the behaviour would be if you tagged an instance and the operation of tagging that instance resulted in a violation of the scheduling policy that the instance was initially launched under?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"a56a9f9dff43e8797d94a89e4b369a8af83cc82c","unresolved":false,"context_lines":[{"line_number":148,"context_line":"   untag a server to remove it from the set of servers you wish to use in"},{"line_number":149,"context_line":"   applying some affinity policy"},{"line_number":150,"context_line":""},{"line_number":151,"context_line":"4. Likewise, since there\u0027s no server group object, in order to relate an"},{"line_number":152,"context_line":"   existing server to another is to simply place a tag on the server."},{"line_number":153,"context_line":""},{"line_number":154,"context_line":"5. The act of applying affinity policies is now directly related to the act of"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_8f53e8e1","line":151,"updated":"2015-05-21 08:09:55.000000000","message":"Does it check the validity of the rule as part of this operation?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"2b3c2d25eeab7d327912e66a0a8e6c856aa39ef3","unresolved":false,"context_lines":[{"line_number":148,"context_line":"   untag a server to remove it from the set of servers you wish to use in"},{"line_number":149,"context_line":"   applying some affinity policy"},{"line_number":150,"context_line":""},{"line_number":151,"context_line":"4. Likewise, since there\u0027s no server group object, in order to relate an"},{"line_number":152,"context_line":"   existing server to another is to simply place a tag on the server."},{"line_number":153,"context_line":""},{"line_number":154,"context_line":"5. The act of applying affinity policies is now directly related to the act of"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_bec246e7","line":151,"in_reply_to":"7a016987_5e17d247","updated":"2015-05-22 00:29:01.000000000","message":"Correct.  We tried to address this in the SG model with some \u0027smarts\u0027 applied to the proposals to add/remove VMs to/from SGs such that it would block the request if whatever action you were about to make would violate the affinity/anti-affinity policy.","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"0773a8870d21e98848951b5cf139a6ddcc54df6e","unresolved":false,"context_lines":[{"line_number":148,"context_line":"   untag a server to remove it from the set of servers you wish to use in"},{"line_number":149,"context_line":"   applying some affinity policy"},{"line_number":150,"context_line":""},{"line_number":151,"context_line":"4. Likewise, since there\u0027s no server group object, in order to relate an"},{"line_number":152,"context_line":"   existing server to another is to simply place a tag on the server."},{"line_number":153,"context_line":""},{"line_number":154,"context_line":"5. The act of applying affinity policies is now directly related to the act of"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_5e17d247","line":151,"in_reply_to":"7a016987_8f53e8e1","updated":"2015-05-21 22:27:48.000000000","message":"Are you asking what the behaviour would be if you tagged an instance and the operation of tagging that instance resulted in a violation of the scheduling policy that the instance was initially launched under?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"a56a9f9dff43e8797d94a89e4b369a8af83cc82c","unresolved":false,"context_lines":[{"line_number":178,"context_line":"REST API impact"},{"line_number":179,"context_line":"---------------"},{"line_number":180,"context_line":""},{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_fd14594d","line":181,"updated":"2015-05-21 08:09:55.000000000","message":"When (if ever) would it be removed?  Or would it just remain present, albeit not be enhanced/supported?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"0773a8870d21e98848951b5cf139a6ddcc54df6e","unresolved":false,"context_lines":[{"line_number":178,"context_line":"REST API impact"},{"line_number":179,"context_line":"---------------"},{"line_number":180,"context_line":""},{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_5e3eb2c1","line":181,"in_reply_to":"7a016987_fd14594d","updated":"2015-05-21 22:27:48.000000000","message":"I\u0027d love for it to be removed eventually, but DEPRECATED would technically mean the code paths would continue to be supported for the old server groups code, but that code would not get any new enhancements to it, and whenver API_MIN_VERSION was incremented above the microversion that the new generic scheduling policies were introduced, then the old server groups code could be removed.","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"a56a9f9dff43e8797d94a89e4b369a8af83cc82c","unresolved":false,"context_lines":[{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"},{"line_number":185,"context_line":"the `near_tag`, `not_near_tags`, `absolute`, and `distance` fields. The new"},{"line_number":186,"context_line":"microversion should validate incoming request body data using the following"},{"line_number":187,"context_line":"JSON-Schema document::"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_8f32c8f0","line":184,"updated":"2015-05-21 08:09:55.000000000","message":"Will there be a way to update this data post boot as well as that was called out as some of the problems we\u0027re trying to address?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"2b3c2d25eeab7d327912e66a0a8e6c856aa39ef3","unresolved":false,"context_lines":[{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"},{"line_number":185,"context_line":"the `near_tag`, `not_near_tags`, `absolute`, and `distance` fields. The new"},{"line_number":186,"context_line":"microversion should validate incoming request body data using the following"},{"line_number":187,"context_line":"JSON-Schema document::"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_9e11ca5f","line":184,"in_reply_to":"7a016987_1e4b9a5a","updated":"2015-05-22 00:29:01.000000000","message":"Is there any way for me to declare that I no longer \"care\" about my affinity/anti-affinity/proximity to other VMs then--without having to update the tags of the other VMs?  A use case would be something like \"When I originally deployed this VM, I wanted to be close to this other VM for network reasons... but I\u0027ve since upgraded my network infrastructure and I no longer care due to the added bandwidth.\"","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":8768,"name":"Chris Friesen","email":"chris.friesen@windriver.com","username":"cbf123"},"change_message_id":"7845ea7b4fbd81822efef6a0e9824664a6a2f616","unresolved":false,"context_lines":[{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"},{"line_number":185,"context_line":"the `near_tag`, `not_near_tags`, `absolute`, and `distance` fields. The new"},{"line_number":186,"context_line":"microversion should validate incoming request body data using the following"},{"line_number":187,"context_line":"JSON-Schema document::"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_6f953331","line":184,"in_reply_to":"7a016987_1e4b9a5a","updated":"2015-05-22 07:06:16.000000000","message":"Wait...I had assumed that by \"tagging\" you meant \"associating a \u0027placement\u0027 with an existing instance\", and by \"untagging\" you meant \"disassociating a \u0027placement\u0027 from an instance\".\n\nI literally just now did some more digging and realized that this spec builds on top of https://review.openstack.org/#/c/177112/7/specs/liberty/approved/tag-instances.rst\n\nIt might make sense to make that explicit early on in this document.\n\nI think that it would make sense to be able to add/remove \"placement\" information to existing instances such that it would take effect when doing a subsequent resize/rebuild/migrate/evac/etc","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"0773a8870d21e98848951b5cf139a6ddcc54df6e","unresolved":false,"context_lines":[{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"},{"line_number":185,"context_line":"the `near_tag`, `not_near_tags`, `absolute`, and `distance` fields. The new"},{"line_number":186,"context_line":"microversion should validate incoming request body data using the following"},{"line_number":187,"context_line":"JSON-Schema document::"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_1e4b9a5a","line":184,"in_reply_to":"7a016987_8f32c8f0","updated":"2015-05-21 22:27:48.000000000","message":"I wasn\u0027t planning on adding a way to update the scheduling policy after an instance is launched, no. I do, however, need to define the behaviour for when an instance is tagged/untagged and that tagging would violate the policy that the instance was originally scheduled under.","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":8768,"name":"Chris Friesen","email":"chris.friesen@windriver.com","username":"cbf123"},"change_message_id":"7845ea7b4fbd81822efef6a0e9824664a6a2f616","unresolved":false,"context_lines":[{"line_number":181,"context_line":"The existing `os-server-groups` API extension will be DEPRECATED."},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"A new API microversion will change the Controller for the servers resource\u0027s"},{"line_number":184,"context_line":"`POST` command to allow the `placement` object in the request body, along with"},{"line_number":185,"context_line":"the `near_tag`, `not_near_tags`, `absolute`, and `distance` fields. The new"},{"line_number":186,"context_line":"microversion should validate incoming request body data using the following"},{"line_number":187,"context_line":"JSON-Schema document::"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_6fbc13bb","line":184,"in_reply_to":"7a016987_9e11ca5f","updated":"2015-05-22 07:06:16.000000000","message":"Would simply removing the \"placement\" from the instance in question satisfy this usecase?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"a56a9f9dff43e8797d94a89e4b369a8af83cc82c","unresolved":false,"context_lines":[{"line_number":223,"context_line":"---------------------"},{"line_number":224,"context_line":""},{"line_number":225,"context_line":"novaclient should add new CLI options for controlling the injection of the"},{"line_number":226,"context_line":"placement preferences in the launch instance request::"},{"line_number":227,"context_line":""},{"line_number":228,"context_line":"    --place-near-tag \u003cSTRING\u003e"},{"line_number":229,"context_line":"    --place-not-near-tags \u003cLIST_OF_STRINGS\u003e"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_0f55b8f8","line":226,"updated":"2015-05-21 08:09:55.000000000","message":"Ditto on the \u0027update\u0027 comment above... i.e., to update all of this information once the VM has been booted?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":8768,"name":"Chris Friesen","email":"chris.friesen@windriver.com","username":"cbf123"},"change_message_id":"7845ea7b4fbd81822efef6a0e9824664a6a2f616","unresolved":false,"context_lines":[{"line_number":223,"context_line":"---------------------"},{"line_number":224,"context_line":""},{"line_number":225,"context_line":"novaclient should add new CLI options for controlling the injection of the"},{"line_number":226,"context_line":"placement preferences in the launch instance request::"},{"line_number":227,"context_line":""},{"line_number":228,"context_line":"    --place-near-tag \u003cSTRING\u003e"},{"line_number":229,"context_line":"    --place-not-near-tags \u003cLIST_OF_STRINGS\u003e"}],"source_content_type":"text/x-rst","patch_set":2,"id":"7a016987_6f31b359","line":226,"in_reply_to":"7a016987_0f55b8f8","updated":"2015-05-22 07:06:16.000000000","message":"Presumably if we were going to support adding/removing \"placement\" information then we would need novaclient commands for that.\n\nAlso, I\u0027m not thrilled about each thing having a separate argument...could we unify these under a single \"--placement\" arg or something?\n\nLastly, if someone specifies both \"near\" and \"not near\" simultaneously, does the specified distance apply to both equally?","commit_id":"63fc0439f8828e4dcf5f6050ea621e58268559d9"},{"author":{"_account_id":9452,"name":"James Penick","email":"penick@verizonmedia.com","username":"epim"},"change_message_id":"4a59ca28f49a7fd9c5484cc2fa69a1a417ea0ac0","unresolved":false,"context_lines":[{"line_number":125,"context_line":"   in different data centers entirely. The deployer could advertise that the"},{"line_number":126,"context_line":"   numbers 0-3 represent those increasing degrees of relative \"distance\" that"},{"line_number":127,"context_line":"   instances would be placed from each other."},{"line_number":128,"context_line":""},{"line_number":129,"context_line":"How does this solve the issues highlighted above? In order, it solves the"},{"line_number":130,"context_line":"issues like so:"},{"line_number":131,"context_line":""}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_91e1a22c","line":128,"updated":"2015-05-22 22:03:00.000000000","message":"Because i\u0027m mean and like to complicate things..\n\nI think this is a lot better than server groups. It does share one limitation,\nhowever: not_near_tags implies that no more than one instance will be placed\nwithin the physical constraints modeled by the tag. However we (yahoo) have\na use case that says \u003c\u003d N instances can be booted within a rack.\n\nTo illustrate my use case:\n  PD1        PD2        PD3\n / | \\      / | \\      / | \\\nR1 R2 R3   R4 R5 R6   R7 R8 R9\n Bob needs to boot 40 instances. He cannot have more than 10 instances in one\npower domain, or more than 5 instances in a single rack within that power\ndomain.\n Without the ability to indicate that up to a certain number of instances\ncould be booted within a rack, AND up to a certain number of instances within\na PDU, Bob\u0027s going to have to make a  lot of separate API calls to boot\ninstances.\n I realize you\u0027re not trying to model physical datacenter elements within this\nspec, but whatever does model those will rely on this.","commit_id":"51c7b27355ee10fddc425f06242d4e4fea0751e7"},{"author":{"_account_id":7166,"name":"Sylvain Bauza","email":"sbauza@redhat.com","username":"sbauza"},"change_message_id":"e329489a7d6f77562e66fdcc3b86251ff4bb607b","unresolved":false,"context_lines":[{"line_number":31,"context_line":"and then, when Alice wishes to launch an instance and make sure it doesn\u0027t land"},{"line_number":32,"context_line":"on a host with another of that instance type, she does the following:"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"nova boot --group $GROUP_UUID ..."},{"line_number":35,"context_line":""},{"line_number":36,"context_line":"There are myriad problems with the above user experience and"},{"line_number":37,"context_line":"implementation. Let me explain them."}],"source_content_type":"text/x-rst","patch_set":4,"id":"fa32b979_eb61e112","line":34,"updated":"2015-06-19 09:56:21.000000000","message":"nit: technically, that\u0027s --hint group\u003d$GROUP_UUID","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7166,"name":"Sylvain Bauza","email":"sbauza@redhat.com","username":"sbauza"},"change_message_id":"e329489a7d6f77562e66fdcc3b86251ff4bb607b","unresolved":false,"context_lines":[{"line_number":38,"context_line":""},{"line_number":39,"context_line":"1. The user isn\u0027t creating a \"server group\" when they issue a nova"},{"line_number":40,"context_line":"   server-group-create call. They are creating a *policy* and *calling it a"},{"line_number":41,"context_line":"   group*. Cognitive dissonance results from this mismatch."},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"2. There\u0027s no way to add an existing server to this \"group\". What this means is"},{"line_number":44,"context_line":"   that the user needs to effectively have pre-considered their environment and"}],"source_content_type":"text/x-rst","patch_set":4,"id":"fa32b979_6b0b912c","line":41,"updated":"2015-06-19 09:56:21.000000000","message":"Well, they just create a logical object, which is technically a collection of instances. Attaching a policy to that object is for the moment the only metadata we allow (and the only usage) but that doesn\u0027t prevent to see a server group for any other usage requiring a collection of instances. I\u0027m just saying that for explaining that either we consider Server Groups as just a pure hack for affinity stuff, and then +1 with your bullet point; or we keep the possibility to group those instances for a future usage, and then it *could* be a reason to keep them.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"d0030aded2e28919ccd27b24690c04b36d02871e","unresolved":false,"context_lines":[{"line_number":50,"context_line":"   don\u0027t end up on the same hosts as the others"},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"No can do, since the first three VMs weren\u0027t started using a --group"},{"line_number":53,"context_line":"scheduler hint."},{"line_number":54,"context_line":""},{"line_number":55,"context_line":"3. There\u0027s no way to remove members from the group"},{"line_number":56,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"1a4dcd0f_f12e1e8b","line":53,"updated":"2015-08-11 18:55:57.000000000","message":"It\u0027s possible I missed it but it didn\u0027t appear this is actually addressed in this proposal either? It\u0027s not one of the use cases discussed further on and all examples focus on scheduling a new instance.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"d0030aded2e28919ccd27b24690c04b36d02871e","unresolved":false,"context_lines":[{"line_number":53,"context_line":"scheduler hint."},{"line_number":54,"context_line":""},{"line_number":55,"context_line":"3. There\u0027s no way to remove members from the group"},{"line_number":56,"context_line":""},{"line_number":57,"context_line":"4. There\u0027s no way to manually add members to the server group"},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"5. The act of telling the scheduler to place instances near or away from some"}],"source_content_type":"text/x-rst","patch_set":4,"id":"1a4dcd0f_b1889672","line":56,"updated":"2015-08-11 18:55:57.000000000","message":"Well yeah, but IIRC remove was going to be fairly simple to implement but was n\u0027acked on account of the fact that providing add was non-trivial and it was desired to only add one if we could also do the other. If this spec doesn\u0027t address applying the policy an already running instance then it\u0027s not clear to me this objection would go away?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7166,"name":"Sylvain Bauza","email":"sbauza@redhat.com","username":"sbauza"},"change_message_id":"e329489a7d6f77562e66fdcc3b86251ff4bb607b","unresolved":false,"context_lines":[{"line_number":54,"context_line":""},{"line_number":55,"context_line":"3. There\u0027s no way to remove members from the group"},{"line_number":56,"context_line":""},{"line_number":57,"context_line":"4. There\u0027s no way to manually add members to the server group"},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"5. The act of telling the scheduler to place instances near or away from some"},{"line_number":60,"context_line":"   other instances has been hidden behind the server group API, which means"}],"source_content_type":"text/x-rst","patch_set":4,"id":"fa32b979_d65b64f5","line":57,"updated":"2015-06-19 09:56:21.000000000","message":"Isn\u0027t that a paraphrase of #2 ?\nI mean, you mentioned that if you spawn VMs without specifying a group, you can\u0027t add an existing server to the group. That sounds really identical to \u0027adding members to a group\u0027, right?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7166,"name":"Sylvain Bauza","email":"sbauza@redhat.com","username":"sbauza"},"change_message_id":"e329489a7d6f77562e66fdcc3b86251ff4bb607b","unresolved":false,"context_lines":[{"line_number":59,"context_line":"5. The act of telling the scheduler to place instances near or away from some"},{"line_number":60,"context_line":"   other instances has been hidden behind the server group API, which means"},{"line_number":61,"context_line":"   that users doing a nova help boot will see a --group option that doesn\u0027t"},{"line_number":62,"context_line":"   make much sense, as it doesn\u0027t describe the scheduling policy activity."},{"line_number":63,"context_line":""},{"line_number":64,"context_line":"Use Cases"},{"line_number":65,"context_line":"----------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"fa32b979_f62a282f","line":62,"updated":"2015-06-19 09:56:21.000000000","message":"That\u0027s even worst. Since there is no --group option (this is passed as a hint), it\u0027s totally obfuscate because it even doesn\u0027t propose that notion (and leaves the user needing to know if the AffinityFilter is enabled or not)","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"f28a234c871b15cea3d1a22b140dba9810d94836","unresolved":false,"context_lines":[{"line_number":61,"context_line":"   that users doing a nova help boot will see a --group option that doesn\u0027t"},{"line_number":62,"context_line":"   make much sense, as it doesn\u0027t describe the scheduling policy activity."},{"line_number":63,"context_line":""},{"line_number":64,"context_line":"Use Cases"},{"line_number":65,"context_line":"----------"},{"line_number":66,"context_line":""},{"line_number":67,"context_line":"There are a number of use cases that this proposed generic scheduling policy"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_785d6d78","line":64,"updated":"2015-05-28 00:23:24.000000000","message":"What about a use case for a user/admin to actually see the groups/relationships of the VMs?\n\nConsider a cloud with thousands of VMs and the user just wants to see which VMs need to be collocated or anti-collocated?\n\nThis was actually one of the pros of server groups--you had a nice way of seeing which VMs were \"related\" without necessarily needing to iterate over all instances in the cloud.\n\nI think this is something we should also consider as part of the new proposal so that user-friendly interfaces can be created?\n\nPerhaps this can be achieved with some convenience methods on the \"tags\" API (also assuming the tags would be indexed for faster look-ups)?  If we went with this approach, it would be nice to have an interface that could query for tags that were *also* used for placement decisions, so that you could weed out tags used for less rigid/structured things?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"f28a234c871b15cea3d1a22b140dba9810d94836","unresolved":false,"context_lines":[{"line_number":64,"context_line":"Use Cases"},{"line_number":65,"context_line":"----------"},{"line_number":66,"context_line":""},{"line_number":67,"context_line":"There are a number of use cases that this proposed generic scheduling policy"},{"line_number":68,"context_line":"spec attempts to serve. In this spec, we will refer to the below use cases by"},{"line_number":69,"context_line":"number when showing how the end solution would look like."},{"line_number":70,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_582631cf","line":67,"updated":"2015-05-28 00:23:24.000000000","message":"A few more thoughts:\n\n1. What about a use case flow that talks about updating the instance such that the affinity/anti-affinity rule(s) should no longer apply?\n\nI bring this up because I think you\u0027d have to have a way to not only remove the tag from the VM, but also a subset (or all) of the \u0027placements\u0027 section from the VM.\n\n2. How do we define the VM\u0027s \u0027placements\u0027 post deploy?  The use case here would be to apply affinity/anti-affinity after the VM\u0027s been booted into the cloud.  I don\u0027t think just tagging the VM is sufficient since it wouldn\u0027t have symmetrical rules?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7166,"name":"Sylvain Bauza","email":"sbauza@redhat.com","username":"sbauza"},"change_message_id":"e329489a7d6f77562e66fdcc3b86251ff4bb607b","unresolved":false,"context_lines":[{"line_number":106,"context_line":"as possible. The definition of distance between instances, from furthest to"},{"line_number":107,"context_line":"nearest, is \"same cabinet\", \"same rack\", \"same host\". If the furthest away"},{"line_number":108,"context_line":"is a compute host within the same rack as other instances, the request should"},{"line_number":109,"context_line":"fail."},{"line_number":110,"context_line":""},{"line_number":111,"context_line":"Project Priority"},{"line_number":112,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"fa32b979_b68020f9","line":109,"updated":"2015-06-19 09:56:21.000000000","message":"That reminds me a specific problem I saw when I reviewed. Sometimes, you have a soft request for placement \"it\u0027s nice to have that affinity\", but in that case, having a placement failure because there was not enough capacity is struggling.\n\nConsidering host-\u003erack-\u003ecabinet proximity with a sense of \u0027do your possible\u0027 sounds definitely needed.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"d0030aded2e28919ccd27b24690c04b36d02871e","unresolved":false,"context_lines":[{"line_number":107,"context_line":"nearest, is \"same cabinet\", \"same rack\", \"same host\". If the furthest away"},{"line_number":108,"context_line":"is a compute host within the same rack as other instances, the request should"},{"line_number":109,"context_line":"fail."},{"line_number":110,"context_line":""},{"line_number":111,"context_line":"Project Priority"},{"line_number":112,"context_line":"-----------------"},{"line_number":113,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"1a4dcd0f_f1fdbeb1","line":110,"updated":"2015-08-11 18:55:57.000000000","message":"Another is N+M groups. That is I have group N and group M. Instances in group N have an affinity policy towards each other, instances in group M have an affinity policy towards each other. Group N has an anti-affinity policy to group M.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9452,"name":"James Penick","email":"penick@verizonmedia.com","username":"epim"},"change_message_id":"d3eb009acd4a9e6312f8ea135663454d328c2b8a","unresolved":false,"context_lines":[{"line_number":107,"context_line":"nearest, is \"same cabinet\", \"same rack\", \"same host\". If the furthest away"},{"line_number":108,"context_line":"is a compute host within the same rack as other instances, the request should"},{"line_number":109,"context_line":"fail."},{"line_number":110,"context_line":""},{"line_number":111,"context_line":"Project Priority"},{"line_number":112,"context_line":"-----------------"},{"line_number":113,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"5afe65bd_1a1a496f","line":110,"updated":"2015-06-01 18:11:42.000000000","message":"Another use case: \n I want to build 20 instances and ensure that they\u0027re spread across (at least) 4 racks in (at least) 2 power domains.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9452,"name":"James Penick","email":"penick@verizonmedia.com","username":"epim"},"change_message_id":"d3eb009acd4a9e6312f8ea135663454d328c2b8a","unresolved":false,"context_lines":[{"line_number":144,"context_line":"            \u0027near_tag\u0027: {"},{"line_number":145,"context_line":"                \u0027type\u0027: \u0027string\u0027"},{"line_number":146,"context_line":"            },"},{"line_number":147,"context_line":"            \u0027not_near_tags\u0027: {"},{"line_number":148,"context_line":"                \u0027type\u0027: \u0027array\u0027,"},{"line_number":149,"context_line":"                \u0027items\u0027: \u0027string\u0027"},{"line_number":150,"context_line":"            },"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5afe65bd_5a10c14e","line":147,"updated":"2015-06-01 18:11:42.000000000","message":"Per the use case I mentioned above, what about a threshold value which controls the maximum number of instances that can schedule within a single compute node/rack/power domain.\n\nDefaults to 1, but if I want to allow up to 5 instances per rack, I can set the value accordingly.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"2828f486d8db4d8d408aa31eb2a74d89b71e3fcf","unresolved":false,"context_lines":[{"line_number":141,"context_line":"                \u0027default\u0027: \u0027compute\u0027,"},{"line_number":142,"context_line":"                \u0027oneOf\u0027: [\u0027compute\u0027]"},{"line_number":143,"context_line":"            },"},{"line_number":144,"context_line":"            \u0027near_tag\u0027: {"},{"line_number":145,"context_line":"                \u0027type\u0027: \u0027string\u0027"},{"line_number":146,"context_line":"            },"},{"line_number":147,"context_line":"            \u0027not_near_tags\u0027: {"},{"line_number":148,"context_line":"                \u0027type\u0027: \u0027array\u0027,"},{"line_number":149,"context_line":"                \u0027items\u0027: \u0027string\u0027"},{"line_number":150,"context_line":"            },"},{"line_number":151,"context_line":"            \u0027absolute\u0027: {"},{"line_number":152,"context_line":"                \u0027type\u0027: \u0027bool\u0027,"},{"line_number":153,"context_line":"                \u0027default\u0027: true"},{"line_number":154,"context_line":"            },"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_56e027cc","line":151,"range":{"start_line":144,"start_character":0,"end_line":151,"end_character":0},"updated":"2015-05-29 09:06:35.000000000","message":"It is a bit confusing to me to be able to provide multiple not_near_tags but a single near_tag in a placement object. As we allow multiple placement objects in a single request we can basically allow to have multiple placement objects with not_near_tags and also multiple placement objects with near_tag in a single request. Therefore I don\u0027t see the reason why we have different cardinality of the different tags in a single placement object as we anyhow need to handle all the possibilities.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":11564,"name":"Chris Dent","email":"cdent@anticdent.org","username":"chdent"},"change_message_id":"b48ced9ba6756a213948b1267c3c4dab1f7254e6","unresolved":false,"context_lines":[{"line_number":175,"context_line":"   `dimension` in the request spec near other instances having a tag."},{"line_number":176,"context_line":""},{"line_number":177,"context_line":" * `not_near_tags`: indicates that the scheduler should place all instances in"},{"line_number":178,"context_line":"   the request spec *not* near instances having any of a list of tags."},{"line_number":179,"context_line":""},{"line_number":180,"context_line":" * `absolute`: (defaults to true) If false, indicates to the scheduler that if"},{"line_number":181,"context_line":"   the scheduler cannot fulfill the preferred placement in absolute terms, that"}],"source_content_type":"text/x-rst","patch_set":4,"id":"bab6814e_3a94bae0","line":178,"updated":"2016-05-25 14:19:16.000000000","message":"Now that I\u0027ve had more time to think about it, I\u0027m more convinced than ever that there\u0027s something fishy about this near_tag and not_near_tags stuff. I understand what it is trying to express and that that that kind of expression needs to be made, but I think it would be bad news for the information to be used as it is structure here.\n\nThe implication is that while doing some loop we\u0027ll try to resolve nearness in section \u0027a\u0027 of the code and then resolve not_nearness in section \u0027b\u0027 of the code. It would be better (for a longevity and flexibility standpoint) to be able to keep it together in the code, and if we\u0027re going to do that, the external representation can probably be more tidy. Rather than a singular distance, each tag gets a distance on a scale -9 to 9 and we solve for optimal solution on the several dimensions named by the tags.\n\nIn fact, the use of the term \u0027dimension\u0027 to describe whether we are in compute, network or storage _space_ seems a bit confusing to me. As I\u0027m understanding things what we\u0027re trying to describe is a location in the world of \"compute\" where we are far from tag a, close to tag b, and mostly indifferent but hey it would nice to be close to tag c.\n\nOh, and of course, this all begs the question of why is near a string and not near a list?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"f28a234c871b15cea3d1a22b140dba9810d94836","unresolved":false,"context_lines":[{"line_number":177,"context_line":" * `not_near_tags`: indicates that the scheduler should place all instances in"},{"line_number":178,"context_line":"   the request spec *not* near instances having any of a list of tags."},{"line_number":179,"context_line":""},{"line_number":180,"context_line":" * `absolute`: (defaults to true) If false, indicates to the scheduler that if"},{"line_number":181,"context_line":"   the scheduler cannot fulfill the preferred placement in absolute terms, that"},{"line_number":182,"context_line":"   the scheduler should attempt a best-effort placement using less-rigid"},{"line_number":183,"context_line":"   placement policies. This is sometimes called \"soft (anti-)affinity\"."}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_d8bc616e","line":180,"updated":"2015-05-28 00:23:24.000000000","message":"Love it!","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"2828f486d8db4d8d408aa31eb2a74d89b71e3fcf","unresolved":false,"context_lines":[{"line_number":272,"context_line":"This request will place instance B on any compute host except compute host 1,"},{"line_number":273,"context_line":"since instance A, also tagged with \u0027foo\u0027, landed on compute host 1."},{"line_number":274,"context_line":""},{"line_number":275,"context_line":"If there were no other compute hosts in rack 1, then the request would fail to"},{"line_number":276,"context_line":"place the new instance."},{"line_number":277,"context_line":""},{"line_number":278,"context_line":"Use Case #3. Same-rack instance placement, not on same compute host"},{"line_number":279,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_968b3f68","line":276,"range":{"start_line":275,"start_character":0,"end_line":276,"end_character":23},"updated":"2015-05-29 09:06:35.000000000","message":"What will happen if there is a proper compute host in rack 2?  For me the answer would be to allow selecting host from rack2 .","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"2828f486d8db4d8d408aa31eb2a74d89b71e3fcf","unresolved":false,"context_lines":[{"line_number":385,"context_line":"This request will place instance A somewhere in the cloud and tag it with the"},{"line_number":386,"context_line":"string \u0027foo\u0027. Placement is indeterminate, since no other instances exist before"},{"line_number":387,"context_line":"this initial request, so we will assume that Instance A is placed on compute"},{"line_number":388,"context_line":"host 1 in rack 1. The placement policy is stored to the instance, along with"},{"line_number":389,"context_line":"the tag \u0027foo\u0027."},{"line_number":390,"context_line":""},{"line_number":391,"context_line":"Instance B is launched using the following::"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_3fcd6285","line":388,"updated":"2015-05-29 09:06:35.000000000","message":"Do you mean host1 rack1 in cabinet1?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"2828f486d8db4d8d408aa31eb2a74d89b71e3fcf","unresolved":false,"context_lines":[{"line_number":403,"context_line":""},{"line_number":404,"context_line":"Let us assume that the datacenter in which instance A was placed has 4"},{"line_number":405,"context_line":"cabinets, each containing 4 racks of compute hosts."},{"line_number":406,"context_line":""},{"line_number":407,"context_line":"This request will place instance B on any compute host in rack 4 of cabinet 4,"},{"line_number":408,"context_line":"which is logically the \"furthest away\" from rack 1 in cabinet 1 of the same"},{"line_number":409,"context_line":"datacenter."}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_ff4fca31","line":406,"updated":"2015-05-29 09:06:35.000000000","message":"We have to assume that we have only a single datacenter as distance \u003d 3 here means we want to put the Instance B to a different datacenter if possible than Instance A.\n\nOr did I misunderstood the meaning of distance \u003d 3?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"2828f486d8db4d8d408aa31eb2a74d89b71e3fcf","unresolved":false,"context_lines":[{"line_number":411,"context_line":"If all compute hosts in rack 4 of cabinet 4 were full, then the new instance"},{"line_number":412,"context_line":"would be placed in a rack 3 in the cabinet 4, since \u0027absolute\u0027 is set to false,"},{"line_number":413,"context_line":"indicating a \"soft\" affinity policy."},{"line_number":414,"context_line":""},{"line_number":415,"context_line":"Use Case #6. As far away from other instances as possible, with limits"},{"line_number":416,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":417,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_bf0a7288","line":414,"updated":"2015-05-29 09:06:35.000000000","message":"I feel the definition of distance is not clear enough. First I thought that the distance defines how far an entity from another entity in the host-rack-cabinet-datacenter hierarchy. But here it seems that the distance can be different for entities on the same hierarchy level.\nI don\u0027t say it is bad, but I think we need a clearer definition of distance above.\n\nSo the my fist simple definition of distance is the following:\n\nh1,h2 are hosts, r1,r2 are racks, c1,c2 are cabinets, h1 in r1 and h2 in r2, r1 in c1 and r2 in c2\n\nd(h1, h2) :\u003d 0 iff h1 \u003d\u003d h2 and r1 \u003d\u003d r2 and c1 \u003d\u003d c2\n\nd(h1, h2) :\u003d 1 iff h1 !\u003d h2 and r1 \u003d\u003d r2 and c1 \u003d\u003d c2\n\nd(h1, h2) :\u003d 2 iff r1 !\u003d r2 and c1 \u003d\u003d c2\n\nd(h1, h2) :\u003d 3 iff c1 !\u003d c2\n\netc...\n\nSo the distance is the number of levels in hierarchy where the two entities differ. For this definition scheduler only needs to understand if two hosts are the same or not, two racks are the same or not, etc...\n\nIn the other hand if we want to define distance between h1 and h2 based on the location of h1 and h2 in the same rack too then the definition of distance becomes more complex and the scheduler has to know not only if the host are the same or not but also the distance between every host pairs.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":11564,"name":"Chris Dent","email":"cdent@anticdent.org","username":"chdent"},"change_message_id":"b48ced9ba6756a213948b1267c3c4dab1f7254e6","unresolved":false,"context_lines":[{"line_number":436,"context_line":"string \u0027foo\u0027. Placement is indeterminate, since no other instances exist before"},{"line_number":437,"context_line":"this initial request, so we will assume that Instance A is placed on compute"},{"line_number":438,"context_line":"host 1 in rack 1. The placement policy is stored to the instance, along with"},{"line_number":439,"context_line":"the tag \u0027foo\u0027."},{"line_number":440,"context_line":""},{"line_number":441,"context_line":"Instance B is launched using the following::"},{"line_number":442,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"bab6814e_354a0bc6","line":439,"updated":"2016-05-25 14:19:16.000000000","message":"In the abstract: Persisting this additional data (beyond the tag) feels heavyweight. If we have to do it fine, but it would be better if a placement policy is ephemeral and we don\u0027t need to refer to it again (after a successful placement).\n\nIs the point that an instance needs to be able to say \"hey, once I\u0027ve landed, don\u0027t put stuff near me because I\u0027m _special_\"? If so, man, that\u0027s unfortunate. So much persisted state and also: when/how is it being used (as bauzas appears to be asking below)?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7166,"name":"Sylvain Bauza","email":"sbauza@redhat.com","username":"sbauza"},"change_message_id":"e329489a7d6f77562e66fdcc3b86251ff4bb607b","unresolved":false,"context_lines":[{"line_number":467,"context_line":"indicating a \"soft\" affinity policy."},{"line_number":468,"context_line":""},{"line_number":469,"context_line":"However, if all compute hosts in all racks in all cabinets other than the rack"},{"line_number":470,"context_line":"1 of cabinet 1 were full, the request would fail."},{"line_number":471,"context_line":""},{"line_number":472,"context_line":"Alternatives"},{"line_number":473,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"fa32b979_562bd494","line":470,"updated":"2015-06-19 09:56:21.000000000","message":"I tried to read all the proposal, but I don\u0027t see how you persist that request-specific information and access it later ?\n\nWhile I like the API, I\u0027m definitely interested in knowing how you plan to model it and access it. IMHO, that\u0027s part of a RequestSpec information which would be related to instances...","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"2828f486d8db4d8d408aa31eb2a74d89b71e3fcf","unresolved":false,"context_lines":[{"line_number":468,"context_line":""},{"line_number":469,"context_line":"However, if all compute hosts in all racks in all cabinets other than the rack"},{"line_number":470,"context_line":"1 of cabinet 1 were full, the request would fail."},{"line_number":471,"context_line":""},{"line_number":472,"context_line":"Alternatives"},{"line_number":473,"context_line":"------------"},{"line_number":474,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_5a8b8cc3","line":471,"updated":"2015-05-29 09:06:35.000000000","message":"We still have problem with the symmetry of the relationship. In every example you assume that the user planned the relationship ahead of time and booted Instance A already with the proper placement objects and tags. This is the same requirement as creating a server-group proactively.\n\nAlso the current proposal does not handle if the user didn\u0027t add placement object (or tag) to Instance A at boot time but realize that he needs to enforce some placement when booting Instance B. So basically we have the same problem as you stated against server-groups: \"4. There\u0027s no way to manually add members to the server group\".\n\nI know that tags can be added retroactively to an instance, but we either need to make sure that placement objects can also be added later or we have to implement automation in nova that adds proper placement objects to Instance A if Instance B booted with a placement object referring to tag Instance A has. If we choose the automation then we have to make sure that placement objects are inserted also when a tag is added to an instance and another instance has placement objects referring to that tag.\n\nAnyhow It seems keeping the placement objects of different instances in sync is quite hard either for the user or for nova in case of automation. Here I think server-group has an advantage that it abstracts some of the complexity out by defining the group concept.","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"f28a234c871b15cea3d1a22b140dba9810d94836","unresolved":false,"context_lines":[{"line_number":529,"context_line":""},{"line_number":530,"context_line":"None"},{"line_number":531,"context_line":""},{"line_number":532,"context_line":"Notifications impact"},{"line_number":533,"context_line":"--------------------"},{"line_number":534,"context_line":""},{"line_number":535,"context_line":"None"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_3808152d","line":532,"updated":"2015-05-28 00:23:24.000000000","message":"Assume the tag/un-tag operations would generate notifications?  i.e., so one can tell when an instance is being marked with tag \u0027foobar\u0027 ?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"},{"author":{"_account_id":7664,"name":"Joe Cropper","email":"jwcroppe@us.ibm.com","username":"jwcroppe"},"change_message_id":"f28a234c871b15cea3d1a22b140dba9810d94836","unresolved":false,"context_lines":[{"line_number":552,"context_line":"increase due to a simpler format that does not need to cross more than a"},{"line_number":553,"context_line":"single table (`tags`) for the non-distance-accounting proximity queries."},{"line_number":554,"context_line":""},{"line_number":555,"context_line":"Other deployer impact"},{"line_number":556,"context_line":"---------------------"},{"line_number":557,"context_line":""},{"line_number":558,"context_line":"Besides documentation that fully covers the deprecation of the old server"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a016987_7b71e7a7","line":555,"updated":"2015-05-28 00:23:24.000000000","message":"Do we need to say anything about what happens if you try to migrate/evacuate/rebuild/resize a VM that\u0027s part of an \u0027absolute\u0027 affinity rule?\n\ne.g., the user has a group of VMs with the \u0027near\u0027 directive all on the same host and s/he tries to perform a migration--shouldn\u0027t we block this (knowing that the absolute rule would become violated)?","commit_id":"047c80c378f87d5fee36b70543926904bbd25dd6"}]}
