)]}'
{"specs/rocky/approved/override-compute-node-uuid.rst":[{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"27a9e4d13c6987a0dac66c37c29aec488095ad0f","unresolved":false,"context_lines":[{"line_number":21,"context_line":"Imagine the ``nova-compute`` and Neutron ML2 agent managing the port resources"},{"line_number":22,"context_line":"on a piece of hardware are deployed in separate containers (or VMs) on a"},{"line_number":23,"context_line":"baremetal host computer. We would like both the virt driver in the"},{"line_number":24,"context_line":"``nova-compute`` as well as the Neutron agent to use the"},{"line_number":25,"context_line":"`update_provider_tree`_ interface for tracking inventory and trait information"},{"line_number":26,"context_line":"for the resource providers that each is responsible for."},{"line_number":27,"context_line":""},{"line_number":28,"context_line":".. _update_provider_tree: https://review.openstack.org/#/c/540111/"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_70adcd86","line":25,"range":{"start_line":24,"start_character":2,"end_line":25,"end_character":33},"updated":"2018-03-12 12:26:45.000000000","message":"Neutron doesn\u0027t use update_provider_tree as that is a nova virt driver call but it is a good hypothetical to simplify talking about the problem via a known interface.","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":5754,"name":"Alex Xu","email":"hejie.xu@intel.com","username":"xuhj"},"change_message_id":"b7fe2574ea9000397df97945be820a06bb371ce0","unresolved":false,"context_lines":[{"line_number":21,"context_line":"Imagine the ``nova-compute`` and Neutron ML2 agent managing the port resources"},{"line_number":22,"context_line":"on a piece of hardware are deployed in separate containers (or VMs) on a"},{"line_number":23,"context_line":"baremetal host computer. We would like both the virt driver in the"},{"line_number":24,"context_line":"``nova-compute`` as well as the Neutron agent to use the"},{"line_number":25,"context_line":"`update_provider_tree`_ interface for tracking inventory and trait information"},{"line_number":26,"context_line":"for the resource providers that each is responsible for."},{"line_number":27,"context_line":""},{"line_number":28,"context_line":".. _update_provider_tree: https://review.openstack.org/#/c/540111/"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_406772be","line":25,"range":{"start_line":24,"start_character":2,"end_line":25,"end_character":33},"in_reply_to":"df7087c5_70adcd86","updated":"2018-03-13 12:46:21.000000000","message":"yea, I don\u0027t see this also.\n\nBut is there anyway to share code between nova and neutron. The shared code is more about the strategy of merge traits thing.","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":5754,"name":"Alex Xu","email":"hejie.xu@intel.com","username":"xuhj"},"change_message_id":"b7fe2574ea9000397df97945be820a06bb371ce0","unresolved":false,"context_lines":[{"line_number":77,"context_line":""},{"line_number":78,"context_line":"A Cyborg, Neutron ML2 or DHCP agent wants to create child resource providers"},{"line_number":79,"context_line":"under the parent compute node resource provider record and must be able to"},{"line_number":80,"context_line":"identify the compute node even if the ``nova-compute`` service worker has yet"},{"line_number":81,"context_line":"to create a resource provider record for the compute node in the placement API"},{"line_number":82,"context_line":"(or even before adding a record to the cell DB\u0027s ``compute_nodes`` table)"},{"line_number":83,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_a0a44ea7","line":80,"range":{"start_line":80,"start_character":70,"end_line":80,"end_character":77},"updated":"2018-03-13 12:46:21.000000000","message":"has not yet?","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":6873,"name":"Matt Riedemann","email":"mriedem.os@gmail.com","username":"mriedem"},"change_message_id":"8c624a677818db6d8b35bade08b9e4139840c16b","unresolved":false,"context_lines":[{"line_number":81,"context_line":"to create a resource provider record for the compute node in the placement API"},{"line_number":82,"context_line":"(or even before adding a record to the cell DB\u0027s ``compute_nodes`` table)"},{"line_number":83,"context_line":""},{"line_number":84,"context_line":"An operator wishes to pre-create a batch of resource provider and inventory"},{"line_number":85,"context_line":"records in the placement API so that they can set allocation ratios and traits"},{"line_number":86,"context_line":"against those resource providers in batches. They want to be able to do this"},{"line_number":87,"context_line":"before any ``nova-compute`` service workers even start."},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Proposed change"},{"line_number":90,"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":1,"id":"df7087c5_0050e35c","line":87,"range":{"start_line":84,"start_character":0,"end_line":87,"end_character":55},"updated":"2018-03-09 15:41:10.000000000","message":"What happens when the inventory the operator put into the yaml file is different from what the virt driver calculates? Today we would override anything set \u0027out of band\u0027.","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"27a9e4d13c6987a0dac66c37c29aec488095ad0f","unresolved":false,"context_lines":[{"line_number":107,"context_line":"    for ``nova-compute`` service workers that are running a virt driver that"},{"line_number":108,"context_line":"    controls a **single** compute node. For virt drivers that manage multiple"},{"line_number":109,"context_line":"    compute nodes, like Ironic and PowerVM, alternate methods of overriding"},{"line_number":110,"context_line":"    compute node UUIDs for multiple compute nodes will need to be considered."},{"line_number":111,"context_line":""},{"line_number":112,"context_line":"Code will be added to the ``nova-compute`` resource tracker\u0027s"},{"line_number":113,"context_line":"``_init_compute_node()`` method that will set the ``nova.objects.ComputeNode``"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_30643522","line":110,"updated":"2018-03-12 12:26:45.000000000","message":"I think the problem even more general. This override only works for the case where a single agent creates a single RP. As soon as more than one RPs are managed by a single agent the mapping between the physical entity (e.g. PF) and the ProviderConfig trying to override the RP for that physical entity is ambiguous. \nExamples:\n1) Neutron SRIOV agent managing two PFs. Operator creates two ProviderConfig files but in the current ProviderConfig proposal there is no way to identify which ProviderConfig belongs to which PF.\n2) Nova compute agent handles NUMA nodes as RPs. Each NUMA node RP will have VCPU and MEMORY_MB resources. The deployer wants to override the reserved value on one of the NUMA nodes. With the current proposal there is no way for the deployer to specify which NUMA node the override should be applied by the nova-compute.\n\nI think we have to introduce either a specific config variable in the agents for each ProvideConfig - physical entity mapping OR modify the ProviderConfig to contain some kind of physical identifier (compute host name, NUMA node id, PF PCI address) that makes the mapping between the config and the overridden entity explicit.","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":6873,"name":"Matt Riedemann","email":"mriedem.os@gmail.com","username":"mriedem"},"change_message_id":"7432472d389e67d62969975e15c551b2d3de7894","unresolved":false,"context_lines":[{"line_number":129,"context_line":"Alternatives"},{"line_number":130,"context_line":"------------"},{"line_number":131,"context_line":""},{"line_number":132,"context_line":"None"},{"line_number":133,"context_line":""},{"line_number":134,"context_line":"Data model impact"},{"line_number":135,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_005a234e","line":132,"updated":"2018-03-09 15:49:04.000000000","message":"What if nova sent out an RPC notification that neutron was listening for when the compute node RP was created, with the host and uuid information, and the neutron server routed that to the appropriate agent to then do it\u0027s setup? I guess if nova-compute and the neutron agent consider themselves on different hosts then that doesn\u0027t help.\n\nTrying to think of ways to avoid needing operators to not only manage their deployment of all of the openstack services, but also now managing yaml inventory files across the deployment just so that the services can understand their relationships to each other. Maybe people using ansible to deploy openstack wouldn\u0027t have a hard time doing this since the inventory information is already available to them.\n\nI just wish this wasn\u0027t based on config. While it\u0027s optional, anyone that wants to use the fancy features based on nested providers would need to do this (unless they are just running the services on the same VM, or baremetal host, and not in containers?).","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":6873,"name":"Matt Riedemann","email":"mriedem.os@gmail.com","username":"mriedem"},"change_message_id":"43c1aaf16ecfeb1d5a1d618f35a1e94ccade29f7","unresolved":false,"context_lines":[{"line_number":129,"context_line":"Alternatives"},{"line_number":130,"context_line":"------------"},{"line_number":131,"context_line":""},{"line_number":132,"context_line":"None"},{"line_number":133,"context_line":""},{"line_number":134,"context_line":"Data model impact"},{"line_number":135,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_e3ecd92a","line":132,"in_reply_to":"df7087c5_005a234e","updated":"2018-03-09 15:59:08.000000000","message":"Another alternative that Dan mentioned in IRC is that if this is just about overriding the UUID, we could define a single config option for that in nova.conf and the compute can use it. Default to None and if None then we do what we do today.\n\nI don\u0027t know if we would add that for UUID or RP.name, Dan was saying name, but I also don\u0027t know if it matters as long as the various agents can figure out the discovery part correctly.","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"27a9e4d13c6987a0dac66c37c29aec488095ad0f","unresolved":false,"context_lines":[{"line_number":129,"context_line":"Alternatives"},{"line_number":130,"context_line":"------------"},{"line_number":131,"context_line":""},{"line_number":132,"context_line":"None"},{"line_number":133,"context_line":""},{"line_number":134,"context_line":"Data model impact"},{"line_number":135,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_4b51f80d","line":132,"in_reply_to":"df7087c5_e3ecd92a","updated":"2018-03-12 12:26:45.000000000","message":"If we only look at the compute RP UUID problem then it seems easy to add a new config option that is present in config of both the nova-compute and the neutron agent. But as soon as more override is needed (e.g. allocation ratio, reserved, traits) adding new config options to each override seems a bit overkill (but not impossible). \nThe current proposal has the advantage that it proposes a unified format. While adding config options would mean that different agents might allow overriding similar things with different config options. (e.g. reserving some resources).","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"},{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"27a9e4d13c6987a0dac66c37c29aec488095ad0f","unresolved":false,"context_lines":[{"line_number":187,"context_line":""},{"line_number":188,"context_line":"We may wish to log a warning if an existing compute host gets a non-None"},{"line_number":189,"context_line":"``CONF.provider_descriptor_path`` value and overrides contained in the provider"},{"line_number":190,"context_line":"descriptor YAML file conflict with the existing compute node\u0027s UUID."},{"line_number":191,"context_line":""},{"line_number":192,"context_line":"Implementation"},{"line_number":193,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":1,"id":"df7087c5_ab41b43b","line":190,"updated":"2018-03-12 12:26:45.000000000","message":"++","commit_id":"dff0703382ad4c1053f6513a58b1490d2fab16ad"}]}
