)]}'
{"/COMMIT_MSG":[{"author":{"_account_id":7448,"name":"Carl Baldwin","email":"carl@ecbaldwin.net","username":"carl-baldwin"},"change_message_id":"05d8aca65f2fc785871414985cf1b101da76da19","unresolved":false,"context_lines":[{"line_number":5,"context_line":"CommitDate: 2015-07-08 17:45:28 +0100"},{"line_number":6,"context_line":""},{"line_number":7,"context_line":"Describe and propose the \u0027routed\u0027 network type"},{"line_number":8,"context_line":""},{"line_number":9,"context_line":"Change-Id: Ie745bb278556c14e826f87a7f7095d6253ce8409"}],"source_content_type":"text/x-gerrit-commit-message","patch_set":2,"id":"9a41bdd9_ab02b93c","line":8,"updated":"2015-07-17 16:28:53.000000000","message":"Could you link the rfe [1] here so that we can traverse either way?  I put \"(rfe)\" after the bug so that reviewers would know it wasn\u0027t a regular old bug [2].  It\u0027d be nice to have that here too.\n\n [1] https://bugs.launchpad.net/neutron/+bug/1472704\n [2] https://review.openstack.org/#/c/196812/","commit_id":"44e12a0c094a5932f5972c62ee4fbb89214af68f"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"01be7fad26b5597a51d44de94a044442d3fa29ff","unresolved":false,"context_lines":[{"line_number":5,"context_line":"CommitDate: 2015-07-08 17:45:28 +0100"},{"line_number":6,"context_line":""},{"line_number":7,"context_line":"Describe and propose the \u0027routed\u0027 network type"},{"line_number":8,"context_line":""},{"line_number":9,"context_line":"Change-Id: Ie745bb278556c14e826f87a7f7095d6253ce8409"}],"source_content_type":"text/x-gerrit-commit-message","patch_set":2,"id":"3a50d1a3_364e48c2","line":8,"in_reply_to":"9a41bdd9_ab02b93c","updated":"2015-07-19 09:19:31.000000000","message":"Sure, I will do that in my next upload, coming very shortly.","commit_id":"44e12a0c094a5932f5972c62ee4fbb89214af68f"}],"doc/source/devref/routed_networks.rst":[{"author":{"_account_id":7715,"name":"Hirofumi Ichihara","email":"ichihara.hirofumi@gmail.com","username":"Hirofumi-Ichihara"},"change_message_id":"372e995b2dcef7f54d37b69ecaabc0edac2b44e6","unresolved":false,"context_lines":[{"line_number":22,"context_line":"- as a practical consequence, it is easier to debug, using standard"},{"line_number":23,"context_line":"  tools such as ping, traceroute, wireshark and tcpdump"},{"line_number":24,"context_line":""},{"line_number":25,"context_line":"- its scale is not limited in the way that VLAN- and VXLAN-based"},{"line_number":26,"context_line":"  networks are, by the practical diameter of the physical underlying"},{"line_number":27,"context_line":"  L2 network."},{"line_number":28,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_ae188076","line":25,"updated":"2015-07-07 05:46:59.000000000","message":"It confused.\n VLAN- -\u003e VLAN-based\n\nI also feel odd VLAN-based network is not limited.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":22,"context_line":"- as a practical consequence, it is easier to debug, using standard"},{"line_number":23,"context_line":"  tools such as ping, traceroute, wireshark and tcpdump"},{"line_number":24,"context_line":""},{"line_number":25,"context_line":"- its scale is not limited in the way that VLAN- and VXLAN-based"},{"line_number":26,"context_line":"  networks are, by the practical diameter of the physical underlying"},{"line_number":27,"context_line":"  L2 network."},{"line_number":28,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_268e7e2e","line":25,"in_reply_to":"ba3cc151_ae188076","updated":"2015-07-08 16:43:58.000000000","message":"Thanks for your comments.\n\n- I\u0027ll change VLAN- to VLAN-based in the next revision.\n\n- Regarding \u0027limited\u0027, my understanding is that L2 segments are limited in practice to a diameter of around 10 switches, for reasons associated with the L2 protocols such as spanning tree.  That\u0027s what I had in mind here.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7448,"name":"Carl Baldwin","email":"carl@ecbaldwin.net","username":"carl-baldwin"},"change_message_id":"e5d658fbce25079d0be37edaa4c1a8ba33ffe430","unresolved":false,"context_lines":[{"line_number":31,"context_line":""},{"line_number":32,"context_line":"A routed network is described on the Neutron API using a new provider"},{"line_number":33,"context_line":"network type: TYPE_ROUTED \u003d \u0027routed\u0027.  The related physical network"},{"line_number":34,"context_line":"and segmentation ID parameters are not meaningful and should be left"},{"line_number":35,"context_line":"unspecified."},{"line_number":36,"context_line":""},{"line_number":37,"context_line":"A routed network can be shared or private; this is indicated as usual"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_2f6b5587","line":34,"updated":"2015-07-03 20:48:47.000000000","message":"These might be useful if different address scopes are in use.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":31,"context_line":""},{"line_number":32,"context_line":"A routed network is described on the Neutron API using a new provider"},{"line_number":33,"context_line":"network type: TYPE_ROUTED \u003d \u0027routed\u0027.  The related physical network"},{"line_number":34,"context_line":"and segmentation ID parameters are not meaningful and should be left"},{"line_number":35,"context_line":"unspecified."},{"line_number":36,"context_line":""},{"line_number":37,"context_line":"A routed network can be shared or private; this is indicated as usual"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_139cd768","line":34,"in_reply_to":"ba3cc151_2f6b5587","updated":"2015-07-08 16:43:58.000000000","message":"I guess what you have in mind is that segmentation ID could be used to indicate the namespace name on the compute host - which in turn would I imagine be derived from the address scope ID.\n\nHowever, if a network is associated with a non-default address scope, isn\u0027t there already another network parameter that specifies that?  If so, wouldn\u0027t it be over-specification to (effectively) duplicate that information by encoding the address scope in segmentation ID as well?  (And also create scope for configuration errors, if the two settings did not match each other.)  I think I\u0027d rather leave segmentation ID unused, and derive the namespace name from the other address scope parameter.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7776,"name":"Paul Carver","email":"pcarver@att.com","username":"pcarver"},"change_message_id":"14a691858558e094d93c499042a36e68948da402","unresolved":false,"context_lines":[{"line_number":40,"context_line":"Neutron router objects are not used with routed networks.  Subject to"},{"line_number":41,"context_line":"security policy, there is automatically potential connectivity between"},{"line_number":42,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":43,"context_line":"routed networks and the outside world."},{"line_number":44,"context_line":""},{"line_number":45,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":46,"context_line":"preceding connectivity point, it is practical to configure two routed"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_ff6dd8cc","line":43,"updated":"2015-07-07 01:42:06.000000000","message":"I haven\u0027t read up on address scopes. How similar are they to VRF? I think there\u0027s a synergy between L3 only networks and BGP VPNs. They\u0027re obviously different things, but I see overlap in the use cases.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":40,"context_line":"Neutron router objects are not used with routed networks.  Subject to"},{"line_number":41,"context_line":"security policy, there is automatically potential connectivity between"},{"line_number":42,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":43,"context_line":"routed networks and the outside world."},{"line_number":44,"context_line":""},{"line_number":45,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":46,"context_line":"preceding connectivity point, it is practical to configure two routed"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_d312efb3","line":43,"in_reply_to":"ba3cc151_ff6dd8cc","updated":"2015-07-08 16:43:58.000000000","message":"In general networking terms I believe address scopes and VRFs are the same idea.  I.e. the same IP address can mean different things in different contexts (aka scopes aka VRFs aka namespaces), and a packet addressed to that IP address can be routed differently in those different contexts.\n\nI\u0027m not yet familiar, though, with all the work within Neutron on VRFs and BGP VPNs.  Could you point me to more details about any relevant work or use case discussion that you think overlaps?","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7448,"name":"Carl Baldwin","email":"carl@ecbaldwin.net","username":"carl-baldwin"},"change_message_id":"e5d658fbce25079d0be37edaa4c1a8ba33ffe430","unresolved":false,"context_lines":[{"line_number":47,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":48,"context_line":"with a range of globally routable IP addresses.  Then, when launching"},{"line_number":49,"context_line":"a VM, it can simply be attached to the latter if it requires inbound"},{"line_number":50,"context_line":"connectivity from the Internet, and to the former if not."},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"Connectivity Implementation - Shared"},{"line_number":53,"context_line":"------------------------------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_4f16d9f2","line":50,"updated":"2015-07-03 20:48:47.000000000","message":"Would all networks be admin created and owned?  Is there any use case for tenant private \"networks\".  At this point, networks are little more than ip blocks (possibly assigned to an address scope).  I guess each must have a gateway ip for the VM instances default route on the network.\n\nWhat about the broadcast address normally reserved for L2 networks?  I guess since the VM still thinks it is connected to an L2 network, it would still need to be reserved.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":47,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":48,"context_line":"with a range of globally routable IP addresses.  Then, when launching"},{"line_number":49,"context_line":"a VM, it can simply be attached to the latter if it requires inbound"},{"line_number":50,"context_line":"connectivity from the Internet, and to the former if not."},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"Connectivity Implementation - Shared"},{"line_number":53,"context_line":"------------------------------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_b6acf99f","line":50,"in_reply_to":"ba3cc151_4f16d9f2","updated":"2015-07-08 16:43:58.000000000","message":"The shared (\u003d\u003d default address scope) networks would be admin created and owned, but yes, there certainly is a use case for tenant private networks as well, so that tenants can use overlapping IPv4 addresses.  So for the latter case, tenants should be able to create private address scopes, and then networks in those scopes.  (I\u0027m partly guessing here about how I imagine scopes work, so I hope I\u0027ve got that right.)\n\nRe the gateway IP - correct, and DHCP communicates this to the instances just as it does for other network implementations.  (Note though that an instance is free to ignore that if it wants and program its default route as a direct one, or an indirect one via any other IP address.  To ensure that instance data is always correctly routed and firewalled anyway, the compute host responds to all ARP requests from the instance, with its own MAC address.)\n\nRe reservation of the broadcast address - yes, correct there too I believe.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7715,"name":"Hirofumi Ichihara","email":"ichihara.hirofumi@gmail.com","username":"Hirofumi-Ichihara"},"change_message_id":"372e995b2dcef7f54d37b69ecaabc0edac2b44e6","unresolved":false,"context_lines":[{"line_number":53,"context_line":"------------------------------------"},{"line_number":54,"context_line":""},{"line_number":55,"context_line":"In the shared routed network case, everything happens in the default"},{"line_number":56,"context_line":"namespace of the relevant compute hosts.  Standard Linux routing"},{"line_number":57,"context_line":"routes VM data, with iptables used to implement the configured"},{"line_number":58,"context_line":"security policy."},{"line_number":59,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_2eea30ad","line":56,"updated":"2015-07-07 05:46:59.000000000","message":"What does namespace? Is it linux network namespace?","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":53,"context_line":"------------------------------------"},{"line_number":54,"context_line":""},{"line_number":55,"context_line":"In the shared routed network case, everything happens in the default"},{"line_number":56,"context_line":"namespace of the relevant compute hosts.  Standard Linux routing"},{"line_number":57,"context_line":"routes VM data, with iptables used to implement the configured"},{"line_number":58,"context_line":"security policy."},{"line_number":59,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_7693d15b","line":56,"in_reply_to":"ba3cc151_2eea30ad","updated":"2015-07-08 16:43:58.000000000","message":"Yes.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7715,"name":"Hirofumi Ichihara","email":"ichihara.hirofumi@gmail.com","username":"Hirofumi-Ichihara"},"change_message_id":"372e995b2dcef7f54d37b69ecaabc0edac2b44e6","unresolved":false,"context_lines":[{"line_number":81,"context_line":" 10.65.0.23      172.18.203.129  255.255.255.255 UGH   0      0        0 eth0"},{"line_number":82,"context_line":" 10.65.0.24      0.0.0.0         255.255.255.255 UH    0      0        0 tapa429fb36-04"},{"line_number":83,"context_line":" 172.18.203.0    0.0.0.0         255.255.255.0   U     0      0        0 eth0"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"This shows one local VM on this host with IP address 10.65.0.24,"},{"line_number":86,"context_line":"accessed via a TAP named tapa429fb36-04; and three VMs, with the .21,"},{"line_number":87,"context_line":".22 and .23 addresses, on two other hosts (172.18.203.126 and .129),"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_ae2ac08e","line":84,"updated":"2015-07-07 05:46:59.000000000","message":"Do you mean that these configurations are set directly on comupute host without network namespace?","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":81,"context_line":" 10.65.0.23      172.18.203.129  255.255.255.255 UGH   0      0        0 eth0"},{"line_number":82,"context_line":" 10.65.0.24      0.0.0.0         255.255.255.255 UH    0      0        0 tapa429fb36-04"},{"line_number":83,"context_line":" 172.18.203.0    0.0.0.0         255.255.255.0   U     0      0        0 eth0"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"This shows one local VM on this host with IP address 10.65.0.24,"},{"line_number":86,"context_line":"accessed via a TAP named tapa429fb36-04; and three VMs, with the .21,"},{"line_number":87,"context_line":".22 and .23 addresses, on two other hosts (172.18.203.126 and .129),"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_f68da131","line":84,"in_reply_to":"ba3cc151_ae2ac08e","updated":"2015-07-08 16:43:58.000000000","message":"Yes, that\u0027s correct.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7715,"name":"Hirofumi Ichihara","email":"ichihara.hirofumi@gmail.com","username":"Hirofumi-Ichihara"},"change_message_id":"372e995b2dcef7f54d37b69ecaabc0edac2b44e6","unresolved":false,"context_lines":[{"line_number":86,"context_line":"accessed via a TAP named tapa429fb36-04; and three VMs, with the .21,"},{"line_number":87,"context_line":".22 and .23 addresses, on two other hosts (172.18.203.126 and .129),"},{"line_number":88,"context_line":"and hence with routes via those compute host addresses."},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"DHCP"},{"line_number":91,"context_line":"----"},{"line_number":92,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_297f8a67","line":89,"updated":"2015-07-07 05:46:59.000000000","message":"Does the compute host has just one interface? It confuses me. I wonder if you could visualize above explanation.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":86,"context_line":"accessed via a TAP named tapa429fb36-04; and three VMs, with the .21,"},{"line_number":87,"context_line":".22 and .23 addresses, on two other hosts (172.18.203.126 and .129),"},{"line_number":88,"context_line":"and hence with routes via those compute host addresses."},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"DHCP"},{"line_number":91,"context_line":"----"},{"line_number":92,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_96953533","line":89,"in_reply_to":"ba3cc151_297f8a67","updated":"2015-07-08 16:43:58.000000000","message":"I\u0027m not sure if you mean its interface (eth0) to other compute hosts, and to other parts of the DC fabric, or its interface (tapa429fb36-04) to a guest VM.\n\nNormally a compute host will have multiple instances on it, and so there will be multiple TAP interfaces like tapa429fb36-04.\n\nFor the fabric interface, however (eth0), it is fine either for the compute host to have just this interface, or for it to be multi-homed onto the fabric.\n\nDoes that make sense?\n\nI don\u0027t think I can easily provide a picture here.  There is a picture at http://docs.projectcalico.org/en/latest/addressing.html though - does it help?","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7448,"name":"Carl Baldwin","email":"carl@ecbaldwin.net","username":"carl-baldwin"},"change_message_id":"e5d658fbce25079d0be37edaa4c1a8ba33ffe430","unresolved":false,"context_lines":[{"line_number":93,"context_line":"DHCP service in this type of network can be provided by Dnsmasq using"},{"line_number":94,"context_line":"its --bridge-interface option.  A patch with the necessary Neutron"},{"line_number":95,"context_line":"DHCP agent modifications can be seen at"},{"line_number":96,"context_line":"https://review.openstack.org/#/c/197578/."},{"line_number":97,"context_line":""},{"line_number":98,"context_line":"Connectivity Implementation - Private"},{"line_number":99,"context_line":"-------------------------------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_efd1ed30","line":96,"updated":"2015-07-03 20:48:47.000000000","message":"Curious, does the DHCP \"port\" still consume an address in this type of network?","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":93,"context_line":"DHCP service in this type of network can be provided by Dnsmasq using"},{"line_number":94,"context_line":"its --bridge-interface option.  A patch with the necessary Neutron"},{"line_number":95,"context_line":"DHCP agent modifications can be seen at"},{"line_number":96,"context_line":"https://review.openstack.org/#/c/197578/."},{"line_number":97,"context_line":""},{"line_number":98,"context_line":"Connectivity Implementation - Private"},{"line_number":99,"context_line":"-------------------------------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_29140ee7","line":96,"in_reply_to":"ba3cc151_efd1ed30","updated":"2015-07-08 16:43:58.000000000","message":"No, it doesn\u0027t.  The DHCP port does need an address - otherwise Dnsmasq doesn\u0027t work - but we can use the subnet\u0027s gateway IP address for this purpose, on every relevant compute host.  (We just need to be careful that BGP doesn\u0027t export a route for that address!)","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7448,"name":"Carl Baldwin","email":"carl@ecbaldwin.net","username":"carl-baldwin"},"change_message_id":"e5d658fbce25079d0be37edaa4c1a8ba33ffe430","unresolved":false,"context_lines":[{"line_number":103,"context_line":""},{"line_number":104,"context_line":"- For each private address scope, there is a corresponding non-default"},{"line_number":105,"context_line":"  namespace on the host, in which the routing for that address scope"},{"line_number":106,"context_line":"  is performed."},{"line_number":107,"context_line":""},{"line_number":108,"context_line":"- The TAP devices for ports in a private address scope are moved into"},{"line_number":109,"context_line":"  the corresponding namespace, on the host side."}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_cfe0a9fd","line":106,"updated":"2015-07-03 20:48:47.000000000","message":"I think this could be reconciled with my work on address scopes as a first-class object in the Liberty time frame.  I think that you should force the use of address scopes with routed networks in order to benefit from the address overlap protection and possibly the address management of subnet pools (from Kilo).","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":103,"context_line":""},{"line_number":104,"context_line":"- For each private address scope, there is a corresponding non-default"},{"line_number":105,"context_line":"  namespace on the host, in which the routing for that address scope"},{"line_number":106,"context_line":"  is performed."},{"line_number":107,"context_line":""},{"line_number":108,"context_line":"- The TAP devices for ports in a private address scope are moved into"},{"line_number":109,"context_line":"  the corresponding namespace, on the host side."}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_89951a33","line":106,"in_reply_to":"ba3cc151_cfe0a9fd","updated":"2015-07-08 16:43:58.000000000","message":"I haven\u0027t looked at that work properly yet, but I will now...","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7715,"name":"Hirofumi Ichihara","email":"ichihara.hirofumi@gmail.com","username":"Hirofumi-Ichihara"},"change_message_id":"372e995b2dcef7f54d37b69ecaabc0edac2b44e6","unresolved":false,"context_lines":[{"line_number":114,"context_line":"  Neutron L2 network types."},{"line_number":115,"context_line":""},{"line_number":116,"context_line":"Work Needed"},{"line_number":117,"context_line":"-----------"},{"line_number":118,"context_line":""},{"line_number":119,"context_line":"For the shared case, the following work is needed for Neutron to"},{"line_number":120,"context_line":"support routed networks."}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_49ba0ee3","line":117,"updated":"2015-07-07 05:46:59.000000000","message":"Doesn\u0027t this proposal need to modify Nova in order to achieve the requirements?","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":114,"context_line":"  Neutron L2 network types."},{"line_number":115,"context_line":""},{"line_number":116,"context_line":"Work Needed"},{"line_number":117,"context_line":"-----------"},{"line_number":118,"context_line":""},{"line_number":119,"context_line":"For the shared case, the following work is needed for Neutron to"},{"line_number":120,"context_line":"support routed networks."}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_a9e81eb4","line":117,"in_reply_to":"ba3cc151_49ba0ee3","updated":"2015-07-08 16:43:58.000000000","message":"Yes, but the required Nova modification is already merged now: please see https://review.openstack.org/#/c/146914/.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7715,"name":"Hirofumi Ichihara","email":"ichihara.hirofumi@gmail.com","username":"Hirofumi-Ichihara"},"change_message_id":"372e995b2dcef7f54d37b69ecaabc0edac2b44e6","unresolved":false,"context_lines":[{"line_number":138,"context_line":"for IP multicast, but we propose to cover those in separate future"},{"line_number":139,"context_line":"phases.  These are not expected to require any further API-level"},{"line_number":140,"context_line":"changes or enhancements."},{"line_number":141,"context_line":"  "},{"line_number":142,"context_line":"References"},{"line_number":143,"context_line":"----------"},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_ae716046","line":141,"updated":"2015-07-07 05:46:59.000000000","message":"Please remove the space.","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"04545e72e9af90895a02669e3a271909719ffdb8","unresolved":false,"context_lines":[{"line_number":138,"context_line":"for IP multicast, but we propose to cover those in separate future"},{"line_number":139,"context_line":"phases.  These are not expected to require any further API-level"},{"line_number":140,"context_line":"changes or enhancements."},{"line_number":141,"context_line":"  "},{"line_number":142,"context_line":"References"},{"line_number":143,"context_line":"----------"},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba3cc151_e9306658","line":141,"in_reply_to":"ba3cc151_ae716046","updated":"2015-07-08 16:43:58.000000000","message":"Done","commit_id":"f20e662abc7a5e4846ae221882eef661cf182926"},{"author":{"_account_id":7776,"name":"Paul Carver","email":"pcarver@att.com","username":"pcarver"},"change_message_id":"cfc9e71eba4f7c378c0175673735c876fcffcfca","unresolved":false,"context_lines":[{"line_number":22,"context_line":"- as a practical consequence, it is easier to debug, using standard"},{"line_number":23,"context_line":"  tools such as ping, traceroute, wireshark and tcpdump"},{"line_number":24,"context_line":""},{"line_number":25,"context_line":"- its scale is not limited in the way that VLAN-based and VXLAN-based"},{"line_number":26,"context_line":"  networks are, by the practical diameter of the physical underlying"},{"line_number":27,"context_line":"  L2 network."},{"line_number":28,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"ba3cc151_b1170aa2","line":25,"updated":"2015-07-09 00:59:25.000000000","message":"I would say that you\u0027re right about the limitations of VLAN-based networks. Large spanning tree networks are a bad idea. However, I\u0027m not sure it\u0027s fair to equate VXLAN with VLAN in this context. A lot can be done with MLAG and related technologies, but an L3 network is much superior than bridging VLANs throughout the whole datacenter or carving the datacenter into a bunch of unequal bridged regions.","commit_id":"44e12a0c094a5932f5972c62ee4fbb89214af68f"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"01be7fad26b5597a51d44de94a044442d3fa29ff","unresolved":false,"context_lines":[{"line_number":22,"context_line":"- as a practical consequence, it is easier to debug, using standard"},{"line_number":23,"context_line":"  tools such as ping, traceroute, wireshark and tcpdump"},{"line_number":24,"context_line":""},{"line_number":25,"context_line":"- its scale is not limited in the way that VLAN-based and VXLAN-based"},{"line_number":26,"context_line":"  networks are, by the practical diameter of the physical underlying"},{"line_number":27,"context_line":"  L2 network."},{"line_number":28,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"3a50d1a3_76b0508c","line":25,"in_reply_to":"ba3cc151_b1170aa2","updated":"2015-07-19 09:19:31.000000000","message":"I think you\u0027re right, so I\u0027ll remove \"and VXLAN-based\" here.\n\n(I was thinking that a VXLAN network still needs to be based on a single underlying L2 segment, _if_ it wants to be able to multicast and broadcast data in the most efficient way possible.  If it isn\u0027t based on a single underlying L2 segment, I believe it either needs to do head-end packet replication, or to require multicast protocol handling (PIM/IGMP etc.) in the underlying fabric.\n\nHowever, all of that is equally true for doing multicast through a routed network, so it\u0027s clearly not a correct point for me to make in this context.)","commit_id":"44e12a0c094a5932f5972c62ee4fbb89214af68f"},{"author":{"_account_id":105,"name":"Kyle Mestery","email":"mestery@mestery.com","username":"mestery"},"change_message_id":"37abfc39d9395a73d05997d959e0e1f6ab41f367","unresolved":false,"context_lines":[{"line_number":35,"context_line":""},{"line_number":36,"context_line":"A routed network can be shared or private; this is indicated as usual"},{"line_number":37,"context_line":"by presence or absence of the \u0027shared\u0027 flag.  The initial"},{"line_number":38,"context_line":"implementation only supports shared, though; private routed networks"},{"line_number":39,"context_line":"will follow later."},{"line_number":40,"context_line":""},{"line_number":41,"context_line":"Neutron router objects are not used with routed networks.  Subject to"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_b536ae01","line":38,"updated":"2015-07-20 14:10:06.000000000","message":"Why not support private from the start?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"d45ce6f5efe99dc82c9d322097da4add4ae242e1","unresolved":false,"context_lines":[{"line_number":35,"context_line":""},{"line_number":36,"context_line":"A routed network can be shared or private; this is indicated as usual"},{"line_number":37,"context_line":"by presence or absence of the \u0027shared\u0027 flag.  The initial"},{"line_number":38,"context_line":"implementation only supports shared, though; private routed networks"},{"line_number":39,"context_line":"will follow later."},{"line_number":40,"context_line":""},{"line_number":41,"context_line":"Neutron router objects are not used with routed networks.  Subject to"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_66e5d407","line":38,"in_reply_to":"3a50d1a3_b536ae01","updated":"2015-07-20 17:47:50.000000000","message":"Interesting question.  We certainly could do, so far as the API is concerned; but the Calico project doesn\u0027t yet have a complete working implementation for private, and there aren\u0027t (yet) any other implementations...\n\nAlso note that there are a couple of possible meanings of \u0027private\u0027, I think.  When a tenant creates a private Neutron network, the privacy has two consequences: firstly that the tenant can use IP addressing that might overlap with other networks, and secondly that the VMs on that network can only (until routers are added in...) communicate to each other, and not to VMs on other networks.\n\nWith a shared, routed network that is provisioned by the operator, I believe it\u0027s still easy for different tenants to come along and get the second privacy aspect.  Each tenant just needs to create its own security group, and use that security group when launching its own instances.\n\nThe first privacy aspect is trickier: we have to use namespaces on the compute hosts to allow overlapping IPs.  We have a pretty good idea how to do that, but no actual full implementation yet.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":748,"name":"Armando Migliaccio","email":"armamig@gmail.com","username":"armando-migliaccio"},"change_message_id":"882b77435a6f981c4b9d161e5fc8ed5c2a644040","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_63bff4fb","line":46,"updated":"2015-07-21 00:36:53.000000000","message":"I find the fact that we got four typical Neutron constructs that do not work with this proposal (segmentation id not meaningful, logical routers, floating ip and overlapping ips) a bit troublesome. The fact that only a single project can support this type of network leads me to believe that the semantic can vary greatly from one implementation to another (when/if another arise), and I wouldn\u0027t want to add further confusion to the user. Ultimately I wonder if this is really \u0027core\u0027 material, or if could be a natural extension to the Calico project. After all, it seems that you\u0027re only adding a type driver, but a type driver can as well live outside this source tree (closer to the calico project) where it makes more sense. Have you considered this?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"f8b0b98b935055143738d537b3542a4d07056064","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_67182b85","line":46,"in_reply_to":"3a50d1a3_0f1ba781","updated":"2015-07-27 16:54:17.000000000","message":"I agree Ryan, this is the nub of the problem.  However, we are stating, fairly clearly, that Neutron is the networking API for OpenStack.  If that is the case, then both L2 and L3 models should be first-class citizens in it, no?  This becomes even more important when we start supporting (for real) containers (where almost all the models are L3 rather than L2 based).  \n\nSo, we have two, realistic options, nudge (as gently as possible) Neutron to provide equivalent function in both L2 and L3 forwarding domains (the, I hope, preferred option), or just come out and say that Neutron is L2, and that a parallel effort is necessary for L3.  \n\nThat, I think, is the worse of the two options, as it would require OpenStack developers to support two separate API infrastructures (and potentially application developers as well).\n\nI don\u0027t think that is something that any of us want.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":105,"name":"Kyle Mestery","email":"mestery@mestery.com","username":"mestery"},"change_message_id":"cb30ee33a74dce56a52a72fa1f743bf1c0136ec6","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_7310fb01","line":46,"in_reply_to":"3a50d1a3_63bff4fb","updated":"2015-07-21 15:06:49.000000000","message":"I like Armando\u0027s idea here.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"8be4d2f2ae202a0f2953bf04633d7760636d087f","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_83ba476d","line":46,"in_reply_to":"3a50d1a3_63bff4fb","updated":"2015-07-24 22:39:23.000000000","message":"Interesting points, Armando.  However, let\u0027s look at these from two different views.  The first view is that of artifacts.  We have the concept of \"segment\u0027 or \"logical router\" in Neutron because it was conceived as an L2 model.  In an L2 model, you need to identify a network segment/broadcast domain/subnet.  Similarly, in a distributed L2 network, you may want a distributed (logical) L3 router.  However, we are talking about a routed network here.  A routed network doesn\u0027t really see things as segments.  It sees them a routing destinations/prefixes.  Similarly, it doesn\u0027t need \"logical routers\" as the network elements are natively routers.\n\nThe other lens is one of function.  I\u0027d group floating ip and overlapping ip in that bucket.  The first, how do I assign a \"service\" IP to an endpoint.  In an L2 network, I usually need to NAT and proxy ARP it.  In an L3 network you simply point another route at that container or VM.  The behavior on the VM is a bit different (in the former, it has to deal with being behind a NAT, in the later, it needs to know to listen on the service address as well as the \"native\" address).  We can certainly support \"floating IP\" in Calico (In the IP world, this is called Anycast, where Any \u003e\u003d 1).  We hesitate to say we support \"floating IP\" because Neutron scoped it to the former NAT/L2 model.  The functional result, however, is the same.  The same can be said for overlapping IP\u0027s.  There are multiple ways of handling that in a routed network, from VRFs to 464XLAT.  So, they can be supported.  I don\u0027t believe we said that overlapping IPs were not supported in this model, btw.\n\nThe larger point here is that the Neutron lens is, at it\u0027s core, an L2 lens.  What we are trying to do here is provide a reasonable way to support L3 models as well as L2 models in Neutron while minimizing the disruption to the existing L2-centric model.   We REALLY believe that some of these concepts NEED to be in Neutron.  If they are not, then Neutron will continue to develop models that are L2-centric, and any L3 options out there will continue to have to adapt to more and more L2 assumptions.\n\nIf we say that Neutron is the networking API for OpenStack, then we need to insure, as a group, that that core infrastructure supports both L2 and L3 models.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"d27ec9ef13e06ec376114cd922345a1add43a732","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_802f5d3d","line":46,"in_reply_to":"3a50d1a3_67182b85","updated":"2015-07-27 17:34:10.000000000","message":"Again, my preference would be to call the L3 forwarding domain what it is and not network.  However, that\u0027s a change that involves other projects and so I\u0027m doubtful it would happen. \n\nGiven that, we are stuck in my apple/orange/banana analogy and given that I like Armando\u0027s idea as a first step...","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"c5ccaf01767535ace3a7ebcb35baf221d8bbb040","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_aa3c7393","line":46,"in_reply_to":"3a50d1a3_802f5d3d","updated":"2015-07-28 23:07:32.000000000","message":"Ryan - just to clarify, by \"Armando\u0027s idea\" do you mean working on this outside of the Neutron core tree, and so not trying to find a way to model this nicely in Neutron?  (I suspect not - so may well be missing either your or Armando\u0027s point here.)","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"f304f90197a613147c372d294250315009787c03","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_4f0661cf","line":46,"in_reply_to":"3a50d1a3_802f5d3d","updated":"2015-07-27 18:58:49.000000000","message":"Ryan, I (sort of) agree with you.  The problem is that we\u0027ve entangled the thing we are delivering (a network) with a specific mechanism (an L2 forwarding environment).  We should call what we are delivering a network that provides connectivity and policy controls.  How that gets rendered is a driver-specific thing (i.e. tunneled L2, native L2 switched, routed, carrier pigeon, etc.).  However, we are where we are.\n\nCan you and/or Armando suggest how changing the semantics into two trains (L2 and L3) will not drive two disjunct code paths in any code that needs to manage a \"Neutron\" network based on which model (L3 vs. L2) is in use?  I don\u0027t see how that could be done, but I may be missing something.\n\nThe model we are supporting does support \"floating ip\", just in a different (more IP friendly way), overlapping IPs can also be supported.  Logical routers are a noop (as everything is a router).  So the only thing that doesn\u0027t make much sense is a segment id.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"88587e1a181c83398a84da8061bb31e740267e83","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_0f1ba781","line":46,"in_reply_to":"3a50d1a3_83ba476d","updated":"2015-07-26 23:02:02.000000000","message":"If the term \"network\" is going to be overloaded, then I tend to agree with Armando.  (Mind you, I wish the network concept had been defined differently, but given that the semantics are what they are, I\u0027m not yet comfortable with saying we are going to change it\u0027s definition to a \"well it\u0027s an apple, unless you choose this particular type, when it becomes an orange, or this other particular type, in which case it becomes a banana [the last is looking down the road to the next thing that wants to change network semantics])","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":17594,"name":"Bjorn Nelson","email":"bjorn@belovedarctos.com","username":"osleep"},"change_message_id":"2ea5117fc54c1f1372c11a9f7f0a81fd08616f6c","unresolved":false,"context_lines":[{"line_number":43,"context_line":"all ports on routed networks in the same address scope; and between"},{"line_number":44,"context_line":"routed networks and the outside world."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Floating IPs are not used with routed networks.  Because of the"},{"line_number":47,"context_line":"preceding connectivity point, it is practical to configure two routed"},{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_8e274462","line":46,"in_reply_to":"3a50d1a3_aa3c7393","updated":"2015-07-29 02:18:38.000000000","message":"Based on our own implementation that is very similar to Project Calico:\n\nsegmention_id: This might actually be useful to Project Calico at scale.  At a certain size, we found a need to separate networks, even though this is a few thousand nodes per segment.\n\nlogical routers: I might be misunderstanding the implementation, but aren\u0027t there benefits to not having a choke point?\n\nfloating and overlapping ips: nat actually prevents some services from working that you might find in an enterprise that has private clouds.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":105,"name":"Kyle Mestery","email":"mestery@mestery.com","username":"mestery"},"change_message_id":"37abfc39d9395a73d05997d959e0e1f6ab41f367","unresolved":false,"context_lines":[{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"},{"line_number":50,"context_line":"a VM, it can simply be attached to the latter if it requires inbound"},{"line_number":51,"context_line":"connectivity from the Internet, and to the former if not."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"Connectivity Implementation - Shared"},{"line_number":54,"context_line":"------------------------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_2757d503","line":51,"updated":"2015-07-20 14:10:06.000000000","message":"Who decides where it attaches? The operator? Based on the fact you don\u0027t support private networks here, I\u0027m inclined to believe the operator creates the network types and users just launch VMs on those.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"d45ce6f5efe99dc82c9d322097da4add4ae242e1","unresolved":false,"context_lines":[{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"},{"line_number":50,"context_line":"a VM, it can simply be attached to the latter if it requires inbound"},{"line_number":51,"context_line":"connectivity from the Internet, and to the former if not."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"Connectivity Implementation - Shared"},{"line_number":54,"context_line":"------------------------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_c6ddc815","line":51,"in_reply_to":"3a50d1a3_2757d503","updated":"2015-07-20 17:47:50.000000000","message":"Yes, for the shared network(s) that\u0027s correct.  But (as I tried to say just above) there could be multiple shared networks provisioned, with different IP addressing, and a user could choose one of those depending on what kind of IP addressing they need.\n\nIf/when we support private networks too, I expect that tenants will be able to create those.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"8be4d2f2ae202a0f2953bf04633d7760636d087f","unresolved":false,"context_lines":[{"line_number":48,"context_line":"networks, one with DC-private (e.g. RFC 1918) IP addresses, and one"},{"line_number":49,"context_line":"with a range of globally routable IP addresses.  Then, when launching"},{"line_number":50,"context_line":"a VM, it can simply be attached to the latter if it requires inbound"},{"line_number":51,"context_line":"connectivity from the Internet, and to the former if not."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"Connectivity Implementation - Shared"},{"line_number":54,"context_line":"------------------------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_febac009","line":51,"in_reply_to":"3a50d1a3_c6ddc815","updated":"2015-07-24 22:39:23.000000000","message":"An example might help, here.  In the shared environment, a operator could create three networks, \"private-v4\", \"public-v4\" and \"v6\".  Those could be 10.64.0.0/10, 198.51.100.0/24, and 2001:db8:1000::/48\n\nA given project/tenant could create VMs and connect as many of them as they wanted to the \"public-v4\" and \"v6\" networks, but only connect a subset of their VMs to the  \"public-v4\" (based on what they had paid for, enforced by quota).  \n\nAn interesting aspect of this is that a VM that needs to be exposed to the public world using v4 just would get a publicly routed v4 address, it would not need NAT.  Similarly, it could connect to private v4 addressed VMs (such as database backends, etc) again, without the use of NAT, because this is a routed infrastructure.  All of this limited, of course, by the application of security groups.\n\nOther models are possible, of course.  But its leveraging standard OpenStack provisioning, resource management and quota infrastructure to do it.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":105,"name":"Kyle Mestery","email":"mestery@mestery.com","username":"mestery"},"change_message_id":"e1d8f5b3ea0fb9664d039f8efd383d73a8b5b7c9","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_787573c9","line":67,"updated":"2015-07-20 17:10:35.000000000","message":"How does this work for Xen, Hyper-V, and VMware compute host drivers? This document seems to be tieing the logical model very closely to the implementation details of this running on a Linux host with a KVM hypervisor.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"ec391e48fd0415b9414248238b9812c7340d772e","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_c1d560e5","line":67,"in_reply_to":"3a50d1a3_06883ec4","updated":"2015-07-27 21:30:52.000000000","message":"The first part would explain historical context.\n\nThe second part would explain what is needed to use this in other hypervisors/OSs and is based on your earlier contribution to this thread.\n\nThere may be other implementations that don\u0027t require those things, but I don\u0027t see why that is a reason not to make the document more accessible to developers who weren\u0027t privy to this discussion thread...","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"d27ec9ef13e06ec376114cd922345a1add43a732","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_a5814b91","line":67,"in_reply_to":"3a50d1a3_270c6393","updated":"2015-07-27 17:34:10.000000000","message":"I would add a \"Background\" section after the introduction that looks sort of like the following (feel free to flesh it out)\n\nBackground\n---------------\n\nWhile the Calico project [#]_ is where current implementation has been incubated, the resulting datapath concepts provide a framework for a more generic L3 networking model.\n\nSpecifically, while the current implementation is linux-based, there are two requirements for adapting this work to another OS/hypervisor pair:\n\n#. The hypervisor needs to support the ability to create virtual interfaces \n   and bond compute instances (VMs, containers or process) to those interfaces.\n#. The OS must have the ability to manipulate and inspect its routing and ACL\n   tables via an API.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"f8b0b98b935055143738d537b3542a4d07056064","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_270c6393","line":67,"in_reply_to":"3a50d1a3_2f06a320","updated":"2015-07-27 16:54:17.000000000","message":"Which points, Ryan?  What Calico supports today, vs what the model would support?  We can do this, but we really don\u0027t want this to be Calico specific, others should be able to use it (we\u0027re trying to make this a general L3 model, not a Calico-only model).  Can you propose some text (or skeleton for same) that you would like to see?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"86cdc3754fc7b4dabef03b5def057e46da530daf","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_a795c133","line":67,"in_reply_to":"3a50d1a3_497096a4","updated":"2015-07-28 00:20:36.000000000","message":"That\u0027s certainly doable.  How about something like\n\"Project Calico is the first public implimentation of the routed network model.  You can study the details of its implementation \u003chere\u003e\"?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"e84b8c0b97d3c04543d9d04081d4888ca9b514a7","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_06883ec4","line":67,"in_reply_to":"3a50d1a3_6fc0fd50","updated":"2015-07-27 21:06:58.000000000","message":"Ryan - I\u0027ve thought about this a bit more, and think that this might confuse things more than it answers them.  What you are suggesting is related to Calico\u0027s implementation.  We view this as a more generic framework.  Using these models, someone could implement a routed network using snabbswitch, OVS, guest VM router, or even SR-IOV to an L3 switch.  None of which require what you are suggesting.\n\nWhat would be clarifying by this addition?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"d45ce6f5efe99dc82c9d322097da4add4ae242e1","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_662394be","line":67,"in_reply_to":"3a50d1a3_787573c9","updated":"2015-07-20 17:47:50.000000000","message":"It\u0027s true that the current agent implementation is very Linux-based.  But the datapath concept isn\u0027t, so in principle I think the agent could be ported to other host types, even Windows.\n\nIn practice I imagine we\u0027d be demand-driven here.  So far I believe all of our current partners for this work are assuming Linux/KVM, but if that changed we\u0027d look at supporting other hypervisors.\n\nI don\u0027t think there\u0027s anything in the API or conceptual model that prevents that, is there?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"8be4d2f2ae202a0f2953bf04633d7760636d087f","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_9ea05ca7","line":67,"in_reply_to":"3a50d1a3_787573c9","updated":"2015-07-24 22:39:23.000000000","message":"Neil\u0027s response is pretty spot on here.  However, there are two questions here, actually.  The first is \"how does this model work with other hypervisors.\"  This model supports any hypervisor OS that supports the ability to create virtual interfaces of some type, and bond VM\u0027s, containers, or processes to those interfaces.  The other support that is required is the ability to manipulate and inspect, via an API, the routing and ACL tables of that OS.  Of hypervisor/OS combinations you mentioned Kyle, most of them support the necessary capabilities, but not all.\n\nThe second part of that question may be \"what does Calico support?\"  While not directly relevant to the spec, I would say that most of our experience, to date, has been on Linux/KVM.  We believe that there would be little, if any work required to support Linux/Xen (based on our implementation of the mechanics).   We also believe that Hyper-V would be possible.   All of this is driven by community demand (or code contributions), of course.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"88587e1a181c83398a84da8061bb31e740267e83","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_2f06a320","line":67,"in_reply_to":"3a50d1a3_9ea05ca7","updated":"2015-07-26 23:02:02.000000000","message":"I think it would be fair to point out these facts in the devref (truth in advertising, as it were)","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"f304f90197a613147c372d294250315009787c03","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_6fc0fd50","line":67,"in_reply_to":"3a50d1a3_a5814b91","updated":"2015-07-27 18:58:49.000000000","message":"That seems reasonable.  Neil, do you want me to take a swing at that?","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"7d5c0a80340d019aebb10f699b92604b4fb7a605","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_0b2ca297","line":67,"in_reply_to":"3a50d1a3_a795c133","updated":"2015-07-28 14:33:40.000000000","message":"+1 to that text :)","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"f8053838aaa1c06893117422c644a253036c3e33","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_d3b8d313","line":67,"in_reply_to":"3a50d1a3_c1d560e5","updated":"2015-07-27 21:47:08.000000000","message":"Would it make more sense to put that documentation in the project calico documents, as it is specific to Calico\u0027s implementation, and not specific to the general routed case we are talking about here?   I don\u0027t want devs to come away thinking that that is required for all routed network implementations, as it isn\u0027t.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":11682,"name":"Ryan Moats","email":"rmoats@us.ibm.com","username":"regXboi"},"change_message_id":"d0ba6102c04a477f194b01b3334a7dec4dd2295d","unresolved":false,"context_lines":[{"line_number":64,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":65,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":66,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"},{"line_number":67,"context_line":"layer."},{"line_number":68,"context_line":""},{"line_number":69,"context_line":"For each local VM, the host programs a route to that VM\u0027s IP"},{"line_number":70,"context_line":"address(es) through the relevant TAP device.  The host also runs a BGP"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_497096a4","line":67,"in_reply_to":"3a50d1a3_d3b8d313","updated":"2015-07-27 22:47:46.000000000","message":"A reference to the calico documentation works for the first part.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"8be4d2f2ae202a0f2953bf04633d7760636d087f","unresolved":false,"context_lines":[{"line_number":109,"context_line":"- The TAP devices for ports in a private address scope are moved into"},{"line_number":110,"context_line":"  the corresponding namespace, on the host side."},{"line_number":111,"context_line":""},{"line_number":112,"context_line":"- Some tunneling or overlay technology is used to connect those"},{"line_number":113,"context_line":"  namespaces, between participating compute hosts.  Options here"},{"line_number":114,"context_line":"  include 464XLAT and any of the tunneling technologies used in"},{"line_number":115,"context_line":"  Neutron L2 network types."}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_1e7c4c1f","line":112,"updated":"2015-07-24 22:39:23.000000000","message":"Strictly speaking, 464XLAT isn\u0027t a tunnel or overlay.  It\u0027s a stateless (computational) translation between v4 and v6 and back again.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":9970,"name":"John Kasperski","email":"jckasper@linux.vnet.ibm.com","username":"jckasper"},"change_message_id":"899f8ce1164cf9c58d4a2024441f1e0ebb7e5cc7","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_c9710646","line":136,"updated":"2015-07-22 03:18:16.000000000","message":"Seems like what is being proposed here is for some additional core neutron changes to be made specifically for the Calico project.  I think that either: \n\n(1) these neutron API and DHCP core changes should also be kept out of tree and with the Calico project   OR \n\n(2) if OpenStack operators are interested in or have been requesting this type function, then all of Calico should be brought into core Neutron.\n\nMy 2 cents","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":17538,"name":"Ramiro Salas","email":"rsalas@pivotal.io","username":"rsalas"},"change_message_id":"cfa87bdd6f740e285506795aa16ec14a00c86dc3","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_c9d586e5","line":136,"in_reply_to":"3a50d1a3_40ae3599","updated":"2015-07-27 23:21:57.000000000","message":"I have to agree with Christopher, Gregory and Nick here. Working on the PaaS side of things (Cloud Foundry), it\u0027s becoming increasingly important to find a way to create homogeneous ways for container-to-container AND container-to-VM communications across multiple technologies. Each model tends to be very opinionated about the \"best\" way to achieve this. L3, OTOH, serves as the logical, vendor neutral glue, effectively making it a low-hanging-fruit for all sort of communications across multi-vendor technologies. \nIn a cloud-native, microservices-driven model there are no hard requirements for L2 adjacency. I\u0027m certain there are use cases when you want it, but plenty other when you just don\u0027t care. From that perspective, by Neutron providing a native L3 model, it would help greatly the OpenStack project as a whole to remain a core substrate for container-based uses cases.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":9879,"name":"Gregory Elkinbard","email":"gelkinbard@mirantis.com"},"change_message_id":"45046571af941937d9142b5977fd0648b4f1be04","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_d7fd666f","line":136,"in_reply_to":"3a50d1a3_b3dace49","updated":"2015-07-25 00:35:35.000000000","message":"Calico is not the only use case. I am aware of at least one more company that is building similar function, just much further away and we have custom built such solutions for the clients in the past. There is actually more apps that do not need L2, then the ones that do. Most modern apps written as microservices make no L2 assumptions. As long as SDN can bring in a verifiable security model, it does not really have to tunnel the packets. Sure there are some apps, that would not run on this type of SDN, but many would.  In both of the NFV and Enterprise use cases this solution would be beneficial.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":14889,"name":"Christopher LILJENSTOLPE","email":"launchpad@cdl.asgaard.org"},"change_message_id":"8be4d2f2ae202a0f2953bf04633d7760636d087f","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_fee1609b","line":136,"in_reply_to":"3a50d1a3_b3dace49","updated":"2015-07-24 22:39:23.000000000","message":"You are correct, Paul.  When we\u0027ve been out talking to folks about Calico, we\u0027ve found a number of large end-user organizations that have built something akin to Calico internally.  Some are OpenStack/Neutron users, some are OpenStack/Nova (quoting the impedence mismatch here as one of the reasons why they haven\u0027t moved), and some are home-grown orchestrators.\nSimilarly, as OpenStack moves to support containers, we will find that most folks in the Container world have very little interest in slinging L2 segments between very large numbers of ephemeral containers.  Most get (and like) the L3 native model.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":7776,"name":"Paul Carver","email":"pcarver@att.com","username":"pcarver"},"change_message_id":"0cdd82bb3b13f94b4479ae5c098d8bdb91f48c33","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_b3dace49","line":136,"in_reply_to":"3a50d1a3_c9710646","updated":"2015-07-22 12:40:13.000000000","message":"I\u0027m not sure that Calico is the only use case, although it may be the only one ready to go. Routed underlay networks are highly desirable from a scalability perspective (and I assume that was a large part of the motivation for Neutron adding GRE and VXLAN networks) but it isn\u0027t clear that we always need to emulate an L2 segment. Especially for NFV use cases, L3 connectivity may be a more natural expectation. For example, if a VM is performing a routing or security function it may have no need for L2 semantics (i.e. \"no switchport\" in Cisco parlance). It would be good to get some input from Open Daylight experts on whether the description in this devref is too specific to Calico or whether it can accommodate other L3 based underlay implementations.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":17594,"name":"Bjorn Nelson","email":"bjorn@belovedarctos.com","username":"osleep"},"change_message_id":"2ea5117fc54c1f1372c11a9f7f0a81fd08616f6c","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_cb1d3e74","line":136,"in_reply_to":"3a50d1a3_c9d586e5","updated":"2015-07-29 02:18:38.000000000","message":"From a customer perspective, we have been using a very similar implementation to what Project Calico does for years.  It is very friendly to big enterprises that have services like nfs and kerberos, specifically because it doesn\u0027t use nat and floating ips.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":1387,"name":"Nick Bartos","email":"nibartos@cisco.com","username":"bartos"},"change_message_id":"fbb7a4fefda26b4c0d4c10527bbf36dcc78fe5ce","unresolved":false,"context_lines":[{"line_number":133,"context_line":"  for a \u0027routed\u0027 network as described above.  This work is out-of-tree"},{"line_number":134,"context_line":"  at https://github.com/Metaswitch/calico and is already fairly"},{"line_number":135,"context_line":"  mature, although it will need some minor adaptation for how the"},{"line_number":136,"context_line":"  in-tree pieces (above) end up."},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"Further implementation work will be needed for the private case, and"},{"line_number":139,"context_line":"for IP multicast, but we propose to cover those in separate future"}],"source_content_type":"text/x-rst","patch_set":3,"id":"3a50d1a3_40ae3599","line":136,"in_reply_to":"3a50d1a3_d7fd666f","updated":"2015-07-27 17:38:55.000000000","message":"I agree with what Christopher and Gregory are saying: This is not just a Calico requirement, Neutron needs better support for the native L3 model.  Having support for this natively in Neutron would help us quite a bit in on the Cisco Intercloud project, by dramatically reducing patching to Neutron itself.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":1387,"name":"Nick Bartos","email":"nibartos@cisco.com","username":"bartos"},"change_message_id":"5914d34a82d06e1460ffcf6baf7e0f84ef5e7991","unresolved":false,"context_lines":[{"line_number":53,"context_line":"A routed network can be shared or private; this is indicated as usual"},{"line_number":54,"context_line":"by presence or absence of the \u0027shared\u0027 flag."},{"line_number":55,"context_line":""},{"line_number":56,"context_line":"As normal in Neutron, \u0027private\u0027 primarily means that that network is"},{"line_number":57,"context_line":"only available for use by the tenant that created it.  It may also"},{"line_number":58,"context_line":"mean that the network uses IP addressing in a non-default address"},{"line_number":59,"context_line":"scope - typically private to the tenant; this permits its IP addresses"}],"source_content_type":"text/x-rst","patch_set":4,"id":"3a50d1a3_745b8122","line":56,"updated":"2015-07-29 17:07:10.000000000","message":"Small nitpick: \"As is normal\" sounds better than \"As normal\" here.  It\u0027s probably not worth changing unless you\u0027re changing else at the same time.","commit_id":"c2f27ce9c43f17c266cd0158eea3558dd704ecd9"},{"author":{"_account_id":7776,"name":"Paul Carver","email":"pcarver@att.com","username":"pcarver"},"change_message_id":"45ff8afe985a99c322bb00361e43488b0f6031a7","unresolved":false,"context_lines":[{"line_number":70,"context_line":"L2-based network types, as the latter would require an explicit"},{"line_number":71,"context_line":"Neutron router object, for mutual or external connectivity."},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"  Note that with this automatic connectivity, it is still easy for a"},{"line_number":74,"context_line":"  particular tenant to get effective isolation for its own group of"},{"line_number":75,"context_line":"  VMs.  The tenant just needs to create its own security group, and"},{"line_number":76,"context_line":"  use that security group when launching its own instances."}],"source_content_type":"text/x-rst","patch_set":4,"id":"3a50d1a3_6e42f869","line":73,"updated":"2015-07-29 02:31:53.000000000","message":"You may want to use the .. note:: syntax to format this as a note in the rendered document.","commit_id":"c2f27ce9c43f17c266cd0158eea3558dd704ecd9"},{"author":{"_account_id":16481,"name":"debedb","email":"grisha@alum.mit.edu","username":"debedb"},"change_message_id":"b26791c423b5fd54e9ca44da6b986d87588147a4","unresolved":false,"context_lines":[{"line_number":92,"context_line":""},{"line_number":93,"context_line":"A VM is \u0027plugged\u0027 with a TAP device on the host that connects to the"},{"line_number":94,"context_line":"VM\u0027s network stack.  The host end of the TAP is left unbridged and"},{"line_number":95,"context_line":"without any IP addresses (except for link-local IPv6).  The host is"},{"line_number":96,"context_line":"configured to respond to any ARP or NDP requests, through that TAP,"},{"line_number":97,"context_line":"with its own MAC address; hence data arriving through the TAP is"},{"line_number":98,"context_line":"always addressed at L2 to the host, and is passed to the Linux routing"}],"source_content_type":"text/x-rst","patch_set":4,"id":"3a50d1a3_660f142f","line":95,"updated":"2015-07-31 20:53:35.000000000","message":"Where is this occurring \"The host is configured to respond to any ARP or NDP requests, through that TAP, with its own MAC address\". Is this just a side effect of adding the dummy interface and bridging it to the tap?","commit_id":"c2f27ce9c43f17c266cd0158eea3558dd704ecd9"},{"author":{"_account_id":748,"name":"Armando Migliaccio","email":"armamig@gmail.com","username":"armando-migliaccio"},"change_message_id":"863b1a04aa375d5d422bc06a204e4f157dbfadb7","unresolved":false,"context_lines":[{"line_number":37,"context_line":"------------------------------"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"A routed network is described on the Neutron API using a new provider"},{"line_number":40,"context_line":"network type: TYPE_ROUTED \u003d \u0027routed\u0027.  The related physical network"},{"line_number":41,"context_line":"and segmentation ID parameters are not meaningful and should be left"},{"line_number":42,"context_line":"unspecified."},{"line_number":43,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"da20952f_f65d397a","line":40,"updated":"2015-09-01 17:23:12.000000000","message":"I don\u0027t seem to see this type of driver.","commit_id":"275c6614aee53387d83c9b5eeb382507bf61b7b2"},{"author":{"_account_id":748,"name":"Armando Migliaccio","email":"armamig@gmail.com","username":"armando-migliaccio"},"change_message_id":"863b1a04aa375d5d422bc06a204e4f157dbfadb7","unresolved":false,"context_lines":[{"line_number":127,"context_line":"its --bridge-interface option.  The following patches enhance the DHCP"},{"line_number":128,"context_line":"agent and IP library to allow it to be used in this way."},{"line_number":129,"context_line":""},{"line_number":130,"context_line":"- https://review.openstack.org/205181"},{"line_number":131,"context_line":"- https://review.openstack.org/206077"},{"line_number":132,"context_line":"- https://review.openstack.org/206078"},{"line_number":133,"context_line":"- https://review.openstack.org/206079"}],"source_content_type":"text/x-rst","patch_set":5,"id":"da20952f_d6e595dc","line":130,"updated":"2015-09-01 17:23:12.000000000","message":"Not sure about the value of showing patches within a devref. I\u0027d suggest you to remove them.","commit_id":"275c6614aee53387d83c9b5eeb382507bf61b7b2"}],"neutron/plugins/ml2/drivers/type_routed.py":[{"author":{"_account_id":7776,"name":"Paul Carver","email":"pcarver@att.com","username":"pcarver"},"change_message_id":"0cdd82bb3b13f94b4479ae5c098d8bdb91f48c33","unresolved":false,"context_lines":[{"line_number":47,"context_line":""},{"line_number":48,"context_line":"    def validate_provider_segment(self, segment):"},{"line_number":49,"context_line":"        for key, value in six.iteritems(segment):"},{"line_number":50,"context_line":"            if value and key !\u003d api.NETWORK_TYPE:"},{"line_number":51,"context_line":"                msg \u003d _(\"%s prohibited for routed provider network\") % key"},{"line_number":52,"context_line":"                raise exc.InvalidInput(error_message\u003dmsg)"},{"line_number":53,"context_line":""}],"source_content_type":"text/x-python","patch_set":3,"id":"3a50d1a3_93f7928a","line":50,"updated":"2015-07-22 12:40:13.000000000","message":"Should this be \"if value is not None and key !\u003d api.NETWORK_TYPE\"  ?  Unless value is a boolean it\u0027s clearer to explicitly test for None rather than make an implicit cast to boolean. On a broader note, is the intent here to say that a for a routed network there will never be ANY valid attributes other than the fact that it is routed? This is probably ok for now, but I think it\u0027s possible to come up with a clearer expression of prohibited vs permitted attributes for a given NETWORK_TYPE.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"},{"author":{"_account_id":13734,"name":"Nell Jerram","email":"nell@tigera.io","username":"neiljerram"},"change_message_id":"c5ccaf01767535ace3a7ebcb35baf221d8bbb040","unresolved":false,"context_lines":[{"line_number":47,"context_line":""},{"line_number":48,"context_line":"    def validate_provider_segment(self, segment):"},{"line_number":49,"context_line":"        for key, value in six.iteritems(segment):"},{"line_number":50,"context_line":"            if value and key !\u003d api.NETWORK_TYPE:"},{"line_number":51,"context_line":"                msg \u003d _(\"%s prohibited for routed provider network\") % key"},{"line_number":52,"context_line":"                raise exc.InvalidInput(error_message\u003dmsg)"},{"line_number":53,"context_line":""}],"source_content_type":"text/x-python","patch_set":3,"id":"3a50d1a3_3700e301","line":50,"in_reply_to":"3a50d1a3_93f7928a","updated":"2015-07-28 23:07:32.000000000","message":"Regarding both points, I was just copying LocalTypeDriver in type_local.py.\n\nI know that\u0027s not the best argument - but consistency is also a virtue, when we\u0027re talking of slightly more stylistic points such as these.\n\nHowever I\u0027m actually going to remove all of the code from the review job, in my next upload, and leave just the devref doc.  In the light of ongoing discussions about the best way to model routed networking in Neutron, it really seems premature to propose a specific code implementation.","commit_id":"fbd05ea0f57e356685fe80770ddacba37c0d1b05"}]}
