| zhenguo | morning mogan! | 00:45 |
|---|---|---|
| *** wanghao has joined #openstack-mogan | 00:48 | |
| *** openstackgerrit has joined #openstack-mogan | 01:00 | |
| openstackgerrit | Zhenguo Niu proposed openstack/mogan master: Make floatingip can be saved to db and can be displayed https://review.openstack.org/475622 | 01:00 |
| *** litao__ has joined #openstack-mogan | 01:13 | |
| *** liujiong has joined #openstack-mogan | 01:14 | |
| *** wanghao has quit IRC | 01:17 | |
| *** wanghao has joined #openstack-mogan | 01:18 | |
| zhenguo | litao__: sorry, I was out of office yesterday | 01:19 |
| litao__ | zhenguo: np, you can review my patch again | 01:22 |
| zhenguo | litao__: ok | 01:22 |
| openstackgerrit | wanghao proposed openstack/mogan master: Refactor exception raise by using mogan exception https://review.openstack.org/475971 | 01:50 |
| liusheng | zhenguo: what do you think we need to research and try to use placement to implement node aggregates now ? | 01:55 |
| zhenguo | liusheng: I talked with alex about the aggregates traits | 01:56 |
| zhenguo | liusheng: he said we can list all rps and set traits to it for now | 01:57 |
| liusheng | zhenguo: yes, sure, I have try some apis about placement | 01:58 |
| zhenguo | liusheng: great | 01:58 |
| liusheng | zhenguo: but I have a doubt that how to set the aggretates, does we need to add a api to call placement to set the rp trats | 01:59 |
| zhenguo | liusheng: no, I don't like proxy api | 01:59 |
| liusheng | zhenguo: hah, yes | 01:59 |
| zhenguo | liusheng: admins should call placement api or CLI to manage resources | 01:59 |
| liusheng | zhenguo: yes, now there is a placement-osc repo in git.openstack.org. but only a base frame | 02:01 |
| zhenguo | liusheng: yes, we can help to enrich it | 02:01 |
| zhenguo | liusheng: only for some that we need, lol | 02:02 |
| liusheng | zhenguo: really, we need to improve the placement osc firstly ? | 02:02 |
| liusheng | zhenguo: hah | 02:02 |
| zhenguo | liusheng: no rush | 02:02 |
| liusheng | zhenguo: ok, we can try to improve mogan now, to use node aggregates traits | 02:03 |
| zhenguo | liusheng: yes | 02:03 |
| zhenguo | liusheng: the placement work on mogan side is almost done, right? | 02:06 |
| zhenguo | liusheng: if so, please also help to enrich the specs :D | 02:06 |
| liusheng | zhenguo: yes, but didn't consider traits yet | 02:07 |
| liusheng | zhenguo: ok | 02:07 |
| zhenguo | liusheng: traits should be done with another specs | 02:07 |
| zhenguo | liusheng: the nodes aggregates one | 02:07 |
| liusheng | zhenguo: ok | 02:07 |
| liusheng | zhenguo: how about only enable placement service without nova in our gate ? | 02:18 |
| zhenguo | liusheng: I remember there's something worong with the tempest if we not enable nova | 02:19 |
| zhenguo | liusheng: not sure if you can succeed to just enable placement | 02:19 |
| liusheng | zhenguo: oh, yes, seems it need the "baremetal" flaovr of Nova | 02:19 |
| zhenguo | liusheng: hah | 02:19 |
| liusheng | zhenguo: we'd better to try to get rid of Nova, play mogan standalone :D | 02:20 |
| zhenguo | liusheng: the problem is ironic depends on nova with tempest | 02:21 |
| liusheng | zhenguo: not sure if we can fix that | 02:21 |
| zhenguo | liusheng: seems not now, but for gate, it's ok to run them together, | 02:22 |
| liusheng | zhenguo: ok, we can try in future | 02:23 |
| zhenguo | liusheng:ｙｅｓ | 02:23 |
| zhenguo | liusheng: I left a comment here https://review.openstack.org/#/c/476325/16 | 02:29 |
| zhenguo | liusheng: if you will delete the RP when node is unavailable, why not just keep the original get_avaialbe_nodes | 02:31 |
| liusheng | zhenguo: actually, the unavailable != not available, hah | 02:31 |
| liusheng | zhenguo: it is not similar with previous implementation | 02:32 |
| zhenguo | liusheng: you mean more filter options? | 02:32 |
| liusheng | zhenguo: the unavailable nodes mean the nodes in abnormal states, cannot be provisioned | 02:33 |
| liusheng | zhenguo: addtionally | 02:33 |
| liusheng | zhenguo: I planed to treat the unavailable nodes and fake nodes in placement in different ways | 02:34 |
| zhenguo | liusheng: how | 02:34 |
| zhenguo | liusheng: as far as I can see, you will just delete the rp | 02:34 |
| liusheng | zhenguo: for the fake nodes(nodes in placement, but have gone in ironic), should be removed from placement, with the allocations inventories | 02:34 |
| liusheng | zhenguo: for the unavailable nodes, we should only delete the resource providers, without allocatios deleted | 02:35 |
| liusheng | zhenguo: that because placement now didn't provide a interface to retrive allocations by a resource provider | 02:36 |
| litao__ | zhenguo: I pull the latest code and mogan raised the db error: "Field 'port_type' doesn't have a default value" | 02:36 |
| liusheng | zhenguo: wdyt ? | 02:36 |
| zhenguo | liusheng: need to restack | 02:36 |
| zhenguo | liusheng: didn't get | 02:37 |
| litao__ | zhenguo: what is the default value of port_type？ | 02:37 |
| zhenguo | liusheng: we have removed it | 02:37 |
| zhenguo | liusheng: from db | 02:37 |
| zhenguo | litao__ | 02:37 |
| litao__ | zhenguo：yes | 02:38 |
| liusheng | zhenguo: yes, because I think the unavailable node should be seleted by scheduler | 02:38 |
| zhenguo | liusheng: why unavailable node should be selected? | 02:38 |
| zhenguo | liusheng: it can't be provisoned | 02:38 |
| liusheng | zhenguo: but maybe a abonormal node can be recover in rionic | 02:38 |
| liusheng | zhenguo: sorry, shouldn't | 02:38 |
| zhenguo | liusheng: what's the benefit | 02:39 |
| liusheng | zhenguo: so for the abnormal nodes (nodes in abnormal state) should keep the allocations in placement | 02:39 |
| zhenguo | liusheng: why should it be kept | 02:40 |
| zhenguo | liusheng: seems in placement we only need to track available and deployed nodes, all others should be removed | 02:42 |
| liusheng | zhenguo: imagine this, a server running on a node, but because some unexpected reason, the node truned into maintance state, we should keep the allocations in placement, (but seems for this situation we don't need to delete resource provide, since one node can only used by one server), once the node recoverd in ironic, periodic task will report the resource provider to placement | 02:43 |
| liusheng | zhenguo: yes | 02:43 |
| liusheng | zhenguo: that's why unavailable != not available | 02:43 |
| liusheng | zhenguo: or, maybe we should mark the nodes in abnormal state shouldn't be seleted in placement, but there isn't a mechanism to avoid seleting node with specific tratis. | 02:44 |
| zhenguo | liusheng: currently, when you call delete_resource_provider, it will keep the allocations? | 02:47 |
| liusheng | zhenguo: yes, there is a TODO in "def delete_resource_provider", and there is a cascade=False paremeter to perform that | 02:48 |
| liusheng | zhenguo: but now, it is need to provider the consumer_id(server_id) to delete allocations | 02:49 |
| zhenguo | liusheng: let me check it | 02:50 |
| liusheng | zhenguo: may it better to implement the interface of "delete allocations of a rp" in placement | 02:50 |
| zhenguo | liusheng: my interpretation is that you will keep all allocations presently, because if some deployed node goes to maintence and come back again, it can't be scheduled to rebuild, right? | 02:57 |
| liusheng | zhenguo: yes, even the provisioned nodes cannot by selected by currect placement, but we can treat the provisioned and unprovisioned nodes together | 03:01 |
| zhenguo | liusheng: but we can only deploy the nodes in 'available' state | 03:01 |
| liusheng | zhenguo: for the fake nodes in placement, which means node maybe removed from ironic, we should remove all the info of node in placement | 03:01 |
| liusheng | zhenguo: yes, but placement record all the nodes | 03:03 |
| zhenguo | liusheng: for nodes in other states other than available should all can't be selected | 03:03 |
| zhenguo | liusheng: the only way to stop it selected it's to set the allocations, right? | 03:04 |
| liusheng | zhenguo: so what's your suggestion ? | 03:04 |
| zhenguo | liusheng: set abnormal state nodes allocations? | 03:04 |
| liusheng | zhenguo: that is right based on current implementation | 03:05 |
| zhenguo | liusheng: seems currently we just delete resource providers for both nodes not list from ironic and in abnormal state | 03:05 |
| liusheng | zhenguo: both we and nova have the requirement in placement to provide an interface about "avoiding select nodes with specific tratis" | 03:05 |
| liusheng | zhenguo: yes | 03:06 |
| liusheng | zhenguo: actually in ironic driver in Nova, it ignore this, it just report all the nodes to placement, that I think is a bug | 03:06 |
| zhenguo | liusheng: you mean we should set a STATE trait to resource provider and filter it? | 03:07 |
| liusheng | zhenguo: yes, filter out the rps with specific traits | 03:07 |
| zhenguo | liusheng: noav ironic driver will report the node as 0 CPU, 0 MEM for abnormal nodes | 03:08 |
| liusheng | zhenguo: no, if there is "resource_class" setted of the nodes, nova will report inventories like we done | 03:09 |
| zhenguo | liusheng: but not scheduled based on that now | 03:09 |
| zhenguo | liusheng: it also report the VCPU, RAM | 03:10 |
| zhenguo | liusheng: as flavor in nova can't pass resource class information now | 03:10 |
| liusheng | zhenguo: oh, yes | 03:11 |
| zhenguo | liusheng: seems it's tricky to handle nodes in abnormal state | 03:13 |
| liusheng | zhenguo: but for us, there is only one way the scheduler can avoid seleting a node is removing the rp in placement or the nodes already allocated | 03:13 |
| *** Xinran has quit IRC | 03:13 | |
| zhenguo | liusheng:ｙｅｓ | 03:14 |
| liusheng | zhenguo: I have tried to set the "total" field in inventory to be 0, but placement api don't allow that | 03:14 |
| liusheng | zhenguo: maybe Nova guys didn't test that, the total in inventory body can not be set to 0 | 03:17 |
| zhenguo | liusheng: maybe that's what expected | 03:18 |
| zhenguo | liusheng: allocations means to do that | 03:18 |
| liusheng | zhenguo: https://github.com/openstack/nova/blob/master/nova/api/openstack/placement/handlers/inventory.py#L39 | 03:18 |
| zhenguo | liusheng: my concern about the current list all nodes is that | 03:19 |
| zhenguo | liusheng: we set detail=True, and limit =0 | 03:19 |
| zhenguo | liusheng: if we have many nodes backend, the API response body will be very huge | 03:19 |
| liusheng | zhenguo: yes, how about improve that in future with pagination query | 03:20 |
| zhenguo | liusheng: sounds good | 03:20 |
| zhenguo | liusheng: for allocation, I think it's ok to always keep it until we delete the server | 03:23 |
| zhenguo | liusheng: no need to treat the nodes differently | 03:23 |
| zhenguo | liusheng: it make sense to delete allocation when delete server | 03:23 |
| liusheng | zhenguo: delete sever should delete the allocation | 03:24 |
| zhenguo | liusheng: so it's ok | 03:24 |
| liusheng | zhenguo: that have impment in the patches, hah | 03:24 |
| zhenguo | liusheng: ok, so for my comments | 03:24 |
| zhenguo | liusheng: I think we don't need to treat nodes unavailble and nodes not in list differently | 03:25 |
| zhenguo | liusheng: just one get_available_interface from driver, and remove rp not in that is ok | 03:25 |
| liusheng | zhenguo: ok, will modify that | 03:26 |
| zhenguo | liusheng: ok, thanks | 03:27 |
| zhenguo | liusheng: and we can move maintenance filter to node list API | 03:27 |
| liusheng | zhenguo: ok | 03:27 |
| openstackgerrit | Merged openstack/mogan-specs master: Track resources using Placement https://review.openstack.org/475700 | 04:03 |
| openstackgerrit | wanghao proposed openstack/mogan master: Refactor exception raise by using mogan exception https://review.openstack.org/475971 | 04:27 |
| litao__ | liusheng, zhenguo | 06:01 |
| litao__ | liusheng, zhenguo : we have removed port_type, but mogan raised 'not found port_type' error in db query. | 06:02 |
| zhenguo | liutao__: a bug? | 06:08 |
| zhenguo | litao__ | 06:08 |
| litao__ | zhenguo: I solve it by | 06:09 |
| litao__ | restarting mogan-scheduler service | 06:09 |
| litao__ | zhenguo: hah | 06:09 |
| zhenguo | litao__: so, it's not an issue, right? | 06:09 |
| litao__ | zhenguo: yes | 06:10 |
| zhenguo | litao__: ok | 06:10 |
| openstackgerrit | Jeremy Liu proposed openstack/mogan-ui master: Drop MANIFEST.in - it's not needed by pbr https://review.openstack.org/480361 | 06:14 |
| openstackgerrit | liusheng proposed openstack/mogan master: Clean the compute_node and compute_port objects and db interfaces https://review.openstack.org/478406 | 06:19 |
| openstackgerrit | liusheng proposed openstack/mogan master: Get rid of listing availability zone api and clean some legacy code https://review.openstack.org/478403 | 06:19 |
| openstackgerrit | liusheng proposed openstack/mogan master: Reporting nodes resource to placement service https://review.openstack.org/476325 | 06:19 |
| openstackgerrit | liusheng proposed openstack/mogan master: Clean the methods about updating node resources to Mogan db https://review.openstack.org/478357 | 06:19 |
| openstackgerrit | liusheng proposed openstack/mogan master: Get rid of node listing api of Mogan https://review.openstack.org/478361 | 06:19 |
| openstackgerrit | liusheng proposed openstack/mogan master: Consume nodes resource in placement https://review.openstack.org/477826 | 06:19 |
| openstackgerrit | liusheng proposed openstack/mogan master: Refactor the scheduler to use placement service https://review.openstack.org/477426 | 06:19 |
| openstackgerrit | Merged openstack/mogan-ui master: Drop MANIFEST.in - it's not needed by pbr https://review.openstack.org/480361 | 06:37 |
| openstackgerrit | Tao Li proposed openstack/mogan master: Fix reverting the network when plug vif failed https://review.openstack.org/477440 | 06:46 |
| *** Xinran has joined #openstack-mogan | 07:57 | |
| *** Xinran has quit IRC | 07:58 | |
| *** Xinran has joined #openstack-mogan | 07:59 | |
| openstackgerrit | wanghao proposed openstack/mogan master: Manage existing BMs: Part-1 https://review.openstack.org/479660 | 08:18 |
| liusheng | zhenguo: hi zhenguo, do you think if it is easy to misleading if we use "available resource" > | 08:41 |
| zhenguo | liusheng: seems it's ok | 08:50 |
| zhenguo | liusheng: it's a driver interface | 08:50 |
| liusheng | zhenguo: hah, I tried to avoid that misleading | 08:51 |
| zhenguo | liusheng: we call that interface to get avalable resources for scheduling | 08:51 |
| liusheng | zhenguo: ok, | 08:51 |
| zhenguo | liusheng: hah | 08:51 |
| liusheng | zhenguo: do you prefer get_available_resources than get_available_nodes ? | 08:54 |
| zhenguo | liusheng: I'm fine with both, | 08:54 |
| zhenguo | liusheng: maybe we need to get other resources like ports, harddrives, so maybe get_availble_resources is better? | 08:55 |
| liusheng | zhenguo: nova call it get_available_nodes, hah | 08:56 |
| zhenguo | liusheng: that's because nova only get node resources, hah | 08:56 |
| liusheng | zhenguo: ok, will update patch as your wish sir. lol | 09:00 |
| zhenguo | liusheng: thanks, hah | 09:00 |
| openstackgerrit | liusheng proposed openstack/mogan master: Clean the compute_node and compute_port objects and db interfaces https://review.openstack.org/478406 | 09:09 |
| openstackgerrit | liusheng proposed openstack/mogan master: Get rid of listing availability zone api and clean some legacy code https://review.openstack.org/478403 | 09:09 |
| openstackgerrit | liusheng proposed openstack/mogan master: Reporting nodes resource to placement service https://review.openstack.org/476325 | 09:09 |
| openstackgerrit | liusheng proposed openstack/mogan master: Clean the methods about updating node resources to Mogan db https://review.openstack.org/478357 | 09:09 |
| openstackgerrit | liusheng proposed openstack/mogan master: Get rid of node listing api of Mogan https://review.openstack.org/478361 | 09:09 |
| openstackgerrit | liusheng proposed openstack/mogan master: Consume nodes resource in placement https://review.openstack.org/477826 | 09:09 |
| openstackgerrit | liusheng proposed openstack/mogan master: Refactor the scheduler to use placement service https://review.openstack.org/477426 | 09:09 |
| zhenguo | liusheng: will it break the gate if just replace the get_availalbe_resources implementation | 09:12 |
| zhenguo | liusheng: as the scheduler still depends on the original get_available_resources method | 09:12 |
| liusheng | zhenguo: hmmm, yes.. | 09:13 |
| zhenguo | liusheng: hah, maybe you can just use get_available_nodes .. | 09:13 |
| zhenguo | liusheng: at least in Pike we will not introduce new resources | 09:13 |
| liusheng | zhenguo: I also considered that, so let name it as get_available_nodes, and we can change it in flowing patch if we need | 09:14 |
| zhenguo | liusheng: yes | 09:14 |
| openstackgerrit | liusheng proposed openstack/mogan master: Clean the compute_node and compute_port objects and db interfaces https://review.openstack.org/478406 | 09:21 |
| openstackgerrit | liusheng proposed openstack/mogan master: Get rid of listing availability zone api and clean some legacy code https://review.openstack.org/478403 | 09:21 |
| openstackgerrit | liusheng proposed openstack/mogan master: Reporting nodes resource to placement service https://review.openstack.org/476325 | 09:21 |
| openstackgerrit | liusheng proposed openstack/mogan master: Clean the methods about updating node resources to Mogan db https://review.openstack.org/478357 | 09:21 |
| openstackgerrit | liusheng proposed openstack/mogan master: Get rid of node listing api of Mogan https://review.openstack.org/478361 | 09:21 |
| openstackgerrit | liusheng proposed openstack/mogan master: Consume nodes resource in placement https://review.openstack.org/477826 | 09:21 |
| openstackgerrit | liusheng proposed openstack/mogan master: Refactor the scheduler to use placement service https://review.openstack.org/477426 | 09:21 |
| *** wanghao_ has joined #openstack-mogan | 09:23 | |
| *** wanghao_ has quit IRC | 09:24 | |
| *** wanghao has quit IRC | 09:26 | |
| * zhenguo brb | 09:39 | |
| *** liujiong has quit IRC | 10:07 | |
| shaohe_feng | https://review.openstack.org/#/c/478406/18..15//COMMIT_MSG | 10:29 |
| shaohe_feng | ^ liusheng: first line need to fix? Why need to fix? | 10:30 |
| zhenguo | shaohe_feng: seems the recommended length should be less than 50 | 10:48 |
| shaohe_feng | zhenguo: got it. | 10:52 |
| zhenguo | shaohe_feng: but not compulsory | 10:52 |
| openstackgerrit | Zhenguo Niu proposed openstack/mogan master: Reporting nodes resource to placement service https://review.openstack.org/476325 | 11:21 |
| zhenguo | shaohe_feng: are you still around? | 11:22 |
| zhenguo | shaohe_feng: please have a look at this https://review.openstack.org/#/c/475302/ and https://review.openstack.org/#/c/475622/ when you got time | 11:22 |
| zhenguo | shaohe_feng: and please help to review the first placement patch https://review.openstack.org/#/c/476325/ , let's try to land placement stuff by this week. | 11:31 |
| shaohe_feng | zhenguo: OK. | 11:34 |
| zhenguo | shaohe_feng: thanks | 11:34 |
| shaohe_feng | really good we can land placement stuff by this week. | 11:34 |
| zhenguo | shaohe_feng: on my test, it works well now | 11:35 |
| zhenguo | shaohe_feng: we need to leave enough time to prepare the first release | 11:35 |
| *** dims has quit IRC | 12:29 | |
| *** dims has joined #openstack-mogan | 12:44 | |
| *** wanghao has joined #openstack-mogan | 14:10 | |
| openstackgerrit | Wangliangyu proposed openstack/python-moganclient master: Drop MANIFEST.in - it's not needed by pbr https://review.openstack.org/480637 | 15:02 |
| *** liujiong has joined #openstack-mogan | 15:03 | |
| *** liujiong has quit IRC | 15:04 | |
| *** wanghao has quit IRC | 16:10 | |
| *** wanghao has joined #openstack-mogan | 16:11 | |
| *** wanghao has quit IRC | 16:11 | |
| *** wanghao has joined #openstack-mogan | 16:11 | |
| *** litao__ has quit IRC | 18:17 | |
| *** wanghao has quit IRC | 22:24 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!