17:00:55 #startmeeting Rally 17:00:55 Meeting started Tue Jan 21 17:00:55 2014 UTC and is due to finish in 60 minutes. The chair is boris-42. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:00:56 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 17:00:58 The meeting name has been set to 'rally' 17:01:04 hi boris-42 17:01:09 hughsaunders hi 17:01:16 who is here?) 17:01:22 hi all! 17:01:24 hi 17:01:25 hi 17:01:39 hi 17:02:23 oh seems I am not alone+) I should care what I am saying =) 17:02:49 topic HOT topics 17:02:54 #topic HOT topics 17:02:57 boris-42: ~362 lurkers in this channel 17:03:09 # 17:03:10 )) 17:03:17 lol=) 17:03:41 So topic list: 17:03:58 1) Rally & Tempest integrration 17:04:17 2) Rally & Running benchmarks against cloud (without admin endpoint) 17:04:34 3) Rally & Multihost deployments 17:04:53 4) Rally & Smart benchmark scenarios 17:04:56 checkers 17:05:09 5) Rally as a Service 17:05:20 6) Improvements in Rally CLI 17:05:34 7) Rally & benchmarking performance of VMs 17:06:30 8) Rally use class instances for benchmark scenarions instead of class objects and classmethods 17:06:47 9) Rally measuring time of atomic actions 17:06:53 ^ Thats all 17:07:04 I will try to use this order in today meeting 17:07:13 #topic Rally & Tempest integration 17:07:25 miarmak pls could you share your results? 17:08:16 (btw guys/girls if you are related to any of this parts prepare your updates) 17:08:21 boris-42: unfortunately, situation has not changed from yesterday 17:08:34 main problem - conf generation 17:08:53 miarmak so you are not able to configure correct conf for tempest ^ 17:08:56 maybe anyone know projects, that have smth same? 17:09:00 probably sdague could help us? ^ 17:09:47 boris-42 mentioned that devstack configures tempest, so this may help: https://github.com/openstack-dev/devstack/blob/master/lib/tempest#L66 17:10:09 hughsaunders ohhhh 17:10:14 hughsaunders how much lines of code.. 17:10:30 boris-42: if it's not a gate issue, I can't really help you this week 17:10:46 sdague sure understand you, thank you for fixing gate! 17:10:53 I'm going to attempt to chef-ify that for the stackforge chef cookbooks project 17:11:14 hughsaunders how to generate proper config for tempest? 17:11:43 boris-42: yeah, not started yet, but planning to base it on that devstack function 17:12:01 miarmak hughsaunders could you probably collaborate around this? 17:12:35 Yeah, if I get anywhere I'll ping miarmak with a link 17:12:44 hughsaunders nice 17:12:45 #topic Rally & Running benchmarks against cloud (without admin endpoint) 17:12:48 ok 17:12:50 thanks 17:13:01 So first of all 17:13:19 #link https://blueprints.launchpad.net/rally/+spec/benchmarking-with-predefined-users 17:13:23 ^ bp 17:13:43 #link https://review.openstack.org/#/q/status:open+project:stackforge/rally+branch:master+topic:bp/benchmarking-with-predefined-users,n,z 17:13:46 patches on review ^ 17:13:54 The idea is simple 17:14:06 Now Rally work in next way 17:14:57 get admin endpoints -> create admin clients -> create tmp tenants & users -> run from temp users benchmarks -> cleanup of cloud -> delete tmp users -> FINISH 17:15:19 So it's work until we would like to test Public clouds 17:15:25 where we don't have admin endpoints 17:15:44 So we will add support of running benchmark from bunch of predefined users 17:16:09 input endpoints of non admin users -> create python clients -> run benchmarks -> make non admin cleanups 17:16:31 So if you will have some free time you can spend it reviewing that patches=) 17:16:56 Does anybody has any questions?) 17:17:39 seems like a good plan 17:18:10 stannie julienvey thoughts? 17:18:16 seems good to me also 17:18:28 okay nice then we will move just to next topic 17:18:45 #topic 3) Rally & Multihost deployments 17:19:07 #link https://blueprints.launchpad.net/rally/+spec/multihost-deploy 17:19:14 bp ^ 17:19:38 #link https://review.openstack.org/#/q/status:merged+project:stackforge/rally+branch:master+topic:bp/multihost-deploy,n,z 17:19:40 ^ patches 17:19:45 This work is already done 17:19:48 and should be merged 17:19:54 Idea is next 17:20:12 we are deploying with separated deployers/provider controller and compute/cinder nodes 17:20:37 so if we choose DevStack it will run N times .stacksh 17:20:46 which will have bad perfromance 17:21:10 so we are deploying controller + N type of Nodes only 1 time 17:21:23 (so N+1) call of stack.sh 17:21:30 and then copy pasting LXC containers 17:21:46 which allows us to deploy 1 controller + 128 compute managers for just a 30 minutes 17:22:30 boris-42: on a single underlying node? 17:22:41 it could use multiply nodes 17:22:47 or even OpenStack provider 17:22:56 or even any other provider 17:23:03 like tripple-o undercloud 17:23:09 hughsaunders but better=) 17:23:36 it doesn't required OpenStack installation 17:23:44 it will work with provider that you put 17:23:56 Dummy/OpenStack/your_custom_cloud_provider 17:24:28 so how do you determine in config how many lxc containers are created on each underlying node? 17:24:31 hughsaunders here is the instruction how to try it http://paste.ubuntu.com/6792218/ 17:24:34 on trello I saw you guys were thinking to integrate Docker, in this case it would simplify this usage. We wouldn't have to copy the LXC containers etc 17:24:48 stannie we need 17:25:06 stannie you should install from custom code your OpenStack installation 17:25:23 stannie with any of Deployers (e.g. DevStack/Anvil/Fuel) 17:25:32 stannie and then copy/paste installed containers 17:25:50 stannie and then fix all copy/pasted lxc containers 17:26:11 stannie there should be fix in nova.conf/cinder.conf/.… to add proper network information (IP/host_name) 17:26:20 stannie and run this components 17:26:20 ye and that's can be done easily with Docker 17:26:31 stannie but I don't see any problems to do it with lxc 17:26:37 ok 17:26:50 it just asks a little more work 17:26:54 stannie https://review.openstack.org/#/c/56222/ 17:27:06 stannie 150 lines of code 17:27:24 ok 17:27:38 stannie but we are as well experimenting with docker 17:27:55 stannie to make another future e.g. cloud deployment snaphsot 17:27:58 stannie to share cloud+) 17:28:14 feature& 17:28:24 stannie that is the case where docker can help us 17:29:13 stannie hughsaunders so you can try as well to use this stuff https://review.openstack.org/#/c/56222/ 17:29:26 stannie hughsaunders I will try to make installation today and merge it if it works=) 17:29:54 we should move to the next topic 17:29:56 #topic 4) Rally & Smart benchmark scenarios 17:30:06 #topic 4) Rally & Smart benchmark scenarios checkers 17:30:16 olkonami1 could you pls share with your resuls 17:30:27 olkonami1 seems like work is almost done 17:30:51 https://review.openstack.org/#/c/67157/ 17:30:52 https://review.openstack.org/#/c/68055/ 17:31:10 olkonami1 I mean could you expelling what have you done=) 17:31:15 olkonami1 in human way=) 17:31:27 olkonami1 btw you should make new blueprint 17:31:31 olkonami1 for your patches 17:32:11 I really like the image_id validator, simple and effective 17:32:46 hughsaunders yep there will be at least 2 more 17:32:53 hughsaunders for flavor checking 17:32:57 I made decorator that add validators for scenario methods 17:33:04 hughsaunders and checking that image could be use with flavor 17:34:24 and this validators should be called when runner starts 17:34:46 ye that's great, we'll be able to add more validator without adding complexity in the method 17:35:04 stannie yep and stop execution before starting it=) 17:35:15 now first validator for image_id is implemented 17:35:43 so I hope we will merge this soon 17:36:14 #topic 5) Rally as a Service 17:36:18 ogelbukh ping 17:36:47 ogelbukh akscram could you guys shares updates? 17:37:24 I will just make now some small introductions. We would like to support 2 versions of Rally in the same time 17:38:00 1 that is service (with deamons) and could be used by corporation to have 1 Rally installation integrated in QA process and simple WEB UI for all operations 17:38:15 another one that is like now just CLI application 17:38:44 so ogelbukh akscram are hard working around rally-python client and rally-as-a-service part 17:39:05 #link https://review.openstack.org/#/q/status:open+project:stackforge/rally+branch:master+topic:bp/rally-service-api-v1,n,z 17:39:55 and here is the first v.0.1 rally 17:39:59 #topic http://docs.rallyapi.apiary.io/ 17:40:11 #topic 4) Rally & Smart benchmark scenarios checkers 17:40:24 #topic 5) Rally as a Service 17:40:29 omg =) 17:40:37 #link http://docs.rallyapi.apiary.io/ 17:40:53 so if somebody is interstened you are welcome to describe API of Rally 17:41:14 #topic 6) Improvements in Rally CLI 17:41:37 stannie julienvey could you share your work 17:42:08 i fixed a small bug on cli output 17:42:18 and currently working on rally use command 17:42:35 (trying to find the bp) 17:43:02 julienvey to find? 17:43:03 https://blueprints.launchpad.net/rally/+spec/rally-use-command 17:43:22 i have had some problem with exporting env vars in python 17:43:28 julienvey bp is approved 17:43:43 boris-42: it was just to share it :) 17:43:55 julienvey =) 17:44:03 julienvey so you are still working around ssh stuff 17:44:08 i should propose it to review tomorow, but i think it will need some rework 17:44:15 julienvey probably just call from python some bash command 17:44:25 julienvey dirty but works=) 17:44:38 actually it doesn't :( 17:44:54 julienvey ah yes.. 17:45:18 julienvey there is another way via DB, but not sure that it's okay 17:45:46 i went with the way proposed here https://bugs.launchpad.net/rally/+bug/1266439/comments/3 17:45:48 julienvey or some extra rally tmp fconfig 17:46:08 use .pam_environment file to store the env var 17:46:24 julienvey yep something like that should work 17:46:58 I am going to work on "deployment check" subcommand. The subcommand will check keystone auth, get all available services and print names etc. 17:47:02 #link https://blueprints.launchpad.net/rally/+spec/deployment-check-subcommand 17:48:24 stannie approved 17:48:29 thanks 17:48:53 stannie julienvey I will be waiting for your patches 17:49:08 stannie julienvey as well it will be good to update/add some extra info on wiki page 17:49:34 boris-42: yes, we have to :) 17:49:57 oaky as we have only 10 minutes moving to next topic 17:50:01 #topic 7) Rally & benchmarking performance of VMs 17:50:17 msdubov & me are going to start working around this 17:50:45 I wrote some interesting document 17:50:50 #link https://etherpad.openstack.org/p/Rally_VMs_Benchmarks 17:50:57 The idea is actually as well pretty simple 17:51:15 using the same benchmark engine 17:51:25 but in scenarios we will use heat & heat templates 17:51:46 that will install HPCC, SPEC and other Benchamrks suits on VMs 17:51:48 run them 17:51:56 then collect data 17:52:29 and have for each such benchmark specific type of output data and data processing tool (viewer) 17:52:46 that will show most important data + raw output of these benchmarks 17:53:30 hughsaunders julienvey stannie ^ 17:53:39 boris-42 How should we implement this viewer? 17:53:48 boris-42: so will your normalise data before it goes into DB? 17:53:56 *you 17:54:12 hughsaunders actually we will go to VMs 17:54:24 hughsaunders then fetch data from bench suuit 17:54:48 hughsaunders then store it in DB as is (but with some kind of marker e.g. specWEB2009) 17:55:06 hughsaunders and we will have viewer that can aggregate this date/or just display as is 17:56:23 so I think that I will publish this to mailing list to get more attention 17:56:31 there is a lot of for discussion 17:56:37 #topic 9) Rally measuring time of atomic actions 17:56:49 Short introduction 17:57:30 just use decorators e.g. @atomic_action("name_of_action") 17:57:31 https://github.com/stackforge/rally/blob/master/rally/benchmark/scenarios/nova/utils.py#L28 17:57:35 in such places ^ 17:57:59 this patch will measure time of call and put this time in special CLS dict 17:58:23 and scenario runner will add extra field https://github.com/stackforge/rally/blob/master/rally/benchmark/runner.py#L60 17:58:27 ^ here 17:58:43 and then improve CLI to display this data as well 17:59:01 as I know stannie is interested in implementing this 17:59:09 stannie I right? 17:59:15 am I right?* 17:59:31 * hughsaunders has to go, laters 17:59:40 we all have to go=) 17:59:46 hughsaunders bye 18:00:04 ok we can continue discussion in rally chat 18:00:10 #endmeeting