17:08:40 <boris-42> #startmeeting Rally 17:08:41 <openstack> Meeting started Tue Jan 28 17:08:40 2014 UTC and is due to finish in 60 minutes. The chair is boris-42. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:08:42 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 17:08:45 <openstack> The meeting name has been set to 'rally' 17:09:00 <boris-42> hughsaunders julienvey ping 17:09:18 <hughsaunders> \o boris-42 17:10:11 <miarmak> hi all 17:11:04 <boris-42> okay let's start probably other will be soon 17:11:22 <boris-42> today we will discuss as usually updates 17:11:30 <boris-42> 1. Changes in CLI 17:11:34 <boris-42> 2. Tempest stuff 17:11:47 <boris-42> 3. LXC engine / Multihost deploy stuff 17:12:12 <boris-42> 4. Benchmark scenario validators 17:12:38 <boris-42> 5. Benchmarking without admin users 17:12:50 <boris-42> 6. OpenDiscussion 17:12:58 <boris-42> #topic Changes in CLI 17:13:23 <boris-42> hughsaunders could you say updates in CLI? 17:13:33 <boris-42> about* 17:13:52 <hughsaunders> so main thing is that julienvey has done some brilliant work on persisting environment 17:14:08 <hughsaunders> which allows cli to set a default deployment and refer to that in future commands 17:14:29 <hughsaunders> this has been added to deployment create and other commands that require a deployment 17:14:51 <hughsaunders> Also added the option of creating a dummy deployment from environment variables 17:15:17 <hughsaunders> any others I've missed? 17:15:54 <boris-42> hughsaunders yep it's all that was merged 17:16:23 <boris-42> hughsaunders ahh no there is work from stannie 17:16:56 <boris-42> hughsaunders "rally deployment check" 17:17:06 <hughsaunders> ah yes, stannie added the deployment check command 17:17:19 <boris-42> that allows us to check that deployment work and returns all available services 17:17:28 <hughsaunders> apologies stannie! 17:17:39 <boris-42> as well he add pretty output if endpoints are wrong 17:18:05 <boris-42> So during this week we are going to rework plot CLI 17:18:18 <boris-42> and add some other types of plots + refactor existing 17:18:45 <boris-42> #topic Tempest stuff 17:19:02 <julienvey> hi, coming in a little late 17:19:05 <stannie> hi, sorry for the delay, was in another meeting :) 17:19:14 <boris-42> julienvey stannie nice =) 17:19:23 <boris-42> glad to see you=) 17:20:34 <boris-42> So the idea of tempest is to have command that will validate cloud 17:20:44 <boris-42> so something like "rally validate start" 17:20:47 <boris-42> that will run tempest 17:20:58 <boris-42> against cloud that we are currently "use" 17:21:05 <boris-42> process results and store them 17:21:27 <boris-42> also as future improvements we would like to add some tasty commands 17:21:46 <boris-42> rally verify list (to show all uuid of validation results) 17:21:54 <boris-42> rally verify show (with detailed infromration) 17:21:54 <stannie> that will be a great addition 17:22:11 <boris-42> rally verify <some specific set> 17:22:40 <boris-42> so procreated sets of tests (for nova/glance/..., short tests/..) and so on 17:23:00 <boris-42> I think this will be very useful for community 17:23:23 <boris-42> so we faced some troubles …. 17:23:29 <boris-42> there is no pypy for tempest 17:23:34 <hughsaunders> :( 17:23:36 <boris-42> and we have to make "git clone" 17:23:39 <boris-42> sdague ^ 17:24:05 <boris-42> So we are going to make some dirty code to handle this.. 17:24:17 <boris-42> I think that we should make command like rally verify install 17:24:33 <boris-42> that will install tempest in some rally directory (e.g. rally/verify/tempest) 17:25:04 <boris-42> miarmak stannie julienvey hughsaunders thoughts ^ 17:25:25 <boris-42> jaypipes ^ 17:25:28 <miarmak> In this situation, I don't have another varients 17:25:35 <hughsaunders> boris-42: I was wondering if there is a good git python library, so we don't have to shell out 17:26:33 <jaypipes> hughsaunders: not really :( 17:26:44 <stannie> can't we build a bundle of tempest with pypy and use it directly ? 17:27:14 <jaypipes> hughsaunders: best approach is to write a tempfile with a shell script and execute a series of git commands by shelling out 17:27:36 <boris-42> jaypipes or just use shell from python .. 17:27:48 <hughsaunders> jaypipes: thats a shame, but I guess reasonably straightforward 17:28:05 <jaypipes> boris-42: right... that's what I described :) 17:28:16 <hughsaunders> I like stannie's idea, I wonder if infra would be up for adding a pypi-publish job for tempest, or if it is deliberately not published 17:28:18 <boris-42> okay so we will use some kind of Popen()? 17:28:32 <jaypipes> boris-42: only if more than a few commands, better performance from just shelling out once and running a script. 17:28:42 <miarmak> mb subprocess? 17:28:52 <jaypipes> boris-42: use openstack.common.processutils? 17:29:28 <jaypipes> hughsaunders: I think stannie was talking about pypy, not pypi :) 17:29:40 <stannie> nah pypi, was a mistake :) 17:29:47 <jaypipes> oh! :) sorry. 17:30:00 <boris-42> so okay guys at this moment we will make … git clone somehow 17:30:15 <miarmak> =) 17:30:16 <boris-42> then we will try to coordinate with openstack-qa guys 17:30:32 <boris-42> to make it cleaner 17:30:34 <jaypipes> boris-42: https://github.com/openstack/oslo-incubator/blob/master/openstack/common/processutils.py#L84 17:30:54 <boris-42> jaypipes ^ I know I know..=) 17:31:09 <boris-42> so next topic 17:31:20 <boris-42> #topic LXC engine / Multihost deploy stuff 17:31:37 <boris-42> sooo seems like redixin is missing second meeting.. 17:31:45 <boris-42> so I will have to make some updates 17:32:38 <boris-42> So for that who don't know what is it 17:32:54 <boris-42> It is actually the super fast way to get your cloud with deployers and servers that you have 17:33:00 <boris-42> Like I have 5 servers 17:33:09 <boris-42> and I would like to deploy on 1 controller 17:33:30 <boris-42> on other 2 I would like to create 100500 cinder nodes 17:33:41 <boris-42> and on other 2 I would like to create 100500 nova nodes 17:34:32 <boris-42> so using Multihost eninge we are able to describe 1 controller section and 2 nodes sections 17:35:02 <boris-42> https://review.openstack.org/#/c/57240/25/doc/samples/deployments/multihost.json 17:35:04 <boris-42> like here ^ 17:35:25 <boris-42> for describing each controller we can use any of deployers in Rally (there will be at least 3 deployers) 17:35:37 <boris-42> dummy (returns existing endpoints) devstack and fuel 17:36:11 <boris-42> for nodes you are able to use actually only DevStack 17:36:19 <boris-42> (I am not sure that we will support fuel) 17:36:42 <boris-42> and you are able to use LXCEngine + DevStack engine to make rappid deployment using approach of copy pasting them 17:37:02 <boris-42> here is the sample https://review.openstack.org/#/c/56222/20/doc/samples/deployments/devstack-lxc-engine-in-dummy.json 17:37:23 <boris-42> so this allows us to get for 20 minutes really huge cloud 17:37:30 <jaypipes> boris-42: so controller/compute/whatever is deployed into an LXC container? 17:37:44 <boris-42> jaypipes it depend 17:37:52 <boris-42> jaypipes if you are using LXCEngine yes 17:38:08 * jaypipes has never been able to install OVS in an LXC container because of the shared kernel. 17:38:29 <jaypipes> curious to see how it was doe. 17:38:31 <boris-42> jaypipes yep with Neutron we can face some problems.. 17:38:31 <jaypipes> done... 17:38:36 <jaypipes> ah... k. 17:39:06 <boris-42> jaypipes I hope we will find the way to resolve them with some kind of Neutron fake OVS driver 17:39:41 <boris-42> jaypipes the goal of such deployment is to figure how works DB/AMQP/CodeInfrastracture/Scheduler at scale 17:39:50 <boris-42> not about networking/virtualization and so on 17:40:29 <boris-42> And with LXC controllers we are able to create really huge clouds in half on hour with tons of physical resources 17:40:35 <boris-42> containers* 17:41:17 <boris-42> so there was an issue that Rally have to work on controller server (that I didn't know) 17:41:26 <boris-42> so I didn't have success yesterday to run it 17:41:38 <boris-42> But Sergey fixed this issue and I will try one more time 17:41:48 <boris-42> If it work I will merge in upstream =) 17:42:08 <boris-42> And we will get weapon against bottlenecks in infrastructure=) 17:42:47 <boris-42> jaypipes hughsaunders julienvey stannie questions? 17:43:03 <jaypipes> none from me, thx boris-42 17:43:25 <hughsaunders> I have no experience with LXC, but should test that patch, sounds good for testing some things at scale 17:43:51 <boris-42> hughsaunders the goal of that stuff in Rally is to avoid learning LXC=) 17:44:02 <boris-42> hughsaunders just put right config and it will do all work for you=) 17:44:07 <hughsaunders> boris-42: I'll test 17:44:14 <boris-42> hughsaunders thanks=) 17:44:31 <stannie> no boris-42, except that would have been interesting to integrate docker for easier LXC manipulation and snapshots feature :) 17:44:49 <boris-42> stannie hmm but it is easy... 17:45:04 <boris-42> stannie I mean I don't see where it will be easier with docker.. 17:45:21 <stannie> you snapshot your openstack infra at a given time, then launch some tests, re-deploy to another snapshot, compare the results 17:45:54 <boris-42> to another snapshot? 17:45:54 <stannie> but anyway as jaypipes said, we will still encounter a lot of issues with neutron, network ovs etc :/ 17:46:13 <boris-42> I mean why not just deploy it from source? 17:46:48 <stannie> from source ? 17:46:52 <boris-42> like now 17:47:02 <boris-42> we are creating 1 lxc container 17:47:14 <boris-42> then ./stack.sh with correct locarl 17:47:17 <boris-42> localrc 17:47:30 <boris-42> then start N from snapshot 17:47:45 <stannie> ok 17:47:56 <stannie> didn't know we were able to do this yet :) 17:48:22 <boris-42> stannie this patch https://review.openstack.org/#/c/56222/ 17:48:53 <stannie> ok didn't see the patch 20 17:49:07 <stannie> patch set* 20 17:49:15 <boris-42> there is no big difference.. 17:49:51 <stannie> ok 17:49:52 <boris-42> stannie so pls test it as well=) 17:50:02 <boris-42> stannie and we will merge it finally=) 17:50:17 <boris-42> #topic 4. Benchmark scenario validators 17:50:21 <boris-42> olkonami hi 17:50:30 <boris-42> olkonami could you share your results? 17:50:35 <olkonami> hi 17:52:14 <olkonami> basic patch with validators adder was merged yesterday 17:52:56 <olkonami> and validator for image_id is on review now 17:52:58 <olkonami> https://review.openstack.org/#/c/68055/ 17:53:54 <boris-42> so guys if you have some time take a look at these patches ^ 17:54:14 <boris-42> 5. Benchmarking without admin users 17:54:20 <boris-42> #topic 5. Benchmarking without admin users 17:54:34 <boris-42> So Mike done already all required patches 17:54:56 <boris-42> the idea is to be able to run part of benchmarks against procreated users without admin role 17:55:09 <boris-42> so there are couple of changes in scenario runner 17:55:35 <boris-42> if there is no admin users we should create new temp users and then delete them 17:55:52 <boris-42> and as well we don't have ability to run part of benchmarks (like keystone benchmarks) 17:56:19 <boris-42> patches are here https://review.openstack.org/#/q/status:open+project:stackforge/rally+branch:master+topic:bp/benchmarking-with-predefined-users,n,z 17:56:56 <boris-42> jaypipes hughsaunders julienvey stannie questions? 17:57:07 <jaypipes> nope 17:57:28 <stannie> nope 17:57:36 <boris-42> #topic atomic actions 17:57:44 <boris-42> stannie do you have any updates? 17:57:45 <hughsaunders> boris-42: not from me, ability to run partial scenario may be useful, so we can avoid creating sub scenarios (like authentication?) 17:58:38 <boris-42> hughsaunders authentifaction sutff is great 17:58:54 <boris-42> hughsaunders (but it should be in keystone benhcmarks) 17:59:03 <hughsaunders> yep 17:59:06 <boris-42> hughsaunders I think we will add some kind of decorator 17:59:16 <hughsaunders> julienvey is good at that :) 17:59:20 <boris-42> hughsaunders e.g. @admin_required for benchmarks that are only from amdins 17:59:27 <boris-42> hughsaunders lol =) 17:59:30 <julienvey> hughsaunders: ;) 17:59:38 <boris-42> stannie do you have any updates?) 17:59:41 <julienvey> hughsaunders: if my patch finally gets merged ! 17:59:58 <boris-42> Okay guys nice meeting 18:00:06 <boris-42> now we should move out this chat 18:00:10 <boris-42> #endmeeting