Tuesday, 2015-12-22

openstackgerritMerged openstack/python-senlinclient: Client support to receivers  https://review.openstack.org/25988000:03
*** Qiming has joined #senlin00:22
openstackgerritjunxu proposed openstack/senlin: Add 'EVENT' checking for scaling policy.  https://review.openstack.org/25882301:05
*** zhenguo has joined #senlin01:17
*** Yanyanhu has joined #senlin01:31
*** Qiming has quit IRC01:47
*** elynn has joined #senlin02:03
openstackgerritMerged openstack/senlin: Remove unused webhook APIs  https://review.openstack.org/26003702:04
openstackgerritMerged openstack/senlin: Add 'EVENT' checking for scaling policy.  https://review.openstack.org/25882302:06
*** Qiming has joined #senlin02:17
openstackgerritQiming Teng proposed openstack/senlin: Remove useless webhook RPC calls  https://review.openstack.org/26029702:23
elynnHi Qiming02:24
elynnAbout senlin resource02:24
elynnWhen I delete cluster, heat will check if cluster goes into deleting status.02:25
elynnBut the first check is so quick and cluster are not ready to put into deleting , still in ACTIVE02:26
elynnSo everytime I delete cluster from heat, I will got an delete failed error because the status are not correct.02:26
Qimingyes, delete will happen sometime in future, we cannot guarantee that action is performed immediately02:26
elynnIs it better to set cluster status immediately when receive the request?02:28
Qimingelynn, I don't like the heat way of checking if a resource is still there02:28
Qimingno, we cannot do that02:28
elynnI notice that now we start an action to set the status.02:28
Qimingchecking cluster status is not reliable02:28
Qimingthe way I would suggest is to follow nova's logic02:29
elynnI might take time until actions actually execute.02:29
Qimingyou do a wait02:29
Qimingwait for the delete to complete02:29
Qimingyou will use this API: http://git.openstack.org/cgit/openstack/python-openstacksdk/tree/openstack/resource.py#n99102:30
Qimingyou cannot require all resources will have a 'status' for you to check02:30
elynnOk, I will try to add something in senlin resource to make sure that update and delete are really starts....02:33
Qimingno, elynn, you cannot predict that02:33
elynnThe logic now is  if the status != deleting will raise error.02:34
Qimingfor delete, the only reliable way of checking is to try get that resource until you get ResourceNotFound or time out02:34
Qimingelynn, exactly, that needs a revision02:35
elynnChange to 'if status in [error, warning]' will raise error02:35
elynnis it better?02:35
Qimingno, you don't check status02:35
elynnso just wait for not_found?02:35
elynnWhat if it can not be deleted, then this resource will forever in in_progress....02:36
Qimingyou can check nova server resource's check logic02:36
Qimingyes, then you time out02:36
elynnok, it's a little weird but acceptable...02:37
Qimingyou can check how heat verify whether a server has been deleted: function check_delete_server_complete in its client plugin02:38
elynnlet me have a check02:39
elynnWill attend a meeting now, talk to you later...02:40
openstackgerritQiming Teng proposed openstack/senlin: Remove webhook from service layer  https://review.openstack.org/26029902:52
openstackgerritYanyan Hu proposed openstack/senlin: Enable update of flavor property of Nova server profile  https://review.openstack.org/25989003:28
elynnHi Qiming03:29
elynnAbout cluster update, it still face the status problem.03:30
elynnIf the cluster start to update, the status of cluster will follow this routine 'active -> updating -> active'03:32
elynnIt's hard for heat to distinguish the first 'active' and the last 'active', which means heat don't know when the updation ends.03:35
Qiminghow does Heat check nova server update?03:36
Qimingnova server doesn't provide a 'update-in-progress' status right?03:36
elynnlet me see03:38
elynnJust check the codes03:40
elynnit just check the status, if status == 'active' will put status to complete.03:41
Qimingso we don't know if the update is performed or not03:42
elynnnova is a little difference , it have a 'VERIFY_RESIZE' before 'active' , heat actually wait for 'verify_resize'03:44
Qimingwhat about changing hostname, changing image?03:44
elynnonly check for 'error' status03:45
elynnSeems heat assume that the first 'active' is omit... nova will change the status immediately, so just wait for the last 'active'.03:47
Qimingthat is not reliable, isn't it?03:47
Qimingwill think about it03:47
elynnThe same for the deletion....03:48
elynnAs a workaround, I can add a counter number for status 'active', and omit any 'active' status before change to 'updating', but the codes will looks a little ugly...03:52
openstackgerritZhuohua Li proposed openstack/senlin: test  https://review.openstack.org/26031304:21
Qimingelynn, don't do that04:55
Qimingit sounds very dangerous to me04:55
Qimingafter the API revision, we are now returning 202 for CREATE and UPDATE04:55
Qimingin that body, we have an action id04:55
Qiminga better alternative is to check if that action has been labelled SUCCEEDED04:56
elynnIt sounds better, let me see how to get the action id from body.04:58
xuhaiwei_hi, when I rebuild my openstack by using devstack, I got this error: Could not determine a suitable URL for the plugin05:00
elynnDo you have any idea about how to get the returned action id?05:00
xuhaiwei_have you ever met this before, sorry to interrupt05:00
elynnxuhaiwei_: Never saw it before, do you have any detailed log?05:01
Qimingme neither05:01
xuhaiwei_no more details, the script failed when creating a image05:01
Qimingseems you need to remove some existing packages, especially keystone related05:02
Qimingincluding keystoneauth05:02
xuhaiwei_I googled it, it seems related to keystone version05:02
xuhaiwei_ok, I will try  it05:02
openstackgerritQiming Teng proposed openstack/python-senlinclient: Use SDK calls when possible  https://review.openstack.org/26032805:45
openstackgerritQiming Teng proposed openstack/python-senlinclient: Directly use SDK calls in client (1)  https://review.openstack.org/26032905:45
Yanyanhuhi, Qiming, around?06:00
YanyanhuI'm trying to remove the name property of os.nova.server profile06:00
YanyanhuI think if user really want to embed a unique str into the name of nova servers created, they can name the cluster using that string06:03
Yanyanhuallowing to specify the name property of nova server profile makes thing complicated06:05
Qimingnova doesn't allow servers to have the same name?06:15
Yanyanhuallows I think06:15
Qimingthen why are are enforcing name uniqueness?06:17
Qimingwhy we are enforcing name uniqueness?06:17
Yanyanhunot enforcing name uniqueness, just don't allow user to define the name of the os.nova.server profile06:17
Yanyanhuactually, the name property is not used before06:18
Qimingif user provides 'name', they mean they want to name their servers using that string06:18
Qimingwe don't care why they are doing that06:18
Qimingif they don't provide 'name', we use cluster name plus a random suffix?06:19
Yanyanhuusing that string like 'NAME-xxx'06:19
Qimingwe shouldn't do that06:19
Yanyanhushouldn't use node name plus a random suffix?06:20
Qimingit is weird when I tell nova to create server named 'web' and I'm getting a 'web-xxxyyy'06:20
Yanyanhuyes, I agree with this06:20
Qimingif 'name' is not provided, it means users don't care about names, we help create random, meaningful names for them06:20
Yanyanhuyou mean directly using name given by user to name all servers in the cluster?06:20
Yanyanhuso all servers will have the same name06:21
Qimingusers know what they are doing06:21
Yanyanhuhmm, makes sense06:21
Yanyanhuok, I think you're right06:21
Yanyanhuwe just follow user's request regardless whether it is weird06:22
Yanyanhusince this is what they want06:22
Qimingat best, we can name a server with '%s%s' % (name, index)06:22
Qimingbut that is not an useful feature I'm afraid06:22
Qimingusers should always get what they want06:23
Qimingrandom suffix at all times is a bad design IMO06:23
YanyanhuI agree with this06:23
Qimingthe key is that we will never use name as the unique ID of a resource06:24
openstackgerritMerged openstack/senlin: Remove useless webhook RPC calls  https://review.openstack.org/26029706:32
openstackgerritMerged openstack/senlin: Remove webhook from service layer  https://review.openstack.org/26029906:32
Qimingdrafting mid-cycle meetup agenda06:46
Qimingplease add topics you want to discuss06:46
Yanyanhugot it06:47
xuhaiwei_I just got a bad news06:48
xuhaiwei_I may not join the meetup last month06:48
Qimingpls ...06:48
xuhaiwei_forgive me sir06:48
xuhaiwei_there is no budget for me :(06:49
Qimingair polution is not that bad indoor06:49
xuhaiwei_really sorry06:50
Qimingwe'll try to sync things online via etherpad06:50
xuhaiwei_not dicided yet, but the chance is very small06:50
xuhaiwei_that will be a great help for me06:50
Qimingplease try convince your manager06:51
Qimingthis is not gonna be just a simple meetup, check the agenda06:52
xuhaiwei_I will go to India for about one week at the end of Jan, I asked my manager can I use that budget to go to Beijing, but he said no, the budget comes from different project06:52
xuhaiwei_the budget management is not good06:53
xuhaiwei_I gonna miss a lot I think06:54
Qimingthese two were outcome from a git commit mistake06:54
Qimingthey are supposed to be in a single patch06:55
Qiminganyway, pls help review06:55
Qimingxuhaiwei_, no worries06:55
Qimingwe will try get you updated06:55
openstackgerritYanyan Hu proposed openstack/senlin: Make name property of os.nova.server profile take effect  https://review.openstack.org/26035507:05
openstackgerritYanyan Hu proposed openstack/senlin: Make name property of os.nova.server profile take effect  https://review.openstack.org/26035507:26
openstackgerritYanyan Hu proposed openstack/senlin: Make name property of os.nova.server profile take effect  https://review.openstack.org/26035507:41
*** lixinhui has joined #senlin08:01
xuhaiwei_about the client patch08:23
xuhaiwei_has commented the patch08:31
xuhaiwei_dont understand why shell.py is also changed08:32
Qimingshell.py is invoking SDK08:32
Qimingclient.py is useless08:32
xuhaiwei_so client will be removed?08:33
Qimingit is there only because there are other services which are used to invoke a xxxclient.client module08:33
Qimingin a perfect world, they should invoke SDK directly as well08:33
Qimingthen we can remove the client.py module as well, leaving only the shell.py08:33
xuhaiwei_but your patch is making double fix of client.py and shell.py08:37
Qimingyes, it is an effort to remove our local models.08:40
xuhaiwei_why not just fix shell.py, since the method like create_profile in client.py is not used any more08:41
Qimingcreate_profile is invoked from heat08:42
Qimingit is invoked from senlin-dashboard08:42
xuhaiwei_that is the point, got it08:43
openstackgerritYanyan Hu proposed openstack/senlin: Enable name property update for os.nova.server profile  https://review.openstack.org/26039208:57
openstackgerritMerged openstack/python-senlinclient: Use SDK calls when possible  https://review.openstack.org/26032809:10
Yanyanhuhi, Qiming, about this patch https://review.openstack.org/#/c/259301/09:11
Yanyanhuso pbr itself can handle the files that will finally be included into the packages?09:12
Qimingcontent in MANIFEST is useless09:13
Yanyanhuwhere pbr gets the file information?09:13
Yanyanhuok, many thanks for explaination09:14
QimingMANIFEST is mainly about which files should be included09:14
Yanyanhuhi, Qiming, about this patch https://review.openstack.org/#/c/260329/09:28
YanyanhuI guess Cluster definition can also be removed from model.py ?09:28
YanyanhuI saw related support has been in sdk?09:28
QimingYanyanhu, not yet, the action apis not yet revised09:30
Yanyanhuok, I see. Thanks09:32
openstackgerritMerged openstack/python-senlinclient: Directly use SDK calls in client (1)  https://review.openstack.org/26032909:35
*** zhenguo has quit IRC10:01
*** zhenguo has joined #senlin10:01
*** Qiming has quit IRC10:16
*** Yanyanhu has quit IRC10:23
*** elynn has quit IRC11:06
*** Qiming has joined #senlin11:17
*** lixinhui_ has joined #senlin11:57
*** yanyanhu has joined #senlin12:51
*** elynn has joined #senlin12:55
yanyanhuelynn, you can try to increase engine number I think14:00
yanyanhuto the number of processor cores you have14:00
Qimingyes, that would be interesting14:00
yanyanhuafter the latest fix of DB concurrency issue14:00
Qiminga single python thread doing sequential work -- deleting nodes14:00
yanyanhuI'm more confident about our multi-engine support14:01
Qimingyou have to wait14:01
elynnSo a cluster can lock an engine entirely?14:01
yanyanhuQiming, yes14:01
yanyanhuelynn, not a cluster, just too many actions14:01
yanyanhumake the engine too busy I think14:01
Qimingcluster is not locked by an engine14:01
yanyanhutoo busy to respond the rpc request before timeout14:02
Qimingelynn, you aware of GIL ?14:02
elynnlet me try multi-engine tomorrow :)14:02
yanyanhuelynn, I set engine number to 2 by default14:03
Qimingthere's no real parallelism in CPython: https://wiki.python.org/moin/GlobalInterpreterLock14:03
yanyanhuyou can increase the number if you server is more powerful :)14:03
lixinhui_elynn and yanyanhu, hope you can provide your experience with me14:03
yanyanhulixinhui_, no problem14:03
elynnlixinhui: no problem14:03
lixinhui_it would be nice if you can send me the configs of your engine server and scale14:04
lixinhui_thanks a lot14:04
yanyanhulixinhui_, sure, will paste my senlin.conf to you tomorrow14:04
lixinhui_okay, yanyanhu14:05
yanyanhuactually I just increase the engine worker number, other part is default :)14:05
elynnhaven't change senlin.conf for now...14:05
lixinhui_I mean the cpu, memory14:05
yanyanhuoh, you mean my work environment14:05
yanyanhuah, it's a VM with 4 VCPU and 16GB memory14:05
lixinhui_I want to recovery that and do some profiling if need14:05
yanyanhuwill do some scalability related test14:06
yanyanhusure, understand :)14:06
elynnI's using i7 16G physical machine14:06
lixinhui_yanyanhu, you just mentioned you can create 5 clusters and more than 1000 nodes each14:07
elynnit should be 8 cores with hyper-threading.14:07
lixinhui_okay elynn, yes that is important14:07
yanyanhuahh, actually with 200 nodes in each cluster...14:07
yanyanhutotally 100014:07
lixinhui_oh, okay14:07
lixinhui_what flavor for this tests to create vm14:08
yanyanhuwill increase the engine worker number to 4 and compare the time consumption of cluster increasing14:08
yanyanhuI used openstack_test cloud backend driver14:08
yanyanhuthis cloud backend driver was developed for functional test14:09
yanyanhuusing it, senlin engine won't talk with other openstack services to create real resource, .e.g nova server or heat stack14:09
lixinhui_cool :)14:09
yanyanhuso we can use it to test the scalability of engine I think14:09
yanyanhuwill show you how to use it tomorrow :)14:10
lixinhui_okay :) thanks in advance, yanyanhu14:10
yanyanhumy pleasure :)14:10
yanyanhuwill leave, talk to you guys later14:11
lixinhui_see you tomorrow14:11
yanyanhusee U14:11
*** yanyanhu has quit IRC14:11
*** elynn has quit IRC14:12
*** lixinhui_ has quit IRC14:46
*** bdrich_ has joined #senlin14:51
*** Qiming has quit IRC15:19
*** bdrich_ has quit IRC15:45
*** openstackgerrit has quit IRC18:02
*** openstackgerrit has joined #senlin18:03
*** bdrich_ has joined #senlin20:47
*** bdrich_ has quit IRC21:36
*** pratikmallya has joined #senlin22:22
*** Qiming has joined #senlin22:45
*** Qiming has quit IRC22:50

Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!