| *** ruijie has joined #senlin | 02:26 | |
| openstackgerrit | Merged openstack/senlin master: Implement policy in code - reno and doc (end) https://review.openstack.org/513634 | 04:05 |
|---|---|---|
| *** maestropandy has joined #senlin | 05:04 | |
| maestropandy | Hi Senlin team, One thing that isn’t quite clear to me is why senlin will scale better than 1. existing ceilometer DB based autoscaling, and 2. Aodh + Gnocchi based autoscaling. Once you answer this in a slide, let’s have a review. Just an info: The default ceilometer solution uses Mongo DB with a data model that isn’t optimized for large scale clouds. The autoscaling queries to get a metric like CPU utilization take too much time. I need to underst | 06:26 |
| maestropandy | Hi Senlin team, One thing that isn’t quite clear to me is why senlin will scale better than 1. existing ceilometer DB based autoscaling, and 2. Aodh + Gnocchi based autoscaling. The default ceilometer solution uses Mongo DB with a data model that isn’t optimized for large scale clouds. The autoscaling queries to get a metric like CPU utilization take too much time. I need to understand what is different in Senlin that will resolve this problem. | 06:27 |
| Qiming | maestropandy, thanks for the question | 06:28 |
| Qiming | let me try give you a little bit background about senlin's design considerations | 06:28 |
| Qiming | the whole auto-scaling idea was derived from Heat (the orchestration project) and borrowed some ideas from AWS auto-scaling service | 06:29 |
| Qiming | I assume you have some backgrounds on both | 06:29 |
| Qiming | to make "auto-scaling" work, you will have to provision a monitoring/metering service | 06:30 |
| Qiming | the openstack community already has ceilometer, monasca, and other monitoring/alarming/metering services | 06:31 |
| Qiming | Senlin was not designed to reinvent that part of the solution, we believe we should leverage the community instead of competing with it | 06:32 |
| Qiming | let the best people do the best thing | 06:32 |
| maestropandy | Qiming: I have few challenges mentioned.. the quering metrics takes much time ... | 06:32 |
| Qiming | Selin is not bound to any monitoring solution, a user has the flexibility to choose whatever solutions he sees right | 06:33 |
| Qiming | be it aodh, the old ceilometer api (now deprecated), monasca, their in-house solution, other 3rd party solutions such as nagios, ... etc. | 06:33 |
| Qiming | Senlin only cares about the "receiving" end, leaving the "triggering" end open | 06:34 |
| maestropandy | Qiming: agreed, metering is not in senlin boundary, but it has relation/dependency with senlin to perform autosclaing, so question araises to evalutate | 06:34 |
| Qiming | we are no experts in harvesting metrics, aggregating and filtering them or storing them in time series database etc etc ... | 06:34 |
| Qiming | as an example, we tried integration with Aodh and documented that process as a sample | 06:35 |
| maestropandy | Qiming: I have seen your videos dictates that you have performed scaling in your org large scale, in that time. how was performance ? | 06:35 |
| Qiming | the large scale deploy was from China Mobile | 06:35 |
| Qiming | I don't think they were using ceilometer | 06:36 |
| Qiming | my vague memory is that ceilometer's scalability is terrible | 06:36 |
| Qiming | that was the reason why they switched over to gnocchi (a time series database) | 06:36 |
| maestropandy | Qiming: yes for that, how was support of senlin, do i have some references or parameters they considered while performing .. like to go some monitoring than ceilometer.. | 06:36 |
| maestropandy | we planned to use gnocchi to avoid ceilometer annoyness | 06:37 |
| Qiming | in a devstack environment, we tried mysql and mongo db, neither one performs well | 06:37 |
| maestropandy | but i need your experience or feedback/suggestion to perform large scale deployment | 06:37 |
| Qiming | better ask some experts in the cloud monitoring domain, they know best what are the recommended parameters | 06:38 |
| maestropandy | can i have some reference points from that china mobile projects performed large scale using senlin .. like parameters ( aodh/gnocchi or whatever they used) | 06:38 |
| Qiming | I myself is not an operator, ..., I listen to what they complain and try fix the mess , :) | 06:39 |
| Qiming | do you know junwei liu ? | 06:39 |
| Qiming | he sits on the openstack board | 06:39 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/python-senlinclient master: Imported Translations from Zanata https://review.openstack.org/514150 | 06:39 |
| maestropandy | junwei.. his IRC name and in which channel i can find him ? | 06:40 |
| Qiming | you can shoot an email to him, and see if he (his team) is willing to share some experiences with you | 06:40 |
| Qiming | btw, that team is a great one ... | 06:40 |
| maestropandy | can i have his complete mail id please .. | 06:40 |
| Qiming | lemme find a link for you | 06:41 |
| maestropandy | okay | 06:41 |
| Qiming | here is his linkedin profile: https://www.linkedin.com/in/%E5%86%9B%E5%8D%AB-%E5%88%98-51409696/ | 06:42 |
| Qiming | here is his openstack profile: https://www.openstack.org/community/members/profile/26694/junwei-liu | 06:42 |
| maestropandy | thanks :) | 06:43 |
| maestropandy | does he has IRC account ? | 06:44 |
| Qiming | besides monitoring, I believe they have solved many other problems related to database, messaging and networking when deploying a large scale cluster | 06:44 |
| Qiming | I'm not sure he is very active on IRC | 06:44 |
| maestropandy | thinking about to use infuxdb inplace of ceilometer.. but need to explore | 06:46 |
| maestropandy | Qiming: also senlin master while installing with specifc branch like stable/ocata in devstack.. it is pointing to master branch automatically, I have raised this question already.. | 06:48 |
| Qiming | you have filed a bug? | 06:49 |
| Qiming | sounds like something to improve in the devstack install script ... | 06:49 |
| maestropandy | its working fine with devstack/master & senlin/master.. if i point stable branches, senlin points to master, where senlin-dashboard remains in stable branch | 06:52 |
| maestropandy | http://paste.openstack.org/show/624333/ | 06:55 |
| maestropandy | Qiming: can i have that documented of Aodh & Gnocchi which you have mentioned to look into.. | 06:58 |
| Qiming | it is all here: http://docs.openstack.org/developer/senlin/ | 06:59 |
| maestropandy | Qiming: whats difference between heat orchestration project and senlin though it derived, it will have some pro's, would like to know, can i have comparison of heat & senlin scaling .. | 07:04 |
| maestropandy | Qiming: are you coming for sydney summit ? | 07:15 |
| openstackgerrit | OpenStack Proposal Bot proposed openstack/senlin-dashboard master: Imported Translations from Zanata https://review.openstack.org/514169 | 07:26 |
| *** AlexeyAbashkin has joined #senlin | 07:37 | |
| *** maestropandy has quit IRC | 07:41 | |
| *** AlexeyAbashkin has quit IRC | 08:12 | |
| *** AlexeyAbashkin has joined #senlin | 08:12 | |
| *** maestropandy has joined #senlin | 09:05 | |
| *** maestropandy has left #senlin | 09:06 | |
| *** jmlowe_ has joined #senlin | 09:07 | |
| *** maestropandy has joined #senlin | 09:07 | |
| *** jmlowe has quit IRC | 09:08 | |
| Qiming | maestropandy, I'm not coming | 09:22 |
| Qiming | as for the difference between heat and senlin | 09:22 |
| maestropandy | Qiming: hmm .. | 09:22 |
| Qiming | heat is for orchestration, constructing complex structs of resources of different types, then deploy and update | 09:23 |
| Qiming | senlin is for management of homogeneous resources only, it creates and manages collections of similar objects | 09:24 |
| Qiming | you can think of heat as the 'struct' in C programming language, senlin as the 'array' data type | 09:24 |
| Qiming | you can create and orchestrate senlin clusters using heat stacks | 09:25 |
| Qiming | you can also create and manage groups of heat stacks using senlin clusters | 09:25 |
| maestropandy | Qiming: Struct & array is very clear example thanks :), but both performs autoscaling ( senlin does using heat), so senlin uses heat features to perform cluster scalings, am right ? | 09:27 |
| Qiming | no | 09:27 |
| Qiming | senlin has its own engine and api for the cluster resizing operation | 09:27 |
| Qiming | you can do, for example, openstack cluster expand, openstack cluster shrink, openstack cluster resize, etc. | 09:28 |
| maestropandy | sorry, orchestration of senlin clusters uses heat ? is it right ? | 09:28 |
| Qiming | yes, if you are managing senlin clusters using heat, you can only use heat apis/commands | 09:29 |
| Qiming | you are then not supposed to manage the clusters directly, heat will get confused if cluster status is changed in the background | 09:29 |
| Qiming | similarly, you are not supposed to operate your nova servers directly if the servers are created as part of a heat stack | 09:30 |
| maestropandy | hmm clear :) | 09:32 |
| *** maestropandy has quit IRC | 10:37 | |
| *** Qiming has quit IRC | 10:50 | |
| *** Qiming has joined #senlin | 10:54 | |
| *** Qiming has quit IRC | 10:58 | |
| *** Qiming has joined #senlin | 11:01 | |
| *** maestropandy has joined #senlin | 12:58 | |
| *** maestropandy has quit IRC | 12:58 | |
| *** jmlowe_ has quit IRC | 13:22 | |
| *** jmlowe has joined #senlin | 13:58 | |
| *** ruijie has quit IRC | 14:08 | |
| *** jmlowe has quit IRC | 15:31 | |
| *** AlexeyAbashkin has quit IRC | 15:37 | |
| *** jmlowe has joined #senlin | 15:40 | |
| *** jmlowe has quit IRC | 16:43 | |
| *** AlexeyAbashkin has joined #senlin | 17:13 | |
| *** AlexeyAbashkin has quit IRC | 17:17 | |
| *** jmlowe has joined #senlin | 17:26 | |
| *** AlexeyAbashkin has joined #senlin | 19:58 | |
| *** AlexeyAbashkin has quit IRC | 20:02 | |
| *** jmlowe has quit IRC | 20:14 | |
| *** jmlowe has joined #senlin | 21:15 | |
| *** jmlowe has quit IRC | 21:50 | |
| *** jmlowe has joined #senlin | 22:08 | |
| *** jmlowe has quit IRC | 22:24 | |
| *** jmlowe has joined #senlin | 22:24 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!