| *** tzn has quit IRC | 00:10 | |
| *** tzn has joined #openstack-bareon | 01:10 | |
| *** tzn has quit IRC | 01:16 | |
| *** tzn has joined #openstack-bareon | 02:13 | |
| *** tzn has quit IRC | 02:19 | |
| *** tzn has joined #openstack-bareon | 04:16 | |
| *** tzn has quit IRC | 04:20 | |
| *** tzn has joined #openstack-bareon | 05:17 | |
| *** tzn has quit IRC | 05:22 | |
| *** tzn has joined #openstack-bareon | 06:18 | |
| *** tzn has quit IRC | 06:22 | |
| *** tzn has joined #openstack-bareon | 07:19 | |
| *** tzn has quit IRC | 07:23 | |
| *** sylwesterB has joined #openstack-bareon | 07:42 | |
| *** sylwesterB has quit IRC | 07:46 | |
| *** sylweste_ has joined #openstack-bareon | 07:46 | |
| *** tzn has joined #openstack-bareon | 07:47 | |
| *** tzn has quit IRC | 08:07 | |
| *** tzn has joined #openstack-bareon | 09:38 | |
| *** tzn has quit IRC | 09:39 | |
| sylweste_ | eli: I've read the document you have provided http://bareon-allocator.readthedocs.org/en/latest/architecture.html | 09:41 |
|---|---|---|
| *** sylweste_ has left #openstack-bareon | 09:41 | |
| *** sylweste_ has joined #openstack-bareon | 09:41 | |
| *** sylweste_ is now known as sylwesterB | 09:41 | |
| sylwesterB | AFAIU | 09:41 |
| sylwesterB | you were looking for a sequence (1,2,4,6...) | 09:42 |
| sylwesterB | which will be this objective function for maximization | 09:43 |
| sylwesterB | which will be used by simplex alg, right? | 09:44 |
| sylwesterB | evgenyl:^^ | 09:44 |
| evgenyl | sylwesterB: yes, the sequence is being used as coefficients for objective function | 09:50 |
| sylwesterB | evgenyl: ok, and sum(c^T x) is a maximization value. Simplex arg should look for the greatest value of it, right? | 09:54 |
| sylwesterB | Simplex alg* | 09:54 |
| *** tzn has joined #openstack-bareon | 10:01 | |
| sylwesterB | I'm just checking if I understand it correctly ;) | 10:02 |
| evgenyl | sylwesterB: correct. It's a function which is used to figure out, which solution is "better" or "optimal". | 10:05 |
| sylwesterB | I like it, very clever. Do you have some PoC implementation of it, so i can test it? | 10:13 |
| sylwesterB | evgenyl: off-top: https://review.openstack.org/#/c/266416/ spec for bareon. It's not visible in gate-bareon-specs-docs http://docs-draft.openstack.org/16/266416/1/check/gate-bareon-specs-docs/c54b3ac//doc/build/html/ | 10:25 |
| sylwesterB | do you know why? | 10:25 |
| *** sylweste_ has joined #openstack-bareon | 10:35 | |
| *** sylwesterB has quit IRC | 10:35 | |
| *** sylweste_ is now known as sylwesterB | 10:36 | |
| evgenyl | sylwesterB: hmm, I don't know why it's not rendered correctly... | 10:48 |
| evgenyl | sylwesterB: there is run.sh script and examples of schemas in /etc directory, after you run the script, you can take a look at the picture in /tmp/bareon.svg directory | 10:49 |
| sylwesterB | evgenyl: which repo you're talking about? | 10:56 |
| evgenyl | https://github.com/rustyrobot/bareon-allocator | 10:56 |
| evgenyl | sylwesterB: ^ | 10:56 |
| sylwesterB | great! Thanks ;) | 10:56 |
| *** asvechnikov has quit IRC | 11:46 | |
| *** asvechnikov has joined #openstack-bareon | 11:50 | |
| asvechnikov | evgenyl: Seems I misunderstood part about "Don’t allocate a single volume on ssd and hdd". I thought that partitions for one volume group shouldn't be allocated on hdd and ssd in the same time. So, Which solution will be used in case when user doesn't specify `best_with_disk` for volume group? | 11:54 |
| evgenyl | asvechnikov: for us all disk are equal, so the user (somebody who wants to write own dynamic schemas) has to explicitly specify the classification of disks and which class to use for which space, so if user doesn't specify best_with_disk it will be treated as a regular disk. | 11:57 |
| evgenyl | asvechnikov: so this algorithm doesn't have "a brain" which can say that one disk is ssd another one is hdd and it can make decisions according to that, it operates with spaces and classes of disks, so on layer above with yaql there is a way to make any kind of classification which is required for specific use-case | 12:00 |
| asvechnikov | evgenyl: Yes it's clear, but I tried to ask another thing. For example we have vg 'test' which contains two physical volumes pv1 and pv2, pv1 is allocated on hdd disk and pv2 is allocated on ssd disk. AFAIK it's a bad practice. Am I right? | 12:15 |
| *** asvechnikov has quit IRC | 12:25 | |
| *** asvechnikov has joined #openstack-bareon | 12:25 | |
| *** asvechnikov has quit IRC | 12:27 | |
| *** asvechnikov has joined #openstack-bareon | 12:27 | |
| *** asvechnikov has quit IRC | 12:29 | |
| *** asvechnikov has joined #openstack-bareon | 12:29 | |
| evgenyl | asvechnikov: correct, allocation of single logical volume over disks with different speeds in general is a bad practice. | 12:37 |
| asvechnikov | evgenyl: will solver have some mechanism to not do it? | 12:38 |
| evgenyl | asvechnikov: yes, user (i.e. sysops/developer) can set correct classification of disks for a space, so it won't make "bad allocation" (if min/max constraints are not violated), so something on level above should set correct classification according to speed of disks, for example it can be done dynamically using YAQL. | 12:44 |
| asvechnikov | evgenyl: Seems that classification should be set very accurately, otherwise there are high probability that solver doesn't find any solution | 12:47 |
| evgenyl | asvechnikov: not really, I've specifically used coefficients of objective function to make sure, that those allocation are not "must" requirements, so if according to provided constraints (min/max) the only solution is to allocate lv on disks with different speeds, it will be done this way, which is not the best, but better than nothing. Also in example with | 13:01 |
| evgenyl | ssd, if there are no ssd, disks set for ceph-jouranl will belong to default class and will be allocated on hdd, if min size is specified. | 13:01 |
| asvechnikov | evgenyl: oh, indeed, there are no zero coefficients, thanks | 13:07 |
| openstackgerrit | LiuNanke proposed openstack/bareon: Put py34 first in the envlist order of tox ,remove py33 https://review.openstack.org/266947 | 14:33 |
| evgenyl | agordeev: there are several patches on py34 addition and removing py26 https://review.openstack.org/#/q/project:openstack/bareon could you please consider merging them? Since Vladimir is seems to be not available at the time. | 14:35 |
| agordeev | evgenyl: yep, sure i'll do. let me take a look at them once again | 14:37 |
| openstackgerrit | Merged openstack/bareon: Put py34 first in the env order of tox, and remove py26/py33 https://review.openstack.org/260476 | 14:51 |
| *** sylwesterB has quit IRC | 15:45 | |
| *** tzn has quit IRC | 17:52 | |
| *** tzn has joined #openstack-bareon | 17:54 | |
| *** tzn has quit IRC | 17:58 | |
| *** ikalnitsky has quit IRC | 18:08 | |
| *** evgenyl has quit IRC | 18:08 | |
| *** agordeev has quit IRC | 18:08 | |
| *** kozhukalov has quit IRC | 18:08 | |
| *** agordeev has joined #openstack-bareon | 18:18 | |
| *** kozhukalov has joined #openstack-bareon | 18:18 | |
| *** evgenyl has joined #openstack-bareon | 18:21 | |
| *** ikalnitsky has joined #openstack-bareon | 18:31 | |
| *** tzn has joined #openstack-bareon | 22:53 | |
| *** tzn has quit IRC | 22:57 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!