Tuesday, 2016-11-08

openstackgerritMerged openstack-infra/shade: Don't fail on trying to delete non-existant images  https://review.openstack.org/38870200:33
*** abregman has joined #openstack-shade06:06
*** abregman has quit IRC07:59
*** abregman has joined #openstack-shade08:03
*** abregman has quit IRC08:27
*** abregman has joined #openstack-shade08:36
*** abregman is now known as abregman|mtg08:36
*** abregman_ has joined #openstack-shade09:08
*** abregman|mtg has quit IRC09:11
openstackgerritArie Bregman proposed openstack-infra/shade: Add compute usage support  https://review.openstack.org/39387311:22
openstackgerritArie Bregman proposed openstack-infra/shade: Add compute usage support  https://review.openstack.org/39387312:23
*** abregman_ is now known as abregman|mtg13:08
*** abregman|mtg is now known as abregman13:24
*** wowbagger42 has joined #openstack-shade13:34
*** openstackgerrit has quit IRC13:48
*** openstackgerrit has joined #openstack-shade13:48
*** abregman is now known as abregman|brb14:32
*** abregman|brb is now known as abregman14:45
*** edtubill has joined #openstack-shade15:04
*** wowbagger42 has quit IRC15:37
*** edtubill has quit IRC16:21
*** edtubill has joined #openstack-shade16:28
openstackgerritMonty Taylor proposed openstack/oaktree: Add a few more docs  https://review.openstack.org/39507616:39
*** abregman has quit IRC16:45
*** edtubill has quit IRC17:30
*** edtubill has joined #openstack-shade17:31
mordredShrews, SpamapS: I just replied to https://review.openstack.org/#/c/393884 - could y'all check it out and see if you agree?18:16
* Shrews reading the novella18:19
*** clarkb has joined #openstack-shade18:21
mordredShrews: I also sent a quick summary to the infra list18:21
Shrewsmordred: SpamapS: so, shade has always tried as best it can to hide cloud differences, but i don't think we've ever promised to make everything the same, especially b/w nova and neutron clouds. we *definitely* haven't promised to implement functionality where lacking. I'm fine with the change as-is.18:22
*** abregman has joined #openstack-shade18:22
Shrewsfor example, shade would NEVER try to implement egress rules for a nova-based cloud18:23
mordredyes. totally agree18:23
mordredI think my main point is that we _can_ make the filtering interface the same for the user for both in this case, it's just that it'll be less efficient for nova-net which I'm personally fine with18:24
clarkbthat, plus its already transparent to the user via the interface you just get nasty fail surprise errors when it breaks which is badness18:24
clarkbif it was an explicit "nova net here and neutron here" fine treat them different but that isn't how its setup, instead its a one size fits all "filter things for me" so we should make it work for both18:25
mordredyah, but I'm also fine with it not happening this instant, because the change to add it later isn't going to break anyone, and the change itself doens't change behavior for current users- it only introduces a new feature that only works for half of the users so far18:25
Shrewsyeah, i don't think we've ever said you can always filter all the things, but we should try to as best we can. i'm fine with it being added piecemeal18:28
mordredI'm more concerned about making sure that normalization and push-down conditions don't step on each other18:29
clarkbShrews: yes there isn't an explicit statement to that effect but the "fail hard with an exception if you don't do the right thing which is completely opaque to the user" is what i htink we should avoid18:29
clarkbbecause if user moves around to different clouds (hey tahts me!) it is incredibly frustrating when I have to go debug failures that show up in 1 of 10 clouds18:29
*** abregman is now known as abregman|nb18:30
mordredtotally. that's why I think the error needs to say "please use search_floating_ips instead of list_floating_ips with a filter for now"18:30
mordredsince search_floating_ips will _always_ work for all of the clouds today18:30
mordredand it's only the list_floating_ips push-down case if explicitly triggered by an end user where there is an issue18:30
Shrewswhy do we have filtering on "list_*" anyway? the original purpose of "list" apis was to list all of them, and "search" apis could filter18:31
Shrewsi'm not sure how the filtering in list* started...18:31
clarkbya mordred brought htat up in the email, basically they should be combined18:31
mordredfiltering on list is so that filters from search can be passed in to list_18:31
mordredbecause if we want to do server-side pushdowns, we have to pass them to list18:31
Shrewshrm, we did that wrong18:32
mordredwhich, now that we've gotten this far shows that actually there's no real difference between the two, since we have to plumb the filters into list anyway18:32
mordredyah18:32
mordredwe didn't do this the right way for sure18:32
Shrewsdamn you, hindsight18:33
*** abregman|nb has quit IRC18:41
mordredShrews: biggest most important question to sort now - which of list_ or search_ do we want to be the 'real' api and which one do we want to be the alias?18:44
mordredShrews: I'd like for my shed to be a nice purple color18:44
Shrewsmordred: the obvious answer is to create new "query_*" apis and have the other two be aliases. Also, green.18:45
mordredmmm. green18:45
jlkmordred: funny, your name in my client is purple.18:45
Shrewseven funnier, in mine, he's green18:45
clarkbI would go with list because that seems to be what a lot of other clients libs do18:45
clarkbwhich will make life a little simpler for anyone moving from eg neutronclient/novaclient18:45
Shrewsmordred: so, not all list_'s filter, but all search_'s do. we'd have to change the list_'s to all filter before we could promote that18:46
Shrewsand clarkb ^^^18:47
clarkboh thats a good point18:47
mordredShrews: yes. I mean, as long as the filter arguments are optional and don't filter anything when missing, it shoudl be a no-op change18:47
clarkbmaybe shade is sufficiently its own thing to say just do search then18:47
mordredoh - but yeah. doy. I understand your words now18:48
Shrewsgood, b/c i did not understand yours18:48
*** root4 is now known as olaph18:48
Shrewsfwiw, i think list_ is the better choice, but requires the code change18:49
Shrewsthen we'd deprecate the search_ apis for a few version18:50
SpamapSclarkb: I understand the "don't fail on arbitrary cloud differences" desire. However, is running 1000x slower not also a failure?18:54
clarkbSpamapS: not in this case where nova net already could be filtered client side18:54
clarkbSpamapS: so I think that ship sailed?18:54
clarkbso this is about making things consistent with what we already have18:55
clarkbnot introducing any new slowness18:55
mordredalso - to be fair to the optimization problem - nodepool does full lists on all things with no pushdowns - and we have a TON of resources on our clouds and it's actually a thing we do to make things faster18:55
clarkband if you do think thats an issue refactor it to make it abundantly clear that one thing is supported and the other is not18:55
mordredso making nova-net fip filtering client-side only, while slower, actually shouldn't be noticable in most cases18:55
clarkblike I siad my biggest concern is its all behind the covers and only a month after writing your code do you discover oh this is completely wrong18:56
SpamapSso let me understand: does shade already filter fips?18:56
mordredyes. client-side18:56
mordredif you use the search_floating_ips method18:57
SpamapSSo the concern is that this new functionality only works for neutrons?18:58
clarkbSpamapS: in a confusing manner because it works over here but not over there and there is no documentation or clear error message for that18:58
clarkbSpamapS: so my suggestion to address that was why not just make it work since it already works over here18:59
clarkband in the process make the code much simpler18:59
clarkbalternate method would be to beef up documentation and error messages18:59
SpamapSI think landing it half-functional is ok. It's not _that_ confusing, but agreed, land docs or the full functionality.19:03
openstackgerritClark Boylan proposed openstack-infra/shade: Add unit test to show herd protection in action  https://review.openstack.org/39212019:09
clarkb^ failed to merge because it depended on old patchset19:09
clarkbrebased and reapproved19:09
mordredclarkb: thanks!19:10
*** abregman has joined #openstack-shade19:55
*** abregman has quit IRC20:49
openstackgerritMerged openstack-infra/shade: Add unit test to show herd protection in action  https://review.openstack.org/39212020:49
openstackgerritMerged openstack-infra/shade: Allow server-side filtering of Neutron floating IPs  https://review.openstack.org/39388421:52
openstackgerritMerged openstack-infra/shade: Use floating-ip-by-router  https://review.openstack.org/39000421:52
openstackgerritMerged openstack-infra/shade: Update floating ip polling to account for DOWN status  https://review.openstack.org/39001421:52
openstackgerritMerged openstack-infra/shade: Refactor out the fallback-to-router logic  https://review.openstack.org/39461321:52
*** edtubill has quit IRC22:14
openstackgerritMerged openstack/oaktree: Add a few more docs  https://review.openstack.org/39507622:34
openstackgerritArie Bregman proposed openstack-infra/shade: Add support for limits  https://review.openstack.org/39523522:44
openstackgerritMonty Taylor proposed openstack/oaktree: Clean a couple of meaningless things  https://review.openstack.org/39524123:00
thingeemordred the requirements repo has this set differently for hacking in test-requirements https://review.openstack.org/#/c/395241/23:37
mordredthingee: yah - there's a patch either up or coming to fix it23:40
mordredthingee: or maybe I'm lying. there was a conversation in infra this morning about it being wrong/too loose23:41
openstackgerritMerged openstack/oaktree: Clean a couple of meaningless things  https://review.openstack.org/39524123:42
openstackgerritMonty Taylor proposed openstack/oaktree: Manual sync with global requirements  https://review.openstack.org/39525823:49
mordredthingee: ^^ did a sync with g-r by hand23:49
thingeemordred setup.py should be pbr>=1.6 instead of 1.8 ?23:53

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