09:00:38 <oanson> #startmeeting Dragonflow
09:00:38 <openstack> Meeting started Mon Sep 19 09:00:38 2016 UTC and is due to finish in 60 minutes.  The chair is oanson. Information about MeetBot at http://wiki.debian.org/MeetBot.
09:00:40 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
09:00:43 <openstack> The meeting name has been set to 'dragonflow'
09:00:46 <oanson> All right. Who is here for the meeting?
09:00:57 <xiaohhui> hi
09:01:00 <lihi> o/
09:01:03 <hujie> Hi
09:01:23 <oanson> nick-ma, are you here?
09:02:14 <oanson> It looks like I forgot to set up an agenda, so we'll follow last week's, and I'll do better next time :)
09:02:22 <oanson> #topic Roadmap
09:02:41 <oanson> The good news is that ML2 fullstack tests are working
09:02:51 <oanson> And it looks like both ML2 and core plugin fullstack tests are stable
09:02:58 <xiaohhui> great
09:03:02 <oanson> That includes security groups.
09:03:16 <oanson> I think this is very good news. Thanks everyone for a great effort in this department.
09:03:43 <xiaohhui> oanson: have you insert some ml2 specific options for ml2 fullstack test?
09:04:06 <oanson> xiaohhui, yes, the fullstack-ml2 uses the mechanism driver + ML2
09:04:20 <oanson> But nothing beyond that.
09:04:24 <oanson> Is there something you think should be there?
09:05:17 <xiaohhui> I am checking the failed fullstack in ml2 for vlan network, I wonder where did the network_vlan_ranges set?
09:05:37 <xiaohhui> It looks like by default it will be empty
09:05:54 <oanson> IIRC (I looked into it a while back) it's a Neutron configuration. We may need to set it in plugin.sh, or understand what's Neutron's defaults.
09:06:30 <xiaohhui> neutron's default is an empty list...
09:07:10 <xiaohhui> https://github.com/openstack/neutron/blob/master/neutron/plugins/ml2/drivers/type_vlan.py#L35
09:07:33 <hujie> what about in /etc/neutron/plugins/ml2/ml2_conf.ini?
09:07:43 <oanson> Same for flat networks: https://github.com/openstack/neutron/blob/master/neutron/plugins/ml2/drivers/type_flat.py#L33
09:08:16 <oanson> Looks like '*' means any flat/vlan networks, and empty list to disable the feature
09:08:34 <oanson> '*' not supported for vlan
09:08:34 <xiaohhui> That is also my understanding
09:09:11 <xiaohhui> hujie: In my ml2 env, the ml2_conf.ini will not have network_vlan_ranges.
09:09:15 <oanson> For the tests, we can add a fake network, with a fake OVS bridge. We can maybe attach to it some tap devices to make sure the network connection works.
09:10:01 <oanson> And use set_override in the configuration for the tests
09:10:58 <oanson> I can help writing something if needed.
09:11:09 <xiaohhui> Yeah, we may need further look to the test. I am not sure if set_override will work in fullstack, because in fullstack, the test runs after everything is set up.
09:11:39 <oanson> Some code to create a br-<vlan network>, attach a tap device to it, attach it to br-int using the physical network code, and see that packets make it as expected
09:12:22 <oanson> We can update the local.conf generated by the fullstack test to include the relevant information
09:12:36 <oanson> And have plugin.sh read the information and update the configuration files.
09:12:47 <oanson> This is something we probably need in devstack anyway
09:12:48 <xiaohhui> yes, I think we should push some fake config to the ml2 fullstack env
09:13:40 <oanson> We can even have devstack attach automatically the external interfaces, but this should wait for a different patch
09:13:50 <oanson> (According to config, of course)
09:14:10 <xiaohhui> I remember devstack have such config for user to config vlan/flat. I just can't recall the details
09:14:29 <xiaohhui> it should be in a different patch
09:14:55 <xiaohhui> We can focus on the current patch to make it pass the fullstack first
09:15:05 <oanson> Yes.
09:15:27 <oanson> But we should align ourselves to devstack configuration. Either in this patch (if possible) or next patch if too complex.
09:15:43 <oanson> We can cut out any default configuration and construction for now.
09:16:13 <oanson> xiaohhui, is this reasonable? Or still too much?
09:16:29 <xiaohhui> I think it makes sense.
09:16:41 <oanson> Great! :)
09:17:02 <oanson> Is this item closed for now?
09:17:13 <xiaohhui> yeah
09:17:21 <oanson> Great. Thanks!
09:17:29 <oanson> The next item is port status notification
09:17:47 <oanson> I think jingting's patch is very close to being ready for merge, with a few minor issues
09:18:23 <jingting> yes, i will fix it soon
09:19:23 <oanson> Packaging will wait for Ocata. it isn't reasonable that it will be ready by the summit. I think there is enough work with VLAN/flat networks, port status, and the other things we're working on.
09:20:43 <oanson> The QoS code is also in review. I was mainly reviewing the North Bound, and I see nick-ma is reviewing the south bound as well.
09:22:01 <oanson> hujie, I see the db consistency patch stopped making progress.
09:22:09 <oanson> Are you still working on it>?
09:23:00 <hujie> sorry, I paid my most time working on HWS+dragonflow test currently
09:23:10 <oanson> I see.
09:23:22 <hujie> and I will try my best to working on it
09:23:33 <oanson> This means that patch https://review.openstack.org/#/c/331132/ is also halted for now (remote port function)?
09:23:37 <hujie> db consistency\remote port\bug fix
09:24:11 <hujie> I'll try to fix remote port this week
09:24:17 <oanson> Because I think it is very close to being finished, and it would be a strong feature for Newton (cloud inter-operability and the such)
09:24:22 <oanson> Great. Thanks!
09:24:49 <hujie> I'll try to working on db consistency next week:)
09:25:03 <oanson> Very good.
09:25:38 <oanson> #topic Performance
09:25:48 <oanson> yuli_s isn't in today
09:25:55 <oanson> But he sent me the results.
09:26:52 <xiaohhui> Is there any work item to improve performance?
09:27:02 <hujie> what's the result??
09:27:02 <oanson> In a test of 4000+ df controller nodes, 1 Neutron server, and redis spread over (I think, because I can't find it in the email) 8 servers
09:27:22 <oanson> xiaohhui, once we have results, we can decide if they are good enough.
09:27:32 <oanson> hujie, one sec, I am trying to read this cryptic email :)
09:27:39 <hujie> ok :)
09:27:53 <oanson> Creation of 50 subnets took ~49 seconds. That means ~1 subnet / second.
09:28:00 <oanson> Which I think is good results.
09:28:33 <oanson> Creating a port per subnet took 17 seconds. i.e. 50 ports in 17 seconds is almost 3 ports/second
09:29:13 <oanson> Creating 5 ports per subnet (in addition) takes ~86 seconds, which is also almost 3 ports per second (2.9).
09:29:22 <oanson> There were 250 ports created in that test.
09:29:24 <hujie> what's the performance data for native Neutron without dragonflow?
09:29:57 <oanson> I don't have that. I can look up the performance benchmarks for the reference implementation for next meeting.
09:30:17 <hujie> ok thanks
09:30:22 <oanson> If we'll manage to keep the environment, we may be able to compare with OVN, opendaylight, etc.,
09:30:36 <oanson> On the other hand, I am not sure this is where we should invest our time.
09:30:59 <oanson> I think it is important first and foremost to verify that we operate fast enough to be deployed in a cloud of thousands of nodes.
09:31:06 <oanson> IIRC, we wanted to reach 10000 :)
09:31:32 <oanson> But I find these numbers very hopeful.
09:32:06 <oanson> We don't have permanent servers for this, but the best thing would be to re-run this benchmark automatically on a weekly basis. But this is a long-term dream for now :)
09:32:24 <oanson> That's what I have for now.
09:32:57 <xiaohhui> Thanks oanson for sharing
09:33:10 <oanson> #topic Bugs
09:33:13 <hujie> do you want to add stress test?
09:33:44 <oanson> hujie, yes. What did you have in mind?
09:34:01 <oanson> I think the tempest and rally should do something like that as well, but they have to be stabalised.
09:34:56 <hujie> I just wonder whether the db cluster is stable enough in the stress test
09:35:46 <oanson> This is definitely something that needs to be tested.
09:35:54 <hujie> ok good thx
09:35:57 <oanson> This may also be the instability we see in the rally and tempest tests.
09:36:26 <hujie> ok let's move on please :)
09:36:31 <oanson> In bugs, I see that all high importance bugs are assigned
09:36:59 <oanson> There were a few Undecided, but I changed that.
09:37:35 <oanson> Since there is nothing critical, I think we can move on.
09:37:39 <oanson> #topic Open Discussion
09:37:44 <oanson> The floor is for the taking
09:38:51 <hujie> what's the most important feature for us to support in the future?
09:39:07 <xiaohhui> And is there a plan in O release?
09:39:18 <oanson> There are plans for O release.
09:39:36 <oanson> I think the most important thing for O release is deployment. I will make sure I have the time to work on this
09:40:04 <oanson> I am looking into openstack-ansible, and nick-ma mentioned that he may be looking into kolla. If anyone wants to look into openstack-puppet or Fuel, that would be great
09:40:11 <oanson> (But not urgent. That's for next cycle)
09:40:31 <oanson> Additionally, there are many Neutron features we can implement, e.g. FW, LB, VPN, etc.
09:40:40 <oanson> Full IPv6 support is important
09:41:03 <oanson> These are the options that immediately spring to mind
09:41:08 <xiaohhui> What about the snat?
09:41:19 <hujie> in the etc...
09:41:25 <oanson> xiaohhui, yes.
09:42:02 <oanson> Whatever potential deployers think is important. I would really like to get our user-base to grow
09:42:33 <oanson> I hope in the Barcelona summit to meet both deployers and developers, and get a clear picture of what are the priorities
09:42:34 <hujie> yes
09:43:14 <oanson> I also have a personal dream of SFC, but I don't know if that would fit Ocata. It may have to wait for P cycle, but I will see if I can catch the networking-sfc people in the summit as well.
09:44:41 <oanson> I will organise a list on etherpad, so we will have a more organised vision
09:45:12 <hujie> ok thx
09:45:26 <oanson> Of course, this is a democracy. If there is something you want, I would love to hear it!
09:46:07 <oanson> Once the etherpad will be ready, you could just add it there as we go.
09:46:21 <hujie> does Ashed still working on dragonflow?
09:47:08 <oanson> hujie, yes.
09:47:13 <hujie> ok good
09:47:27 <oanson> But he is taking a less technical role, and a more community role
09:47:45 <hujie> ok got it :)
09:48:03 <oanson> Anything else?
09:48:24 <xiaohhui> Nothing from me
09:48:35 <hujie> It's the supper time in beijing :)
09:48:35 <oanson> All right.
09:48:41 <oanson> The enjoy your meal.
09:48:50 <xiaohhui> Thanks~
09:48:52 <oanson> It's lunch time here, so we're hungry too :)
09:48:53 <hujie> thx :)
09:49:04 <oanson> Thanks for coming, and for a great effort!
09:49:08 <oanson> #endmeeting