19:00:09 #startmeeting Poppy Weekly Meeting 19:00:11 Meeting started Thu Aug 28 19:00:09 2014 UTC and is due to finish in 60 minutes. The chair is amitgandhinz. Information about MeetBot at http://wiki.debian.org/MeetBot. 19:00:12 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 19:00:14 The meeting name has been set to 'poppy_weekly_meeting' 19:00:32 #topic roll call 19:00:43 welcome, who do we have today for the poppy meeting? 19:00:53 0/ 19:00:58 o/ 19:01:00 guimarin from fastly 19:01:16 Fred from MaxCDN 19:01:36 hi guimarin 19:01:41 hi Fred 19:01:46 hey guimarin and Fred-MaxCDN ! 19:01:53 o/ 19:01:53 hi all! 19:02:12 Hey Guys 19:02:26 me here... 19:02:32 Hi Fred. 19:02:49 Hi guimarin 19:02:53 ok, most of us are here 19:03:06 fyi megan_w is out the next couple of meetings 19:03:38 #topic Review last weeks action items 19:03:51 following the agenda here: #link: https://wiki.openstack.org/wiki/Meetings/Poppy#Next_meeting 19:04:08 #link: http://eavesdrop.openstack.org/meetings/poppy_weekly_meeting/2014/poppy_weekly_meeting.2014-08-21-19.02.html 19:04:28 so action items from last week, going down the list... 19:04:48 1. amitgandhinz to investigate MaxCDN CDN Manager API for master/sub accounts 19:05:01 i still havent had the chance to do that, sorry 19:05:27 i'll add it on to my todo for the next week. 19:05:49 #action: amitgandhinz to investigate MaxCDN CDN Manager API for master/sub accounts 19:06:13 related to this, guimarin - do you have any updates on the fastly side around master/sub accounts 19:06:25 iirc some work had been done, and you were working on exposing apis? 19:06:40 just a second, had to put sandwich down 19:06:56 so have you guys got into contact with the other divs in rackspace about Fastly 19:07:12 my understanding since going back to my team is that you're building out an account management system on your side 19:07:24 using our existing api's 19:07:38 i can have edwardb who just joined reach out to megan_w about this 19:07:46 im not aware of those discussions. 19:07:54 edward-fastly can you comment on that 19:08:03 ok please cc me on that email also - megan_w is out for then next 10 days 19:08:07 Here - a little late - sorry. 19:08:18 hi catherine_r 19:09:12 #action guimarin, edward-fastly, and amitgandhinz to look into what "building out an account management system" is about. 19:09:31 guimarin: ok we can take that offline 19:09:35 sounds good 19:09:36 thanks 19:09:55 ok next action item: amitgandhinz to investigate bp permissions 19:10:18 after working with obul we found that anyone can create a bp. to be assigned a bp in launchpad, only the PTL can assign work 19:10:35 are there any other questions around this item? 19:10:48 FYI: we can edit the whiteboard, link bug reports etc to a bp 19:11:00 I think this is mostly what we need 19:11:16 Seems developer cannot change the state of a BP. 19:11:17 basically assignment is done by the PTL as they are the one who sets the direction of the project 19:12:00 tonytan4ever: agreed. its a little annoying, but basically the PTL has to change the state also. Through these meetings and other discussions in the #openstack-poppy channel 19:12:17 ok item 3 is the same as 2 19:12:29 4) obulpathi to get on the atlanta openstack meetup schedule for poppy 19:12:35 obulpathi: can you please comment on that 19:12:51 So I had a discussion with the organizers of the local openstack meetup 19:13:07 I proposed the poppy meeting for october 16 19:13:30 to discuss about poppy architecture and what it means for openstack and to gather early feedback from openstack users 19:13:38 its still going on 19:13:48 only one organizer has responded so far 19:14:08 hey amitgandhinz are you saying megan is out today and for 10 more days? 19:14:13 I will keep bugging them and schedule the talk 19:14:39 edward-fastly: megan_w returns on Sep 9 19:14:53 thats it for now on action item 4 19:15:02 ok thanks obulpathi 19:15:22 #action: obulpathi to keep bugging the atlanta openstack meetup organizers to get on the schedule 19:15:34 got it 19:16:00 #topic: Updates on Blueprints and Bugs 19:16:15 #link: https://blueprints.launchpad.net/poppy 19:16:48 how is the progress on bp/create-service, tonytan4ever 19:17:21 I just finished design discussion and started coding since yesterday. 19:17:40 ok, i'll leave the state as started 19:17:49 Sounds good to me. 19:18:09 #link: https://blueprints.launchpad.net/poppy/+spec/flavors 19:18:19 this one is in review 19:18:38 will review it 19:18:39 could use some review love 19:18:42 thanks obulpathi 19:18:50 :) 19:19:09 #link: https://blueprints.launchpad.net/poppy/+spec/poppy-models 19:19:18 poppy-models: https://blueprints.launchpad.net/poppy/+spec/poppy-models 19:19:19 this one is under review as well. 19:19:50 I got a few review comments, and I will address them ASAP. 19:20:16 ok thanks tonytan4ever. i had posted a large comment there. did those make sense? 19:20:24 Yes. 19:20:48 Ok cool. let me know if you need more discussion on this. we can talk about it in the normal channel 19:21:18 #link: https://blueprints.launchpad.net/poppy/+spec/add-docstrings 19:21:19 Sure. 19:21:29 ametts doesnt appear to be in the room 19:21:40 doesnt appear to be started yet 19:21:54 #link: https://blueprints.launchpad.net/poppy/+spec/health 19:21:59 obulpathi: any update on this 19:22:04 health end point is almost done 19:22:24 I got side tracked due to opencafe bug and cassandra testing 19:22:44 ok, lets try to get this one done (avoid sidetracking if possible lol) 19:22:44 so the status is still the same for that bp 19:22:47 ok 19:22:55 sure .. will get it done asap 19:23:01 do you have what you need to check for the health of provider api's? 19:23:35 right now I am just hitting cloudfront uri and checking if its alive 19:23:44 I need to do the same thing to fastly and maxcdn 19:24:19 I have not investigated that part yet 19:25:00 guimarin: Fred-MaxCDN: do your api's have a health endpoint to check if your api's are up and behaving? *i havent looks at the docs on that yet 19:25:13 s/looks/looked 19:25:24 you can subscribe at status.fastly.com 19:25:28 for that information 19:25:54 oh cool 19:25:59 I will check that 19:26:02 we use a third party service, but they have a 'webhook' 19:26:03 thanks guimarin 19:26:07 interface that you can have a deamon on 19:26:25 obulpathi: the rss feed maybe the way to go #link: http://status.fastly.com/history.atom 19:26:37 right if you want to scrape it 19:26:42 ok 19:26:44 but we have the ability to notify by email, sms 19:26:45 etc. 19:26:51 but if you go to the webhook notification 19:26:55 we have it on staus.maxcdn.com but we can add historical data if you wish 19:26:57 they will looks like send you a json call 19:27:38 guimarin: cool. 19:27:48 Fred-MaxCDN: is the a programmatic way to access the status? 19:27:50 thanks Fred-MaxCDN. will look in to status.maxcdn.com 19:27:55 and it looks like maxcdn has the same service as fastly 19:28:04 looks like they just need to enable that same json webhook thingy 19:28:12 fyi this is something that would be used by an operator to check all systems that poppy talk to are operating optimally 19:28:56 ok cool, they both look similar 19:29:21 great 19:29:23 alright, lets move on to the next topic 19:29:31 yes we have it. I need to get the public page up for you guys 19:29:51 #topic Load Testing tools and Best Practices 19:30:34 ok, what we are looking to do is deploy a bunch of different websites for testing (static sites, word press sites, file download sites, etc) 19:30:58 then enable CDN on them using Poppy with the various providers 19:31:24 then load test those sites to ensure that the cdn configurations are set up correctly and the cdn providers are handling the traffic at the edge 19:31:38 there are a number of tools out there, which malini1 is starting to investigate 19:32:01 when you say load test, do you mean accessing from servers around the world? 19:32:07 i am interested to hear from guimarin and Fred-MaxCDN on their recommended approaches 19:32:11 malini1: yes 19:32:26 so do you mean for setting up services or for testing them? 19:32:45 so accessing the sites from around the world, and also hitting the sites hard enough to ensure the edge is serving the traffic etc 19:33:08 guimarin: testing that the service was configured correctly and that the provider is doing what its supposed to 19:33:15 eg honoring restrictions/conditions 19:33:17 I am looking at http://www.webpagetest.org/, & seems like fastly/maxCDN already partner with them 19:33:21 actually service content from the edge 19:33:26 s/service/serving 19:33:37 right so those are two seperate things 19:33:41 we can use load impact , also catchpoint for global availability 19:34:24 Fred-MaxCDN: are those opensource tools? 19:34:34 no 19:34:35 right so checking that the CDN is working is different from how well is the CDN working 19:34:38 what i typically do is this 19:35:03 https://gist.github.com/guimarin/9280636 19:35:09 and piont it at the CDN and the origin 19:35:30 now you can get fancier if you want to spend money, but i find running this on VPSs around the world give me a pretty good idea 19:35:46 guimarin: what is a VPS? 19:35:54 virtual private server 19:36:03 linode/digital ocean/ec2 19:36:04 etc 19:36:18 thanks! 19:37:05 ok so that will help in testing the conditions from geo's etc 19:37:08 very cool 19:37:10 and simple! 19:37:20 guimarin: do u have any recommendation on how the VPSs are geographically distributed etc.? 19:37:44 so the thing about this 19:37:50 is that testing is done by like 4 companies 19:37:58 and it's really easy to er, modify the tests 19:38:06 so that' swhy i have this oss script 19:38:06 malini1: i think we would just spin up servers with a range of providers around the world depending on what we are testing 19:38:08 that people can use 19:38:29 what you really want is a huge Real user metrics RUM setup 19:38:54 i fyou want to do comparisons, etc. 19:39:13 bc then you see what actual users see, but the apachebenchmark test is free to set up and cheap to run on your own 19:39:21 I will ping you with more questions in #openstack-poppy 19:39:23 if you guys want some more info on this we can take it offline 19:39:32 im assuming if we didnt have RUM then we would have to have enough load generators and be careful of generating hotspots 19:39:38 I am familiar with ab, but just need to figire out the geo part 19:40:11 so webpagetest is another really interesting tool, but it also needs to have some toggles toggled to get something that you want as opposed to the default 19:40:34 webpagetest looked really interesting -mainly because of the geo footprint 19:40:49 I used to use multiple vps boxes but at most I could cover 10 or 20 location , it is good when you deal with large file delivery but for small file you need more locations. 19:41:15 yes it's open source and our cpo steve sauders says it's the best middle ground for people that want to run tests on verifiable systems 19:41:31 we also look at cedexis and catch point but you have topay for them 19:41:45 and their methodology needs to be understood so you undestand the limits of the test 19:41:55 eg. if the testing box is in teh same data center as the CDN pop it's kind of useless 19:42:32 ok, malini1 do you have enough info on this topic, we can talk about it more in the normal channel 19:42:38 thanks guimarin & Fred-MaxCDN! 19:42:45 yup thanks, really useful info! 19:42:46 yes amitgandhinz 19:42:59 sure thing 19:43:02 #topic: Log Delivery 19:43:14 ok, so this is a topic that i have started to investigate 19:43:49 basically i am looking at how the various providers are distributing the logs (so that poppy can retrieve them), and later use for both analytics, and metering 19:44:29 so far i have looked at fastly, maxcdn, and cloudfront api's 19:44:43 cloudfront dumps their logs into S3 buckets 19:45:03 fastly has a range of options (S3 included) - do you plan to support Swift/CloudFiles 19:45:07 Fastly: under the model we discussed, each customer with themselves seperate and their own service, each can stream real-time logs via syslog ( about 1s latency ) to a your endpoint, a bunch of third party endpoints, or over FTP ( minimum 1min latency ) to the same sinks. 19:45:21 and maxcdn appears to have a realtime api to query and get back a json response 19:45:38 we also have the s3 option 19:45:54 Fred-MaxCDN: ah, thanks i didnt see that option 19:45:55 sweet, real time analytics. 19:46:11 supporting cloudfiles is the goal 19:46:44 I think with poppy, it needs to be flexible on the sinks it provides, as each provider is bound to be different 19:47:00 +1 19:47:06 s3 is definitely a common one i am seeing 19:47:23 looking at the roadmap there is a DNS issue that needs to be resolved before we add cloudflies 19:47:50 my concern with providing an endpoint for syslog to stream to is the amount of info that will be streamed 19:48:10 guimarin: do you have data on the size of each log that is streamed? 19:48:52 that depends on how much you get to add 19:48:58 the logs are just apache common format logs 19:49:02 just streamed using syslog 19:49:07 true, i saw the format options 19:49:46 but back to the other point, i think defaulting to S3 FTP or similar is probably a good diea, but i would give people the option to do syslog streaming if they wanted 19:50:04 hmm...i will have to talk to our ops folks on the streaming concept though 19:50:04 especially if s3 support is so widespread thorugh all the providers 19:50:13 i like the concept, im just concerned about how fat the pipe will be 19:50:37 tha'ts reasonable, it will also be as fat as the # of qps 19:50:54 let me tell you the guys doing 10k qps are certainly getting a lot of logs 19:52:04 ok, i will talk to our ops team, but we can experiment with this 19:52:15 im only at the beginning of looking at the logging story 19:52:19 gzipping the logs before shoot could help 19:52:27 but im getting a good idea about it 19:52:39 Fred-MaxCDN: do you guys have plans to support CloudFiles/Swift also? 19:53:10 +1 for zipping logs 19:53:11 I need to check the dev roadmap 19:54:34 ok we have 5 minutes left... 19:54:34 just fyi fastly ftp log transfer which is one of the options we're discussing, currently supports gziping up to level 9 19:54:59 same with s3 19:55:19 cool. 19:55:22 that will def helpt 19:55:29 * amitgandhinz really cant type today 19:55:43 ok lets move on to the last agenda item 19:55:53 #topic: switching PUT to POST 19:56:08 so i kind of thought about this yesterday and am interested in feedback 19:56:24 currently when we PUT /service with pass the servicename in both the url and the body 19:56:27 seems redundant 19:56:39 its in the body to keep consistent with the GET which will return the servicename 19:56:50 same applies to /flavors 19:57:05 i propose we change it to POST /v1.0/services 19:57:13 and pass the servicename in the body 19:57:26 let the server decide the location (which will end up being v1.0/services/servicename 19:57:39 and then we no longer have duplication 19:57:42 thoughts? 19:57:52 +1 for change to POST with a shortened url. 19:58:12 looks like a good idea 19:58:35 malini1: ?? 19:58:58 ok 1 minute remaining =P 19:58:58 will get be GET /services/servicename? 19:59:03 obulpathi: yes 19:59:09 ok 19:59:35 ok i will document the change then 19:59:45 i will take malini1's silence as a vote for it =P 19:59:50 hahaha 19:59:57 ok time is up 20:00:04 thanks everyone, that was a lot to cover today 20:00:09 was very productive and helpful 20:00:15 #endmeeting