| redbo | Well, yeah. We're trying to make a mod_python app not lock up entirely with a bunch of idle connections right now. Each connection just uses too many resources. | 00:04 |
|---|---|---|
| rdw | oh wow, yeah, mod_python | 00:05 |
| redbo | we have plans to phase it out, but it's being a pain point right now, so we're thinking about putting nginx in front of it so idle connections can just sit there all they want. | 00:07 |
| *** henriquetft_ is now known as henriquetft | 00:08 | |
| justinsb | eday: Sorry about delay. I got File Exists errors with eventlet. Where should I put the os.fork calls to get it to work? | 00:08 |
| justinsb | eday: Twisted improved marginally, but I'd argue that it wasn't worth the price. e.g. deferreds everywhere, multiple processes, etc | 00:09 |
| eday | justinsb: if you use the example on my blog, i put os.fork() right after the server = listen(...) line | 00:09 |
| justinsb | eday: If we're aiming for raw performance, we should use C++ (and ScaleStack!) But we're making a trade-off between productivity and performance... | 00:10 |
| eday | justinsb: yup, but as the other folks have been talking about, with threads you eat more memory and performance gets worse with more CPU bound loads because of the GIL | 00:10 |
| justinsb | eday: On the eventlet thing, I am working from your blog, and I put the os.fork right after the eventlet.listen line | 00:11 |
| eday | justinsb: this simple echo test helps the mulit-threaded case a bit since nothing is really happening in the threads | 00:11 |
| eday | justinsb: I should say, happening in the threads while in python interpreter | 00:12 |
| justinsb | eday: I just think we should find a benchmark where Twisted or Eventlet is worth the price before we adopt them. I was looking to find where the cut off point lay, and whether Jython helped, but I've changed my mind... | 00:12 |
| justinsb | Also, I agree that threading has a huge cost in terms of _virtual_ memory... but not necessarily real memory :-) | 00:13 |
| eday | justinsb: it's still more real memory too, but yeah, not as much as vm. but that difference may be enough to adopt eventlet | 00:15 |
| justinsb | eday: I think we should let the twisted proponents demonstrate the value of twisted. | 00:15 |
| eday | justinsb: work up some tests that does more processing of the buffered data in python. just xor all the bytes, one at a time, in the buffer or something | 00:16 |
| justinsb | eday: And in the absence of contrary evidence, we should adopt whatever avoids lock in. | 00:16 |
| justinsb | eday: So we can use Threads, Threads with GIL fix, Jython or Eventlet - whatever is best in 2 years | 00:16 |
| justinsb | eday: I'm sure we can construct a benchmark that can favor any of the tests :-) But the only benchmark that matters is our use case. | 00:17 |
| justinsb | eday: So we should write our use case so that we can pick the winner when one appears | 00:17 |
| justinsb | eday: So today's story is eventlet or threading based on your load | 00:17 |
| eday | justinsb: yup, but any useful use case is going to have more processing than echo :) | 00:17 |
| justinsb | eday: Except that nova-api is likely more I/O than anything else | 00:18 |
| justinsb | eday: So your test is probably a very accurate depiction of what nova-api will do! | 00:18 |
| eday | justinsb: hmm, not really. there is xml/json encoding/decoding, and then waiting on redis/message queues (which is more protocol packing before getting out to socket) | 00:19 |
| eday | justinsb: also the URL mapping/parsing to route requests, and WSGI layers (at least the new stuff) | 00:19 |
| redbo | what machine did you test threads vs eventlet with? I see a HUGE difference just pulling down a 512k file with http concurrently. | 00:20 |
| justinsb | eday: So that's what we'll have to benchmark. And the only realistic way to benchmark that is to use nova-api. And the only way to do that is to make sure that the same code can run against the various strategies. | 00:21 |
| justinsb | eday: Which is fine for evenlet | 00:22 |
| justinsb | redbo: I haven't run your benchmark yet.. if you send me the source code I can try it. I've just been running eric day's benchmark from here: http://oddments.org/?p=494 | 00:22 |
| redbo | on what sort of hardware and over what sort of network? | 00:23 |
| justinsb | redbo: I ran the benchmark on a quad-core over loopback. It's the same as eday ran, though I have 4 cores vs his 2. | 00:23 |
| *** rnewson has quit IRC | 00:24 | |
| *** blpiatt has quit IRC | 00:35 | |
| *** randybias has joined #openstack | 00:37 | |
| *** pvo has joined #openstack | 00:37 | |
| *** pvo has joined #openstack | 00:37 | |
| *** ChanServ sets mode: +v pvo | 00:37 | |
| *** randybias has quit IRC | 00:38 | |
| *** _anm has joined #openstack | 00:39 | |
| *** binaryWarrior has joined #openstack | 00:42 | |
| _anm | as an ISP, i'm thinking about what nova needs to do really well for me and my customers - who is working on network management of VMs in openstack? where/how to network configurations get administered from? how about VLAN management? | 00:42 |
| _anm | so far, in my trying to get nova running from source, the network is a very mysterious black box | 00:43 |
| *** binaryWarrior has quit IRC | 00:44 | |
| eday | _anm: a patch just went in that pushes the network config off to workers. This enables you to write different workers to handle your network differently | 00:45 |
| eday | For example, our needs (I work at rackspace) are different than nasa's, so we'll have a different networ setup | 00:45 |
| eday | _anm: start here for how it works today: http://nova.openstack.org/network.html | 00:45 |
| _anm | eday: very good. my first thought is that my end user and admin management wrapper will be a web.py or django framework, and I'll store my network configs in a database, but that may be way off the mark | 00:47 |
| _anm | eday: is there work to make network settings for vm's persistent, so one "assigned" from my management wrapper, they survive vm rebots, migrations, etc? | 00:49 |
| eday | _anm: yeah. right now they are DHCP based, we're working on making them persistent (a feature we need) | 00:50 |
| *** pvo has quit IRC | 00:52 | |
| _anm | as a hosting company, that will be one of the most critical pieces - i run a modest xen based cluster now for VMs, and network management is difficult to manage | 00:52 |
| _anm | i "lose" customrt VMs all the time | 00:52 |
| _anm | i literally don;t know where they are | 00:53 |
| *** nettok has joined #openstack | 01:00 | |
| mtaylor | justinsb: good email, btw | 01:58 |
| *** jaypipes has quit IRC | 02:03 | |
| *** Jim_Shew has joined #openstack | 02:50 | |
| *** Jim_Shew has quit IRC | 02:58 | |
| *** Jim_Shew has joined #openstack | 04:04 | |
| _anm | where is documentation needed the most? | 04:06 |
| _anm | my attempt at keeping a current "install from source" page is rough | 04:07 |
| _anm | source is moving sooo fast, and I don't know how to keep up with changes | 04:07 |
| _anm | how do I clean out the redis db while hacking nova? It is full of instance data that needs to be blown away | 04:34 |
| _anm | just delete /var/lib/redis/dump.rdb? | 04:34 |
| _anm | also, WHY am I up at 23:35 on a saturday working on this? | 04:35 |
| *** _anm is now known as nikmartin | 04:36 | |
| nikmartin | redis-cli FLUSHALL cleans up the redis-server database | 04:38 |
| *** Jim_Shew has quit IRC | 05:04 | |
| nikmartin | is torndo still required in nova/trunk? | 05:09 |
| nikmartin | tornado | 05:09 |
| comstud | yeah | 05:11 |
| comstud | the api is tornado still | 05:11 |
| nikmartin | ok | 05:13 |
| *** henriquetft has quit IRC | 05:17 | |
| *** rajijoom has joined #openstack | 05:37 | |
| *** jaker74 has quit IRC | 06:01 | |
| *** jamiew has joined #openstack | 06:49 | |
| *** jtdowney has joined #openstack | 06:59 | |
| nikmartin | i notice rabbitmq consumes an awful lot of cpu and ram | 07:03 |
| eday | nikmartin: there are still some optimizations to be done around queuing | 07:04 |
| nikmartin | yeah, th eqiki says rabbitmq may be temporary | 07:04 |
| nikmartin | wiki | 07:04 |
| nikmartin | dang i'm tired | 07:04 |
| *** nikmartin is now known as nickmartin-away | 07:09 | |
| *** allsystemsarego has joined #openstack | 07:23 | |
| *** rajijoom has quit IRC | 07:30 | |
| *** sophiap has quit IRC | 07:31 | |
| *** sophiap has joined #openstack | 07:38 | |
| *** kuttan_ has joined #openstack | 07:40 | |
| *** kuttan_ has left #openstack | 07:42 | |
| *** sophiap_ has joined #openstack | 07:46 | |
| *** kuttan_ has joined #openstack | 07:48 | |
| *** sophiap__ has joined #openstack | 07:48 | |
| *** sophiap has quit IRC | 07:48 | |
| *** sophiap__ is now known as sophiap | 07:49 | |
| *** jaker has joined #openstack | 07:52 | |
| *** sophiap_ has quit IRC | 07:52 | |
| *** chmouel has left #openstack | 08:17 | |
| *** chmouel has joined #openstack | 08:17 | |
| *** nettok has quit IRC | 08:18 | |
| *** rnewson has joined #openstack | 09:19 | |
| *** kuttan_ has left #openstack | 09:36 | |
| *** metoikos has joined #openstack | 09:36 | |
| *** jamiew has quit IRC | 10:03 | |
| *** jtdowney has quit IRC | 10:29 | |
| *** nettok has joined #openstack | 10:39 | |
| *** nettok has quit IRC | 11:04 | |
| *** kuttan_ has joined #openstack | 12:29 | |
| *** tjyang has joined #openstack | 12:35 | |
| *** tjyang has joined #openstack | 12:38 | |
| *** nickmartin-away is now known as nikmartin | 12:52 | |
| nikmartin | so, which api is the "native" nova api, from which would prefer to use in my attempt at a modest web.py based UI for nova? | 13:00 |
| nikmartin | so far, Ive found the makings of a rackspace api, aws ec2 api, and bits and pieces of some others | 13:06 |
| *** rnewson has quit IRC | 13:26 | |
| *** rnewson has joined #openstack | 13:26 | |
| *** rnewson has quit IRC | 14:59 | |
| *** kuttan_ has left #openstack | 15:02 | |
| *** rnewson has joined #openstack | 15:05 | |
| *** metoikos has quit IRC | 15:08 | |
| *** sophiap has quit IRC | 16:00 | |
| eday | nikmartin: ec2 api curretly is the way, we're actively working on the rackspace version (which will evolve into the "OpenStack API") | 16:13 |
| nikmartin | eday: so, adminclient.py is where i should go? | 16:30 |
| *** zul has quit IRC | 16:31 | |
| eday | nikmartin: you probably want to use the boto package for talking to nova, but yeah, some things are wrapped in adminclient too | 16:39 |
| nikmartin | ok, i see references to boto, which is a python api for ec2 - got it | 16:40 |
| nikmartin | so, my web.py ui will use boto to talk to nova-api? | 16:41 |
| nikmartin | Id hate to spend much time wrapping up boto only to have the ec2 api ripped out of nave, maybe I'll wait for the native api that i can use directly | 16:44 |
| *** jamiew has joined #openstack | 16:49 | |
| eday | nikmartin: it probably won't be ripped out anytime soon (if ever), but the focus is certainly shifting to an API we have control over :) | 16:52 |
| nikmartin | That's waht I mean i guess, I don't think nova will be ready for production before the openstack api is well documented, giving guys like me time to develop a UI and assist in documentation on the openstack wiki | 16:54 |
| nikmartin | so i may just keep helping with testing, and just wiring up some other parts of the UI/control system; the billing interface, and the UI itself with a support database | 16:55 |
| *** rnewson has quit IRC | 16:56 | |
| *** rnewson has joined #openstack | 17:06 | |
| *** rnewson has joined #openstack | 17:06 | |
| *** sophiap has joined #openstack | 17:07 | |
| *** hornbeck has quit IRC | 17:18 | |
| *** rnewson has quit IRC | 17:34 | |
| *** rnewson has joined #openstack | 17:35 | |
| *** jamiew has quit IRC | 17:40 | |
| *** RobertLJ has joined #openstack | 17:53 | |
| *** metoikos has joined #openstack | 18:16 | |
| *** rajijoom has joined #openstack | 18:19 | |
| *** binaryWarrior has joined #openstack | 18:39 | |
| *** binaryWarrior has quit IRC | 18:44 | |
| *** binaryWarrior has joined #openstack | 18:45 | |
| *** binaryWarrior has quit IRC | 18:48 | |
| *** binaryWarrior has joined #openstack | 18:56 | |
| *** rajijoom has quit IRC | 18:57 | |
| *** jonesy_ has quit IRC | 19:04 | |
| *** rajijoom has joined #openstack | 19:37 | |
| *** heckj has joined #openstack | 19:49 | |
| *** stewart has quit IRC | 19:54 | |
| *** rajijoom has quit IRC | 20:02 | |
| *** heckj has quit IRC | 20:07 | |
| *** BartVB has quit IRC | 20:12 | |
| *** kim0 has joined #openstack | 20:22 | |
| *** rnewson has quit IRC | 20:49 | |
| *** rnewson has joined #openstack | 20:50 | |
| *** rnewson has joined #openstack | 20:50 | |
| *** stewart has joined #openstack | 21:04 | |
| *** sophiap has quit IRC | 21:16 | |
| *** RobertLJ has quit IRC | 21:28 | |
| *** jtimberman has joined #openstack | 21:52 | |
| kim0 | hi, what's the minimum number of machines I can get openstack running on for playing and hacking (ideally only my laptop) ? | 21:53 |
| *** binaryWarrior has quit IRC | 21:58 | |
| *** HyperJohnGraham has left #openstack | 22:05 | |
| *** stewart has quit IRC | 22:06 | |
| *** allsystemsarego has quit IRC | 22:14 | |
| *** binaryWarrior has joined #openstack | 22:15 | |
| *** Daviey has quit IRC | 22:16 | |
| *** Daviey has joined #openstack | 22:21 | |
| chmouel | kim0: for swift there is the swift-aio 'all in one' guide on the wiki which as the name suggest allow you to playd with only one VM with swift | 22:24 |
| kim0 | chmouel: I'm more interested in nova I think :) | 22:25 |
| kim0 | http://nova.openstack.org/getting.started.html doesn't mention if all nodes can be the same box | 22:26 |
| kim0 | since http://www.joinfu.com/2010/07/developing-nova-on-linux-getting-started/ runs all tests successfully .. does this mean all components have been started on a single box ? | 22:28 |
| justinsb | kim0: It's definitely possible to run everything on one box | 22:28 |
| justinsb | kim0: (For nova) | 22:28 |
| kim0 | great to hear | 22:28 |
| justinsb | kim0: Indeed, I think some things might not work on one box | 22:28 |
| *** sandywalsh has joined #openstack | 22:28 | |
| justinsb | kim0: Sorry, might not work on more than one box :-) | 22:28 |
| kim0 | :D | 22:29 |
| kim0 | I was confused :) | 22:29 |
| justinsb | kim0: The one issue you might have... what OS do you want to run it on? | 22:29 |
| kim0 | so nova was never deployed to more than one box ? | 22:29 |
| kim0 | ubuntu 10.10 | 22:29 |
| justinsb | kim0: Ubuntu 10.10 should work great - I think that's what most of the developers run | 22:29 |
| kim0 | justinsb: so I sould follow the getting-started doc on one box .. thanks | 22:30 |
| justinsb | kim0: Well I know NASA has it deployed on multiple boxes | 22:30 |
| justinsb | kim0: But one box is definitely simplest :-) | 22:30 |
| kim0 | absolutely cool | 22:30 |
| *** jonesy_ has joined #openstack | 22:32 | |
| *** sandywalsh has quit IRC | 22:32 | |
| *** nikmartin has quit IRC | 22:49 | |
| *** binaryWarrior has quit IRC | 22:52 | |
| *** rnewson has quit IRC | 23:04 | |
| *** kim0 has quit IRC | 23:12 | |
| *** miclorb_ has joined #openstack | 23:15 | |
| *** rdw has quit IRC | 23:28 | |
| *** binaryWarrior has joined #openstack | 23:31 | |
| *** sophiap has joined #openstack | 23:43 | |
| *** dmd17 has joined #openstack | 23:44 | |
| *** justinsheehy has quit IRC | 23:52 | |
| *** rdw has joined #openstack | 23:57 | |
| *** justinsheehy has joined #openstack | 23:59 | |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!