imbrandon | zomg , jcastro you round ? | 02:51 |
---|---|---|
* imbrandon just figured out a nginx loadbal config that allows the site to do 600+ concurrent req/s AND stay under 50ms ( yes 50 not 500 ) response time | 02:53 | |
* imbrandon does a lil dance | 02:53 | |
imbrandon | ( oh, off a m1.small still ) | 02:54 |
=== flepied_ is now known as flepied | ||
SpamapS | so quiet in here all weekend | 06:29 |
SpamapS | imbrandon: clearly you have been drinking :) | 06:29 |
imbrandon | nah | 06:29 |
imbrandon | lol | 06:29 |
imbrandon | why is that tho ? | 06:29 |
SpamapS | figured maybe you were a quiet drunk | 06:30 |
imbrandon | :) | 06:32 |
flepied | any help needed with charms ? | 06:33 |
imbrandon | SpamapS: check this hottness out | 06:33 |
imbrandon | http://paste.ubuntu.com/942087/ | 06:33 |
imbrandon | put that bad boy in the /etc/nginx/loadbalancer config and boom , sweetness enchews | 06:34 |
imbrandon | even works with admin logins :) | 06:34 |
imbrandon | well chang the IP's to something other than my internal ones but yes | 06:35 |
imbrandon | next trick is gonna see what the limit on the cache size is, so i can throw /mnt/nginx on a ramdisk | 06:36 |
imbrandon | safely | 06:36 |
imbrandon | and no, i am a quiet coder, loud thinker, been coding alot this weekend ( php ) , more than normal | 06:38 |
imbrandon | SpamapS: think we could try to psudo implment that new way of storage at UDS ? i'm keen on getting it done but i havent grasp the juju codebase yet, mostly due to me still not clear on a few of the basics and trying to run before i walk but I'm sure 10 min of facetime with marcoceppi, you, or someone else will clear that right up at uds ( just how i initially learn things i guess ) | 06:42 |
imbrandon | oh and Go rocks, not as much like php as i first thought, but still rocks | 06:42 |
* imbrandon goes back to his php code for the moment :) | 06:44 | |
imbrandon | SpamapS: oh and my non scientific test do show that for some strange reason unix:/some/php.sock is indeed slower ( well not slower but fails faster with 503's ) than 127.0.0.1:9000 , no idea WHY yet, like if its open file descriptors or what is the issue because that just _seems_ wrong in my gut, but yea | 06:47 |
SpamapS | why would you do a ramdisk? can't nginx just cache in RAM? | 06:47 |
imbrandon | ummm not certain, probably not without a sep module like the memcache module | 06:47 |
imbrandon | it does most everything filebased by default and most things are modules | 06:48 |
SpamapS | memcache as in, memcached, or it has another cache module for RAM cache? | 06:48 |
imbrandon | that said there likely is noe | 06:48 |
imbrandon | one* | 06:48 |
imbrandon | memcached | 06:48 |
imbrandon | only used as a example cuz its a sep module in nginx | 06:48 |
SpamapS | that would be silly | 06:49 |
imbrandon | what would ? | 06:49 |
imbrandon | that it can only do file based in core ? | 06:49 |
imbrandon | nginx tries to be very very slim in code and make all extras modules, even some of the stuff we use like proxy is technicaly a module and not in core proper | 06:50 |
imbrandon | its like php and all its blah.so extensions | 06:50 |
SpamapS | imbrandon: well if its one big file like varnish does, that would make sense... you can just use VFS then | 06:50 |
imbrandon | but they are done at compile time only iirc | 06:50 |
imbrandon | nah its tons of small files | 06:50 |
* SpamapS notes that his laptop has but 10 minutes of battery left | 06:50 | |
SpamapS | imbrandon: thats a huge fail | 06:51 |
SpamapS | IMO | 06:51 |
SpamapS | closing a file == syncing a file | 06:51 |
imbrandon | like /mnt/nginx/[1-9]/[a=f] | 06:51 |
imbrandon | a-f | 06:51 |
SpamapS | yeah, so thats nice for simpler coding, but you dn't get to take much advantage of vfs | 06:51 |
imbrandon | well yea i'm sure there are better ways , but thats just how i set it up to begin with , see how its specified on line .... | 06:52 |
imbrandon | proxy_cache_path /mnt/nginx levels=1:2 keys_zone=microcache:5m max_size=1000m; | 06:52 |
imbrandon | so i'm sure there is optinos to that line where we could make it even more uber | 06:52 |
imbrandon | but even like that it did 625 concurrent users and 0 errors or time outs from blitz.io | 06:53 |
imbrandon | and stayed under 100ms , mostly 50ms | 06:53 |
imbrandon | response time from clai , and the server was in vegas | 06:53 |
imbrandon | s/was/is | 06:53 |
imbrandon | cali* | 06:53 |
SpamapS | imbrandon: nice. | 06:54 |
imbrandon | but yea , that was just the "refrence" line from the wiki, i'm sure we can tweak it lots more | 06:54 |
imbrandon | and research a little etc | 06:54 |
SpamapS | ok, battery is about to go dead even with powertop showing 8.0W of power usage | 06:54 |
imbrandon | heh ok, gnight | 06:55 |
SpamapS | imbrandon: should just thieve varnish's mmap method | 06:55 |
imbrandon | still in san fran ? | 06:55 |
imbrandon | kk | 06:55 |
SpamapS | nooo | 06:55 |
imbrandon | lol | 06:55 |
SpamapS | I was only there from the morning to evening | 06:55 |
imbrandon | ahh | 06:55 |
SpamapS | kids.. wife.. all get in the way of fun stuff like conferences ;) | 06:56 |
imbrandon | cool cool, i got some stuff to do early in the morning, but i've already woken for the day so just ping me later and i'll show yta the whole config set | 06:56 |
imbrandon | that i'm working on | 06:56 |
imbrandon | hahaha yea | 06:56 |
imbrandon | btu yea that was a great start i thought, and its a dropin for our current LB config on omg | 06:57 |
imbrandon | but* | 06:57 |
SpamapS | need to start thinking in more generic terms | 06:58 |
imbrandon | so really anything like apache or rails could sit on the other side of that upstream in this config too | 06:58 |
SpamapS | I want to get omg's specific stuff into a subordinate | 06:58 |
SpamapS | anyway | 06:58 |
imbrandon | yea i have been, i've been making sure all the stuff is sep | 06:58 |
SpamapS | battery.. dead.. later | 06:58 |
imbrandon | infact even the web and lp i want sep | 06:58 |
imbrandon | later | 06:58 |
imbrandon | lb* | 06:58 |
imbrandon | hrm , yea infact i'm going to make this into a sub charm right now | 06:59 |
imbrandon | its in a perfect state to do so | 06:59 |
imbrandon | as a nginx loadbal microcache proxy | 07:00 |
* imbrandon looks for that charm init command | 07:00 | |
=== amithkk is now known as Andrew__ | ||
=== Andrew__ is now known as AndrewStrong | ||
=== AndrewStrong is now known as ubuntu__ | ||
=== ubuntu__ is now known as xubuntu | ||
=== xubuntu is now known as a | ||
=== a is now known as b_ | ||
=== b_ is now known as amithkk | ||
marcoceppi | imbrandon: I hope that's in the omg-wp repo | 11:09 |
imbrandon | i'm actually commiting it now | 11:09 |
imbrandon | or trying to . | 11:09 |
imbrandon | had some conflicts | 11:09 |
imbrandon | but yea | 11:09 |
imbrandon | it is or will be in less than 5 min | 11:09 |
imbrandon | marcoceppi: good weekend ? | 11:09 |
marcoceppi | imbrandon: excellent | 11:09 |
marcoceppi | yes, lots of little wordpress tweaks, but if this does what it needs to we'll be golden | 11:10 |
imbrandon | marcoceppi: yea i got lots of impovements i've tested over the weekend while it was quite, bad part is i'm not QUITE ready like inches away though, but i got somewhere to be this morning till like lunch | 11:10 |
marcoceppi | That's fine, we can wait to deploy later this afternoon | 11:11 |
imbrandon | nice, and i was lookign too its best if we do use totalcache to use apc backend not file, then nginx dont have to be touched as far as total cache, its just a normal setup + our other tweaks at that point | 11:11 |
* imbrandon heads to the other room | 11:12 | |
jcastro_ | m_3: jamespage when we make the juu blueprints, it needs to be "servercloud-juju-blah-blah" | 13:05 |
jcastro_ | m_3: I have 3 charm growth blueprints I will file too. | 13:06 |
=== jcastro_ is now known as jcastro | ||
jamespage | jcastro, servercloud-q-juju-blah-blah please :-) | 14:07 |
jcastro | jamespage: you don't need the q, but whatever | 14:14 |
jamespage | jcastro: hmm - has that changed? | 14:15 |
jcastro | lp tracks the UDS series, it's just people always put in the letter anyway | 14:16 |
jcastro | you don't need it, but if you want to do it that way that's fine too | 14:16 |
jamespage | robbiew specifically asked us todo it that way for the servercloud track | 14:16 |
jcastro | oh ok then, do what he tells you | 14:17 |
jamespage | jcastro, BTW can you change the message for this channel - might be a little out-of-date 'Charm content ends monday!...' | 14:18 |
=== jcastro changed the topic of #juju to: Charms at http://jujucharms.com || Want to write a charm? http://juju.ubuntu.com/Charms | ||
jamespage | jcastro, ta! | 14:32 |
jcastro | jamespage: you were correct on the animal letter btw, mail sent to -devel again | 14:33 |
jamespage | jcastro, glad to be of nit-picking service! | 14:34 |
jcastro | robbiew: hey, can we put juju backports in the shiny new cloud repository? | 15:32 |
jcastro | that would be cool I think | 15:32 |
robbiew | jcastro: No, I'd prefer folks use the PPAs and/or Backports for juju | 15:35 |
jcastro | ok | 15:35 |
robbiew | I would have pushed openstack the same way | 15:35 |
robbiew | but there will be times when folks want to freeze on a certain version | 15:35 |
robbiew | I guess the same could be said about juju | 15:36 |
robbiew | jcastro: tbh, there's really no difference between the cloud archive and a ppa | 15:37 |
robbiew | just naming | 15:37 |
robbiew | but people get worried when you ask them to add a ppa :P | 15:38 |
jcastro | ok, I was just wondering if delivering juju with that at the same time would be a good idea like milestone wise, etc. | 15:38 |
jcastro | "you not only get the backend, you get the front piece too!" | 15:38 |
robbiew | juju needs a cadenced release first ;) | 15:39 |
jcastro | well I was hoping that would make us do that, heh | 15:39 |
SpamapS | robbiew: won't the cloud archive also use the regualar ubuntu signing key? | 16:27 |
SpamapS | jcastro: juju is going to have something similar, but it should not be coupled to openstack. | 16:28 |
robbiew | SpamapS: I guess...it's just a PPA mirrored off archive.ubuntu.com ;) | 16:29 |
SpamapS | robbiew: so they'll still have to add an apt key | 16:31 |
SpamapS | which, IMO, would freak people out *more* than a PPA | 16:31 |
robbiew | SpamapS: your opinion in the minority ;) | 16:32 |
robbiew | it's not about the tech...it's about the "name" | 16:33 |
robbiew | Personal Package Archive scares people | 16:33 |
robbiew | I admit it might not be logical...but neither are people :P | 16:33 |
jcastro | https://bugs.launchpad.net/charms/+bug/966484 | 16:34 |
_mup_ | Bug #966484: New Charm: Kusaba X <new-charm> <Juju Charms Collection:Fix Committed> < https://launchpad.net/bugs/966484 > | 16:34 |
jcastro | https://bugs.launchpad.net/charms/+bug/964936 | 16:34 |
_mup_ | Bug #964936: Drupal Charm: superchared Drupal charm with nginx,, apc, php-fpm, all setup to scale to the moon and be Best Practices. <new-charm> <Juju Charms Collection:Fix Committed by imbrandon> < https://launchpad.net/bugs/964936 > | 16:34 |
jcastro | 2 in the queue | 16:34 |
jcastro | SpamapS: we can get these in by thursday right? For some sort of cycle-ending WOO? | 16:34 |
SpamapS | robbiew: I suppose the one win is that you'll get it as part of the mirroring of the archive and won't have to poke a hole in a firewall to get to ppa's | 16:41 |
robbiew | exxxxxactly ;) | 16:42 |
SpamapS | jcastro: need to be available to do 0-day SRU reviews but I'll try to squeeze these in | 16:46 |
=== carif_ is now known as carif | ||
_mup_ | juju/retry-client-for-cli r532 committed by kapil.thangavelu@canonical.com | 18:57 |
_mup_ | use retry client for cli | 18:57 |
SpamapS | negronjl_: hey, how would you feel about moving your REST api into juju-jitsu? | 18:57 |
avoine | negronjl: great works by the way (jrapi) | 19:24 |
=== hspencer is now known as hspencer[afk] | ||
flepied | is there any plan to support dynamic creation of vm in juju based on load or other metrics ? | 19:55 |
SpamapS | flepied: you can do that now. stick the client on a thing that knows what the load is, and when load goes up, 'juju add-unit X' | 19:58 |
flepied | SpamapS, any example of this ? | 19:59 |
SpamapS | flepied: none I've seen, but its such a simple concept.. you can do it with icinga/nagios | 20:00 |
flepied | SpamapS, seems easy, I'll try | 20:03 |
flepied | another question is there any plan to use chef/puppet with juju ? | 20:03 |
SpamapS | flepied: I created puppet charms last week... | 20:04 |
SpamapS | flepied: and some people have toyed with using puppet to assert the system state they want from install/etc. | 20:04 |
SpamapS | flepied: juju really doesn't want to define that.. it executes stuff ... you choose what it executes :) | 20:04 |
flepied | SpamapS, yes I got that. Just wanted to see if this was going to be the way to write charm in the future or not... | 20:06 |
SpamapS | flepied: If we define a single way, we will be losing all the existing goodness that people have for their services. | 20:12 |
SpamapS | flepied: the main advantage of juju is that we can connect these unrelated stacks to form a new, bigger stack | 20:13 |
flepied | SpamapS, ok makes sense but you also need to ease the integration of charms together so having standard ways to do thing could help | 20:15 |
SpamapS | flepied: right, we've been doing a lot of shell scripts so they're easy to read for that reason. | 20:16 |
flepied | SpamapS, is there any way I can help ? I have some spare time... | 20:18 |
SpamapS | flepied: are you an expert on any service that might be useful in the cloud? | 20:25 |
SpamapS | flepied: if so, write a charm... using puppet or chef would be a bonus so people can see how that works | 20:25 |
SpamapS | flepied: one method I think that will work is to use the facter custom facts plugin to have a charm's hooks store relation/config settings data, and then just keep running the puppet ruleset using those facts. | 20:26 |
flepied | SpamapS, ok I'll look at that | 20:28 |
SpamapS | flepied: Also the charms we have now are all kind of untested.. we need people to put them through the paces and apply real world scaling/configurability to them | 20:29 |
SpamapS | flepied: oh, and if nobody said this to you yet.. WELCOME! :) | 20:31 |
flepied | SpamapS, thx :) | 20:31 |
flepied | SpamapS, I'll try to see how to test charms. I saw there is a ftest service running | 20:33 |
SpamapS | flepied: yeah that just bootstraps and runs a few of the basic charms | 20:37 |
flepied | do you have a jenkins server testing the charms ? | 20:37 |
=== lamont` is now known as lamont | ||
flepied | SpamapS, from this table http://jujucharms.com/interfaces, it seems monitoring (nagios) and ngnix (gunicorn) cannot work as they require something that is not provided anywhere... | 20:46 |
SpamapS | flepied: actually we just landed a feature which will enable that | 20:47 |
SpamapS | flepied: its called 'subordinate charms' and we need to work on the documentation a bit :) | 20:47 |
SpamapS | flepied: re charms in jenkins, 'http://charmtests.markmims.com' .. but there is one running precise charms now too | 20:48 |
SpamapS | m_3: ^^ I can has precise, preeeaassseeee | 20:48 |
flepied | SpamapS, yes I experimented it the other day and it's working fine | 20:48 |
SpamapS | flepied: also nagios I think can be modified to be able to relate to anything since there is now a 'juju-info' relation for every service | 20:50 |
flepied | SpamapS, yes that's the trick to avoid having explicit relations | 20:52 |
SpamapS | flepied: though I think there should be a general 'monitoring' interface which can feed back basic instructions on what to monitor... | 20:53 |
SpamapS | flepied: like, mysql can tell nagios to monitor it, and if the nagios charm has mysql specific monitors, it will use them | 20:53 |
flepied | SpamapS, yes that'd cool | 20:54 |
SpamapS | flepied: make it generic enough and we can sub in newer, more scalable monitoring stuff :) | 20:55 |
* SpamapS pushes support for subordinates into charm-tools | 20:55 | |
SpamapS | m_3: why didn't I think of this. the charmtester should run 'charm proof' against all charms as well | 20:56 |
SpamapS | bac: hrm, I feel like we left something undone with charm helpers and your buildbot stuff | 22:03 |
bac | SpamapS: gmb and i have a couple of proposed branches and there was the packaging issue. | 22:04 |
bac | SpamapS: it is on our radar but we haven't bugged you about it assuming you were swamped ATM | 22:05 |
SpamapS | bac: I seem to recall that we wanted to get python-shelltoolbox packaged on its own | 22:06 |
bac | SpamapS: yes, and portable to lucid | 22:07 |
bac | as needed by our buildbot slaves inside the containers | 22:08 |
xmltok | when using juju, are people using chef/puppet to do the base system configuration? I understand writing charms to do an application, but when configuring the system (ldap, accounts, mounts, etc), it seems like chef/puppet make more sense then a bunch of charms. I'd think that there should be a charm to deploy/configure/run chef/puppet, right? | 22:09 |
SpamapS | xmltok: There's a new way to deploy puppet/chef/??? along side your charms to do things like system policy. | 22:10 |
SpamapS | xmltok: http://jujucharms.com/search?search_text=puppet | 22:10 |
SpamapS | xmltok: I'm clint-fewbar btw ;) | 22:10 |
xmltok | aha! perfect | 22:11 |
SpamapS | http://jujucharms.com/~clint-fewbar/precise/puppet | 22:11 |
SpamapS | xmltok: its still just a proof of concept though. | 22:11 |
xmltok | exactly what i was thinking | 22:11 |
xmltok | oh hey cool, mod_spdy charm. nice | 22:12 |
SpamapS | xmltok: yeah that needs some work too.. it just assumes 127.0.0.1:80 will have a service to forward to :) | 22:12 |
=== hspencer[afk] is now known as hspencer[mtg] |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!