opscamp lightning talks - by sponsors? first paid commerical annoucements! the rule of two feet not explained Alot of cloud cloud cloud OpsCode (John from OpsCode) (deployment, management) ------- Why opscode? - Ops guy named Bob who dies (hit by bus, i presume) - operations should be code -- eclipse, cookbooks, SCM, etc - Do you have Bob? - Internet 10 (top sites) -- infra as code - shows lamp stack -- define those as nodes, roles, etc -- apply recipes - example is scaling wordpress architecture -- cookbook for WP -- include recipe for apache2 -- etc, etc. - list of offereings -- Chef -- Cookbooks [what about state/schema upgrades, service managment, etc]? rPath (shawn edmonson) (deployment, management, lifecycle of app) ---------------------- - drafts in emacs, draws in pencil - lots of open source, but people don't know what it does - Priciples -- Drive change from a complete model -- Version control everything -- best way to deal with change - Open source tech called Canary -- binary artifact revision control -- everything goes in it (patches, OSes, configs, applications, middleware) --- build manifest out of this -- can make images from this for rapid provisioning (hardware, VMDK, AMI) -- incremental updates/patches from manifests -- audit compliances (diff from what is out there and the model) -- promote through lifecycle (dev->test->prod) - sits underneath Chef - unified hierarchal model (?) -- used to build system diversity [what about state/schema upgrades, service managment, etc]? [can you push environents]? Dyn, Inc. (Lisa Hagemann) ------------------------- - lots of topics, her fav is disaster response -- don't forget about the DNS --- they've been doing DNS for 11 years -- some stats on their DNS work -- more sales data -- integrates with Chef - handles GSLB, CDN, Anycast, traffic management, etc - show some data about saving time by having someone else management of DNS -- customer saved 100ms off DNS resolution time - "talk to people who think DNS is sexy" (they have shirts) [very salesy] [but they sponsored the booze] Phase 2 (still no rule of two feet, we're all locked in at the rec center) "Unpanel" -------- Panel is Matt from Hubspot, Jay from Brightcove, Vladimir from mocospace, Cory from Dyn * What happens when all the ops jobs move to India? * How would cloud adoption effect outsource of ops? - answering both question, Jay says - problem with outsources is that you make the problem of ops/dev division much worse, brightcove has - operations system and development team all wearing pagers, all involved in deploy/automation, all respond to outages. If you have no interactions you can't do this. [moderator says that mature companies aren't the same] [lots of discussion on this, not taking those notes] [this will be broken out] * Why won't my ops people self provision like I can from the cloud? - Matt from Hubspot, still need standardization, change control, actually gives developers more control if there is automation - Vladimir adds that the best thing to do is discuss what can be done, who is responsible, what techniques can you do to make each others' life easier [comment from audience] - the way we made it work is that developers need complete understanding of version/feature impact, provably * Patch or rebuild? Push new images? Vladimir is a patch man, rebuilding can be to disruptive, but depends on the application design. [audience] - do both, depends on the software rPath guy disagrees, sees it all patches [audience] [breakout session is called for] * Cost of ops in IT? What is the trend for that? IT assets to people admining them? [will be a breakout] * What is the connection between the talks we just heard and the cloud? - cory says most policy/process hasn't changed, but work is different (API, self provisioning vs. hardware acquisition) * Mixing dedicated infrastructure with other public cloud resources? Should the management be central? vladiimir says - you should manage it together matt says - not only servers, but VPNs, etc, etc rPath joins the conversations and takes a seat on the panel and talks about rPath doing this (common API between all provisioning) Moved on to session planning ---------------------------- Future of operations employees - arguments, arguments, arguments - apollo is very excited about ops being a profit center -- revenue production from ops -- people are using automation to reduce cost -- some places don't have development teams -- more arguing Why can't devs just pull a credit card out and deploy to amazon? ("Why should I let the developers ruin my business? -or- Why are ops people assholes?") - ran this, so don't have notes Provisioning in the cloud ------------------------ - will we get an infrastructure deployment? DNS, authentication, monitoring, etc. -- Chef does this - do we patch, or do we build new? - when is it okay to drift from the gold standard of an image? -- when the business case for the software is there -- -- in research in a small environment they add new software to old servers by need -- how quickly do these things change? -- majority of changes are not emegency patching, they are "routine" -- discussions on dependency management, chef, puppet, cfengine