Month: January 2015

CoreOS: an introduction – Brandon Philips

• Reference to the “Datacenter as a Computer Paper“
• Intro to containers
• cAdvisor – API of what resources are used by a container
• Rocket
  • Multiple implementations of container spec , rocket is just one implementation
• Operating system is able to make less promises to applications
• Kernel API is really stable
• Making updates easy
  • Based on ChromeOS
  • Update one partition with OS version. Then flip over to that.
  • Keep another partition/version ready to fail back if needed
  • Safer to update the OS seperated from the app
  • Just around 100MB in size. Kernel, very base OS, systemd
• etcd
  • Key value store over http (see my notes from yesterday)
  • multiple, leader election etc
  • Individual server less critical since data across multiple hosts
• Scheduling stuff to servers
  • fleet – very simple, kinda systemd looking
  • fleetctl start foo.service   – sends it off to some machine
  • meso, kubernetes, swam other alternative scedulers
• Co-ordination
  • locksmith
• Service discover
  • skydns, discoverd, conf
  • Export location of application to DNS or http API
  • Need proxies to forward request to the right place (for apps not able to query service discovery directly)
• It is all pretty much a new way of thinking about problems

 

Why you should consider using btrfs, real COW snapshots and file level incremental server OS upgrades like Google does. – Marc Merlin

• Worked at netapp, hooked on snapshots, lvm snapshots never worked too well , also lvm partitions not too good
• Switched laptop to btrfs to 3 years ago
• Why you should consider btrfs
  • Copy on Write
  • Snapshots
  • cp -reflink=always
  • metadata is redundant and checksummed, data checksummed too
  • btrfs underlying filesystem [for now]
  • RAID 0, 1, 5, 6 built in
  • file compression is also built in
  • online background scrub (partial fsck)
  • block level filesystem diff backups(instead of a slow rsync)
  • convert difectly from ext3 (fails sometimes)
• Why not use ZFS instead
  • ZFS more mature than ZFS
  • Same features plus more
  • Bad license. Oracle not interested in relicensing. Either hard to do or prfer btrfs
  • Netapp sued sun for infringing patents with ZFS. Might be a factor
  • Hard to ship a project with it due to license condistions
• Is it safe now?
  • Use new kernels. 3.14.x works okay
  • You have to manually balance sometimes
  • snapshots, raid 0 , raid 1 mostly stable
  • Send/receive mostly works reliably
• Missing
  • btrfs incomplete, but mostly not needed
  • file encryption not supported yet
  • dedup experimental
• Who use it
  • openSUSE 13.2 ships with it by default
• File System recovery
  • Good entry on bfrfs wiki
  • btrfs scrub, run weekly
  • Plan for recovery though, keep backups, not as mature as ext4/ext3 yet, prepare beforehand
  • btrfs-tools are in the Ubuntu initrd
• Encryption
  • Recommends setup encryption on md raid device if using raid
• Partitions
  • Not needed anymore
  • Just create storage pools, under them create sub volumes which can be mounted
  • boot: root=/dev/sda1  rootflags=solvol=root
• Snapshots
  • Works using subvolumes
  • Read only or read-write
  • noatime is strongly recommended
  • Can sneakily fill up your disk “btrfs fi show” tells you real situation. Hard to tell what snapshots to delete to reclaim space
• Compression
  • Mount option
  • lzo fast, zlib slower but better
  • if change option then files changed from then on use new option
• Turn off COW for big files with lots of random rights in the middle. eg DBs and virtual disk images
• Send/receive
  • rsync very slow to scan many files before copy
  • initial copy, then only the diffs. diff is computed instantly
  • backup up ssd to hard drive hourly. very fast
• You can make metadata of file system at a different raid level than the the data
• Talk slides here. Lots of command examples

 

Bob Young

• Warns that some stories might not be 100% true
• ”  Liked about Early Linux – Nobody was very nice to each other but everybody was very respectful of the Intel Microprocessor “
• CEO of Redhat 1992 – 2000
• Various stories, hard to take notes from
• One person said they walked out of the Keynote when they heard the quote “it was a complete meritocracy” re the early days of Linux.
• Others didn’t other parts of the talk. General tone and some statements similar to the one above.
• “SuSe User Loser” proviked from laughs and a Suse Lizzard being thrown at the speaker
• Reasons the publishing industry rejects books: 1. no good; 2. market not big enough; 3. They already publish one on the subject.

Alerting Husbandry – Julien Goodwin

• Obsolete alerts
  • New staff members won’t have context to know was is obsolete and should have been removed (or ignorened)
• Unactionable alerts – It is managed by another team but thought you’d like to be woken up
• SLA Alerts – can I do something about that?
• Bad thresholds ( server with 32 cores had load of 4 , that is not load ), Disk space alerts either too much or not enough margin
• Thresholds only redo after complete monitoring rebuilds
• Hair trigger alerts ( once at 51ms not 50ms )
• Not impacting redundancy ( only one of 8 web servers is down )
• Spamming alerts, things is down for the 2925379857 time. Even if important you’ve stopped caring
• Alerts for something nobody cares about, eg test servers
• Most of earlier items end up in “don’t care” bucket
• Emails bad, within a few weeks the entire team will have a filter to ignore it.
• Undocumented alerts – If it is broken, what am I supposed to do about it?
• Document actions to take in  “playbook”
• Alert acceptance practice, only oncallers should e accepting alerts
• Need a way to silence it
• Production by Fiat

 

 

Managing microservices effectively – Daniel Hall

• Step one – write your own apps
• keep state outside apps
• not nanoservices, not milliservices
• Each should be replaceable, independantly deployable , have a single capability
• think about depandencies, especially circular
• Packaging
  • small
  • multiple versions on same machine
  • in dev and prod
  • maybe use docker, have local registry
  • Small performance hit compared to VMs
  • Docker is a little immature
• Step 3 deployment
  • Fast in and out
  • Minimal human interaction
  • Recovery from failures
  • Less overhead requires less overhead
  • We use Meso and marathon
  • Marathon handles switches from old app to new, task failure and recover
  •  Early on the Hype Cycle
• Extra Credit Sceduling
  • Chronos within Mesos
  • A bit newish

 

Corralling logs with ELK – Mark Walkom

• You don’t want to be your bosses grep
• Cluster Elastisearch, single master at any point
• Sizing best to determine with single machine, see how much it can hadle. Keep Java heap under 31GB
• Lots of plugins and clients
• APIs return json. ?pretty makes it looks nicer. The ” _cat/* ” api is more command line
• new node scales, auto balancers and grows automatic
• Logstash. lots of filters, handles just about any format, easy to setup.
• Kibana – graphical front end for elastisearch
• Curator, logstash-forwarder, grokdebugger

FAI — the universal deployment tool – Thomas Lange

• From power off to applications running
• It is all about installing software packages
• Central administration and control
• no master or golden image
• can be expanded by hooks
• plan your installation and FAI installs the plan
• Boot up diskless client via PXE/tftp
• creates partitions, file systems, installs, reboots
• groups hosts by classes, mutiple classes per host etc
• Classes can be executables, writeing to standard output, can be in shell, pass variables
• partitioning, can handle LVM, RAID
• Projected started in 1999
• Supports debian based distributions including ubuntu
• Supports bare metal, VM, chroot, LiveCD, Golden image

 

Documentation made complicated – Eric Burgueno

• Incomplete, out of date, inconsistent
• Tools – Word, LibreOffice  -> Sharepoint
• Sharepoint = lets put this stuff over here so nobody will read it ever again
• txt , markdown, html. Need to track changes
• Files can be put in version control.
• Mediawiki
• Wiki – uncontrolled proliferation of pages, duplicate pages
• Why can’t documentation be mixed in with the configuration management
• Documentation snippits
  • Same everywhere (mostly)
  • Reusable
• Transclusion in mediawiki (include one page install another)
• Modern version of mediawiki have parser functions. display different content depending on a condition
• awesomewiki.co

Mass automatic roll out of Linux with Windows as a VM guest – Steven Sykes

• Was late and missed the start of the talk

etcd: distributed locking and service discovery – Brandon Philips

• /etc distributed
• open source, failure tolerant, durable, watchable, exposed via http, runtime configurable
• API – get/put/del  basics plus some extras
• Applications
  • Locksmith, distributed locks used when machines update
  • Vulcan http load balancer
• Leader Election
  • TTL and atomic operations
  • Magical stuff explained faster than I can type it.
  • Just one leader cluster-wide
• Aims for consistence ahead of raw performance

 

Linux at the University – Randy Appleton

• No numbers on how many students use Linux
• Peninsula Michigan
• 3 schools
• Michigan Tech
  • research, 7k students, 200CS Students, Sysadmin Majors in biz school
  • Linux used is Sysadmin courses, one of two main subjects
  • Research use Linux “alot”
  • Inactive LUG
  • Scripting languages. Python, perl etc
• Northern Michigan
  • 9k students, 140 CS Majors
  • Growing CIS program
  • No Phd Programs
  • Required for sophomore and senior network programming course
  • Optional Linux sysadmin course
  • Inactive LUG
  • Sysadmin course: One teacher, app of the week (Apache, nfs, email ), shell scripting at end, big project at the end
  • No problem picking distributions, No problem picking topics, huge problem with desperate incoming knowledge
  • Kernel hacking. Difficult to do, difficult to teach, best students do great. Hard to teach the others
• Lake Superior State
  • 2600 students
  • 70 CS Majors
  • One professor teaches Sysadmin and PHP/MySQL
  • No LUG
  • Not a lot of research
• What is missing
  • Big power Universities
  • High Schools – None really
  • Community college – None really
• Usage for projects
  • Sometimes, not for video games
• Usage for infrastructure
  • Web sites, ALL
  • Beowuld Clusters
  • Databases – Mostly
• Obstacles
  • Not in High Schools
  • Not on laptops, not supported by Uni
  • Need to attract liberal studies students
  • Is Sysadmin a core concept – not academic enough
• What would make it better
  • Servers but not desktops
  • Not a edu distribution
  • Easier than Eclispe , better than visual studio

Untangling the strings: Scaling Puppet with inotify – Steven McDonald

• Around 1000 nodes at site
• Lots of small changes, specific to one node that we want to happen quickly
• Historically restarting the puppet master after each update
• Problem is the master gets slow as you scale up
• 1300 manifests, takes at least a minute to read each startup
• Puppet internal caching very coarse, per environment basis (and they have only one prod one)
• Multiple environments doesn’t work well at site
• Ideas – tell puppet exactly what files have changed with each rollout (via git, inotify). But puppet doesn’t support this
• I missed the explan of exactly how puppet parses the change. I think it is “import” which is getting removed in the future
• Inotify seemed to be more portable and simpler
• Speed up of up to 5 minutes for nodes with complex catalogs, 70 seconds off average agent run
• implementation doesn’t support the future parser, re-opening the class in a seperate file is not supported
• Available on github. Doesn’t work with current ruby-inotify ( in current master branch )

 

 

Configuration Management – A love Story – Javier Turegano

• June 2008 – Devs want to deploy fast
• June 2009 – git -> jenkins -> Puppet master
• But things got pretty complicated and hard to maintain
• Remove puppet master, puppet noop, but only happens now and then lots of changes but a couple of errors
• Now doing manual changes
• June 2010 – Thngs turned into a mess.
• June 2011 – Devs want prod-like development
• Cloud! Tooling! Chef! – each dev have their own environment
• June 2012 – dev environments for all working in ec2
• dev no longer prod-like. cloud vs datacentre, puppet vs chef , debian vs centos, etc
• June 2013 – More into cloud, teams re-arranged
• Build EC2 images and deploy out of jenkins. Eaither as AMI or as rpm
• Each team fairly separate, doing thing different ways. Had guilds to share skills and procedures and experience
• June 2014 – Cloudformation, Ansible used by some groups, random

Healthy Operations – Phil Ingram

• Acquia – Enterprise Drupal as a service. GovCMS Australian Federal Government. 1/4 are remote
• Went from working in office to working from home
• Every week had phone call with boss
• Talk about thing other than with work, ask home people are going, talk to people.
• Not sleep, waking up at night, not exercising, quick to anger and negative thinking, inability to concentrate
• Hadn’t taken more than 1 week off work, let exercise work, hobbies was computer stuff
• In general being in Ops not as much of an option to take time off. Things stay broke until fix
• Unable to learn via Osmosis, Timing of handing over between shifts
• People do not understand that computers are run by people not robots
• Methods: Turn work off at the end of the day, Rubber Ducking, exercise

Developments in PCP (Performance Co-Pilot) : Nathan Scott

• See my slides from yesterday for intro to PCP
• Stuff in last 12 months
  • Included in supported in RHEL 6.6 and RHEL 7
  • Regular stable releases
  • Better out of the box experience
  • Tackling some long-standing problems
• JSON access – pmwebd , interactive web charts ( Graphite, grafana )
• zero-install look-inside containers
• Docker support but written to allow use by others
• Collectors
  • Lots of new kernel metrics additions
  • New applications from web devs (memcached, DNS, web )
  • DB server additions
  • Python PMDA interfaces
• Monitor work
  • Reporting tools
  • Web tools, GUIs
• Also improving ease of setup
• Getting historical data from sar, iostat
• www.pcp.io

Security options for container implementations – Jay Coles

• What doesn’t work: rlimits, quotas, blacklisting via ACLs
• Capabilities: Big list that containers probably shouldn’t have
• Cgroups – Accounting, Limiting resource usage, tracking of processes, preventing/allowing device access
• App Armor vs selinux – Use at least one, selinux a little more featured

Last spoke 10 years ago in Canberra Linux.conf.au

Things have improved in the last ten years

• $10s of billions of value have been lost in software patent war
• But things have been so bad that some help was acquired, so worst laws have been pushed back  a little
• “Fear of God” in industry was enough to push open Patent pools
• Judges determined that Patent law was getting pathological, 3 wins in Supreme court
• Likelihood worst patent laws will be applied against free software devs has decreased
• “The Nature of the problem has altered because the world has altered”

The Next 10 years

• Most important Patent system will be China’s
• Lack of rule of law in China will cause problems in environment of patents
• Too risky for somebody too try and stop a free software project. We have “our own baseball bat” to spring back at them

The last 10 years

• Changes in Society more important changes in software
• 21st century vs 20th century social organisations
  • Less need for hierarchy and secrecy
  • Transparency, Participation, non-hierarchical interaction
• OS invented that organisation structure
• Technology we made has taken over the creation of software
• “Where is BitKeeper now?” – Eben Moglen
• Even Microsoft reorganises that our way of software making won
• Long term the organisation structure change everywhere will be more important than just it’s application in Software
• If there has been good news about politics = “we did it”, bad news = “we tried”

Our common Values

• “Bridge entire environment between vi and emacs”

Snowden

• Without PGP and free software then things could have been worse
• The world would be a far more despotic place if PGP was driven underground back in 1993. Imagine today’s Net without HTTPS or SSH!
• “We now live in the world we are afraid of”
• “What stands between them and us is our inventions”
• “Freedom itself depends on how we make use of the technologies we are creating.” – Eben Moglen
• “You can’t trust what you can’t read”
• Big power in the wrong is committed against the first law of robotics, they what technology to work for it.
• From guy in twitter – “You can’t trust what you can’t read.” True, but if OpenSSL teaches us anything you can’t necessarily trust what you can
• Attitudes in under-18s are a lot more positive towards him than those who are older (not just cause he looks like Harry Potter)
• GNU Project is 30 years old, almost same age is Snowden

Oppertunity

• We can’t control the net but opportunity to prevent others from controlling it
• Opportunity to prevent failure of freedom
• Society is changing, demographics under control
• But 1.6 billion people live in China, America is committed to spying, consumer companies are committed to collecting consumer information
• Collecting everything is not the way we want the net to work
• We are playing for keeps now.

 

 

AWS OpsWorks Orchestration War Stories – Andrew Boag

• Autoscaling too slow since running build-from-scratch every time
• Communications dependencies
• Full stack rebuild in 20-40 minutes to use data currently in production
• A bit longer in a different region
• Great for load testing
• If we ere doing again
  • AMI-based better
  • OPSWorks not suitable for all AWS stacks
  • Golden master for flexable
• Auto-Scaling
  • Not every AMI instance is Good to Go upon provisioning
  • Not a magic bullet, you can’t broadly under-provision
  • needs to be throughly load-tested
• Tips
  • Dual factor authentication
  • No single person / credentials should be able to delete all cloud-hosted copies of your data
• Looked at Cloudformation at start, seemed to be more work
• Fallen out of love with OpsWorks
• Nice distinction by Andrew Boag: he doesn’t talk about “lock-in” to cloud providers, but about “cost to exit”.   – Quote from Paul

 

Slim Application Containers from Source – Sven Dowideit

• Choose a base image and make a local version (so all your stuff uses the same one)
• I’d pick debian (a little smaller) unless you can make do with busybox or scratch
• Do I need these files? (check though the Dockerfile) eg remove docs files, manpages, timezones
• Then build, export, import and it comes all clean with just one layer.
• If all your images use same base, only on the disk once
• Use related images with all your tools, related to deployment image but with the extra dev, debug, network tools
• Version the dev images
• Minimise to 2 layers
  • look at docker-squash
  • Get rid of all the sourc code from your image, just end up with whats need, not junk hidden in layers
• Static micro-container nginx
  • Build as container
  • export as tar , reimport
  • It crashes 🙁
  • Use inotifywait to find what extra files (like shared libraries) it needs
  • Create new tarball with those extra files and “docker import” again
  • Just 21MB instead of 1.4GB with all the build fragments and random system stuff
  • Use docker build as last stage rather than docker import and you can run nginx from docker command line
  • Make 2 tar files, one for each image, one in libs/etc, second is nginx

 

Containers and PCP (Performance Co-Pilot) –  Nathan Scott

• Been around for 20+ years, 11 years open source, Not a big mindshare
• What is PCP?
  • Toolkit, System level analysis, live and historical, Extensible, distributed
  • pmcd daemon on each server, plus for various functions (bit of like collectd model)
  • pmlogger, pmchart, pmie, etc talk (pull or poll) to pmcd to get data
• With Containers
  • Use –container=  to grab info inside a container/namespace
  • Lots of work still needed. Metrics inside containers limited compared to native OS

 

The Challenges of Containerizing your Datacenter – Daniel Hall

• Goals at LIFX
  • Apps all stateless, easy to dockerize
  • Using mesos, zookeeper, marathon, chronos
  • Databases and other stuff outside that cloud
• Mesos slave launches docker containers
• Docker Security
  • chroot < Docker < KVM
  • Running untrusted Docket containers are a BAD IDEA
  • Don’t run apps as root inside container
  • Use a recent kernel
  • Run as little as possible in container
  • Single static app if possible
  • Run SELinux on the host
• Finding things
  • Lots of micoroservices, marathon/mesos moves things all over the place
  • Whole machines going up and down
  • Marathon comes with a tool that pushes it’s state into HAProxy, works fairly well, apps talk to localhost on each machines and haproxy forwards
  • Use custom script for this
• Collecting Logs
  • Not a good solution
  • can mount /dev/log but don’t restart syslog
  • Mesos collects stdout/stderror , hard to work with and no timestamps
  • Centralized logs
  • rsyslog log to 127.0.0.1 -> haproxy -> contral machine
  • Sometimes needs to queue/drop if things take a little while to start
  • rsyslog -> logstash
  • elasticsearch on mesos
  • nginx tasks running kibana
• Troubleshooting
  • Similar to service discover problem
  • Easier to get into a container than getting out
  • Find a container in marathon
  • Use docker exec to run a shell, doesn’t work so well on really thin containers
  • So debugging tolls can work from outside, pprof or jsonsole can connect to exposed port/pid of container

Clouds, Containers, and Orchestration Miniconf

 

Cloud Management and ManageIQ – John Mark Walker

• Who needs management – Needs something to tie it all together
• New Technology -> Adoption -> Proliferation -> chaos -> Control -> New Technology
• Many technologies follow this, flies under the radar, becomes a problem to control, management tools created, management tools follow the same pattern
• Large number of customers using hybrid cloud environment ( 70% )
• Huge potential complexity, lots of requirements, multiple vendors/systems to interact with
• ManageIQ
  • Many vendor managed open source products fail – open core, runt products
  • Better way – give more leeway to upstream developers
  • Article about taking it opensource on opensource.com. Took around a year from when decision was made
  • Lots of work to create a good open source project that will grow
  • Release named after Chess Grandmasters
  • Rails App

 

LXD: The Container-Based Hypervisor That Isn’t –  Tycho Andersen

• Part of Openstack
• Based on LXC , container based hypervisor
• Secure by default: user namespaces, cgroups, Apparmor, etc
• A EST API
• A daemon that doesn’t hypervisory things
• A framework for maintaining container based applications
• It Isn’t
  • No network configuration
  • No storage management – But storage aware
  • Not an application container tool
  • handwavy difference between it and docker, I’m sure it makes sense to some people. Something about running an init/systemd rather than the app directly.
• Features
  • Snapshoting – eg something that is slow to start, snapshot after just starts and deploy it in that state
  • Injection – add files into the container for app to work on.
  • Migration – designed to go fairly fast with low downtime
• Image
  • Public and private images
  • can be published
• Roadmap
  • MVP 0.1 released late January 2015
  • container management only

 

Rocket and the App Container Spec – Brandon Philips

• Single binary – rkt – runs everywhere, systemd not required
• rkt fetch – downloads and discovers images ( can run as non-root user )
• bash -> rkt -> application
• upstart -> rkt -> application
• rkt run coreos.com/etcd-v2.3.1
• multiple processes in container common. Multiple can be run from command line or specified in json file of spec.
• Steps in launch
  • stage 0 – downloads images, checks it
  • Stage 1 – Exec as root, setup namespaces and cgroups, run systemd container
  • Stage 2 – runs actual app in container. Things like policy to restart the app
  • rocket-gc garbage collects stuff , runs periodicly. no managmanent daemon
• App Container spec is work in progress
  • images, files, compressed, meta-data, dependencies on other images
  • runtime , restarts processes, run multiple processes, run extra procs under specified conditions
  • metadata server
  • Intended to be built with test suite to verify