#coreos

/

      • calavera joined the channel
      • danderson
        hmm, is etcd 0.5.0-alpha4 supposed to be compatible with 0.4.x data? (at least in the upgrade direction)
      • I dropped etcd 0.5.0-alpha4 onto a stable coreos machine, and it complains: "unknown wal version in data dir /var/lib/etcd"
      • which suggests etcd 0.4.x can't be upgraded to 0.5.0... which doesn't sound right?
      • moonflux has quit
      • AvianFlu joined the channel
      • robszumski
        danderson: that will be fully supported, im not sure if it's fully working yet though
      • pelox has quit
      • danderson
        robszumski: ah, gotcha.
      • (fwiw, /var/lib/etcd on this machine contains only an empty snap/ directory - no data in the cluster)
      • but for toying around, I'll just nuke and pave. Thanks.
      • haploid joined the channel
      • robszumski
        danderson: some of those edge case scenarios might be polished up yet, i thought i had seen a working demo of upgrading very recently
      • danderson
        fwiw, the crash-loop when starting on an empty /var/lib/etcd (w/ snap directory removed): http://pastie.org/private/zhp8zbyrellazlag7xokta
      • argh, journalctl truncated it. Sec.
      • first time round, it fails to find itself. All following times, it complains about a bad wal version.
      • danderson digs through open bugs
      • aha, okay.
      • so, etcd 0.5 won't start up correctly on an empty /var/lib/etcd . You have to run etcd 0.4 at least once to set up the 0.4 state.
      • at that point, etcd 0.5 will happily come up and migrate that state.
      • but bringing up etcd 0.5 on a clean slate fails.
      • harish joined the channel
      • robszumski
        danderson: what are you seeing on the blank slate failure? it works fine for me when i was playing around
      • danderson
        robszumski: that would be the last pastie I linked above.
      • the first run fails with "etcd: couldn't find local name "1a105f4e7d354ecc9343e9f060e27569" in the initial cluster configuration"
      • jkyle1 joined the channel
      • robszumski
        are you not specifying a discovery url?
      • korya joined the channel
      • if not, the bootstrapping process changed a little
      • danderson
        correct, not specifying any config, just running the etcd binary
      • (I'm using the 0.4 etcd.service, with just s#/usr/bin/etcd#/opt/etcd#
      • so, okay, if that's not meant to work properly, mea culpa. I should go read moreabout 0.5 :)
      • robszumski
        danderson: if youre just running ./etcd it should work fine, but using a new binary with the old unit file probably wont work
      • danderson
        ack. Is there a sample .service in git somewhere?
      • I guess not, that'd probably be in another coreos repo.
      • korya has quit
      • bitshuffler joined the channel
      • justinsevens joined the channel
      • zz_rabisg is now known as rabisg
      • rodlogic joined the channel
      • doublerr has quit
      • justinsevens has quit
      • robszumski
      • haploid has quit
      • danderson
        robszumski: thanks!
      • robszumski
        thats a fleet unit FYI
      • pblittle joined the channel
      • cnbuff410 has quit
      • Tranquility has quit
      • cnbuff410 joined the channel
      • danderson
        yup. I'm just looking at the etcd flags, and the 0.5 docs in git
      • NOTICE: [coreos-overlay] marineam pushed 2 new commits to master: http://git.io/TbIV4Q
      • NOTICE: coreos-overlay/master 1966e6b Oliver Kurth: add patch to set USE_SLASH_PROC macro
      • NOTICE: coreos-overlay/master e8f43f2 Michael Marineau: Merge branch 'open-vm-tools-9.4.6-procps-fix' of git://github.com/vmware/coreos-overlay...
      • nyarla is now known as nyarla_away
      • NOTICE: [coreos-overlay] marineam comment on issue #965: Took care of the merge conflicts, so this is in now. :) http://git.io/NNT4ZA
      • AvianFlu has quit
      • AvianFlu joined the channel
      • tsandall joined the channel
      • cnbuff410 has quit
      • chance
        im curious for when etcd 0.5 comes out how one would configure using systemd a set of 3 cluster nodes, and then a bunch of proxies for all other nodes.
      • itxx00 joined the channel
      • i imagine you would need to leverage fleet, but atm i just have etcd setup via my cloud-init
      • robszumski
        chance: i dont know how it will end up working, but having a proxy attribute in cloud-init seems easy
      • chance
        an attribute?
      • inconshr_ has quit
      • inconshreveable joined the channel
      • tsandall has quit
      • pelox joined the channel
      • cnbuff410 joined the channel
      • itxx01 joined the channel
      • itxx00 has quit
      • yexingok joined the channel
      • calavera has quit
      • whitebook joined the channel
      • harish has quit
      • graeme_ joined the channel
      • alanwill joined the channel
      • yexingok1 joined the channel
      • em-dash has quit
      • robszumski
        yeah
      • so you'd have a cloud config for active peers and one for proxies
      • yexingok has quit
      • chance
        oh right, multiple cloud configs
      • yeah.
      • pblittle has quit
      • i keep forgetting that there's a way to do things less simply than in my simple 3 node vagrant cluster xD
      • i should deploy to some cloud provider a larger cluster to get a better feel for a real deployment.
      • fullstacker joined the channel
      • cnbuff410 has quit
      • cnbuff410 joined the channel
      • viklas joined the channel
      • fullstacker has quit
      • NOTICE: [coreos-overlay] marineam comment on issue #973: [![LGTM](http://www.lgtm.in/p/WXjnfn3iq)](http://www.lgtm.in/i/WXjnfn3iq) http://git.io/oDsM1g
      • fayablazer has quit
      • cnbuff410 has quit
      • gabe3 joined the channel
      • gko has quit
      • gko joined the channel
      • tomh- has quit
      • calavera joined the channel
      • tsandall joined the channel
      • bitshuffler has quit
      • xuxi joined the channel
      • duck_cpd has quit
      • tsandall has quit
      • xuxi
        is there an example tutorial somewhere showing how to connect a database, a backend/api, and a web frontend together
      • chance
        using coreos?
      • xuxi
        chance, yes
      • chance
        ive followed some tutorials of getting elasticsearch, logstash, and kibana hooked up on coreos
      • which is sort-of like that.
      • a little
      • xuxi
        chance: if you have links to those article, that would be helpful as well
      • kibana reads from a static json file, so it's not quite the same
      • chance
      • yeah, its not quite the same
      • xuxi
        if i recall correctly
      • chance
        but it's pretty adaptable.
      • and kibana does have to talk to elasticsearch
      • so it does have a backend, of sorts.
      • but your right, its a little more complex with a full stack web app
      • xuxi
        i'm also wondering if anyone here has considered the points mentioned here: http://consul.io/intro/vs/zookeeper.html
      • are there tools available for extending the heartbeat fucntionality of etcd?
      • graeme_ has quit
      • gabe3 has quit
      • nyarla_away is now known as nyarla
      • nyarla is now known as nyarla_away
      • robszumski
        xuxi: if you're looking for heartbeating + load balancing, vulcand has it built in and reads from etcd natively
      • chance
        i really should look into vulcand.
      • Sinjek has quit
      • nyarla_away is now known as nyarla
      • Sinjek joined the channel
      • johanneswuerbach has quit
      • cnbuff410 joined the channel
      • rsampaio_ joined the channel
      • Damm
        yeah but it's not https
      • danderson
        the other option for proxying in etcd 0.5, iiuc, is to just point all etcds at the same discovery url after setting up the active peers
      • the others would see that the cluster is full, and switch to proxy mode
      • Damm
        i don't know off hand how do to this in docker; but ucarp is fairly good
      • otherwise you would need to use something like anycast and have exabgp announce from a container
      • for a different kind of load balancing
      • danderson
        ooh, exabgp looks neat.
      • Damm
        and besides anycast/routing layer load balancing is a lot easier to scale
      • than trying to scale haproxy
      • i did that once; 25 proxy servers for 450 nodes
      • really insane
      • chance
        danderson: wait nodes can determine the cluster is full with proxy mode?
      • ahh interesting..so you can set the size in the discovery url
      • thats awesome.