Fork me on GitHub
#ldnclj
<
2015-07-31
>
mccraigmccraig12:07:12

awesome : elasticsearch, cassandra and kafka all have actively maintained mesos frameworks now simple_smile

tcrayford12:07:33

#rip sweet riak

tcrayford12:07:50

@mccraigmccraig: guessing that means zookeeper does as well right? kafka needs an zk…

mccraigmccraig12:07:41

@tcrayford: if you are running mesos, you already have a zk... i've used the same zk for kafka as for mesos

tcrayford12:07:15

seems real weird to me 😉

tcrayford12:07:29

(I don't use mesos, just sorta keep an eye on it from time to time)

tcrayford12:07:58

like "we're running everything on mesos" "except for one dependency of a thing we run on mesos"

tcrayford12:07:13

I mean, kafka doesn't do a lot of zk traffic, so it's maybe ok 😉

tcrayford12:07:41

just seems like an odd thing, esp if you model it as the ideal future: "AWS provides me mesos"

mccraigmccraig12:07:31

zk is a mesos dependency, so unless that changes there is always going to be one around... but if the mesos-zk was hidden from the deployed containers (by AWS or whatever) then you could run another zk inside mesos... i haven't tried this, but given zk's requirement for persistence, a framework might well be needed

tcrayford12:07:00

yeah, makes sense simple_smile

jamiei16:07:27

mccraigmccraig: What sort of scale is your mesos usage (if you don't mind me asking)

jamiei16:07:47

Been keeping an eye on, but have refrained from any prod use thus far

mccraigmccraig16:07:52

jamiei: we've got an AWS cluster with 3 t2.small masters and 3 m3.large slaves. we're using marathon and chronos for running most of the application instances, while storage is elasticsearch, which is running on the cluster but not managed by mesos, and postgresql via RDS

mccraigmccraig16:07:48

so a small cluster, but with enough components and services that it was worth using mesos+marathon

mccraigmccraig16:07:01

a while back we only had 2 slaves... adding a 3rd was as trivial as you might hope (just run the pallet converge to create the new instance, then let marathon do the rest)

mccraigmccraig16:07:41

shrinking a cluster is perhaps not so simple... not tried it yet

mccraigmccraig16:07:11

jamiei: if you want to play, i use this to deploy a cluster : https://github.com/trampoline/clustermap-mesos/

mccraigmccraig16:07:37

to upgrade i just migrate to a new cluster, which is much less scary than trying to do it in-place

mccraigmccraig21:07:32

@malcolmsparks: i'm using buddy JWS token auth with a yada app... finding that i need to set custom headers {"Access-Control-Allow-Headers" "Authorization"} on resources for CORS requests... is that expected ?