|
JewKiller 3000 posted:javascript and go are alike in that they both suck
|
# ? Oct 8, 2016 06:18 |
|
|
# ? May 4, 2024 13:35 |
|
uncurable mlady posted:we're active contributors to the dashboard, I can talk to some people and see where that's roadmapped at if it isn't out there already I decide to skim through the kubernetes GitHub; some comments on multizone HA but looks like the focus at the moment is making it not so lovely to set up HA in the first place. will be interesting as more ubernetes related stuff lands in HEAD.
|
# ? Oct 8, 2016 06:34 |
|
the talent deficit posted:it takes like five minutes to setup efs with ecs efs is loving epic you dumb shithead
|
# ? Oct 8, 2016 09:45 |
|
30 cents per GB for a multi-az replicated netapp is 'bad' lol stfu
|
# ? Oct 8, 2016 09:46 |
|
yeah i like it a lot, though my bbs app doesnt like it because it's :vomit: sqlite3 :vomit: and that stuff shits the bed hard over nfs anything i wonder if i can just port the bbs to use an aws product for the DB, that would be a good experiment. what would i be looking at, dynamo probably? im not so good with databases.
|
# ? Oct 8, 2016 09:50 |
|
RDS mysql
|
# ? Oct 8, 2016 09:56 |
|
pram posted:efs is loving epic you dumb shithead Welcome
|
# ? Oct 8, 2016 10:11 |
|
Jonny 290 posted:yeah i like it a lot, though my bbs app doesnt like it because it's :vomit: sqlite3 :vomit: and that stuff shits the bed hard over nfs anything If it's sqlite can you not use RDS ? Has mysql and postgres options pram posted:efs is loving epic you dumb shithead Hey pram is back
|
# ? Oct 8, 2016 15:15 |
|
has anyone said "Docker? I barely knew 'er!" yet?
|
# ? Oct 8, 2016 15:59 |
|
dynamo is cool if you don't need transactions
|
# ? Oct 8, 2016 16:01 |
|
so i've never used docker. i gather that the idea is that you don't have to install a particular version of a language / database / whatever on your local machine, and you can run your programs against a known and standard environment. and for databases and stuff like that, you would include a set of standard data either in the docker container or outside it that loads each time you start it up. then you write your code outside and run it in the docker environment. is this close?
|
# ? Oct 8, 2016 16:06 |
|
JewKiller 3000 posted:javascript and go are alike in that they both suck
|
# ? Oct 8, 2016 17:43 |
|
Somebody post prams old av
|
# ? Oct 8, 2016 17:45 |
|
The Leck posted:so i've never used docker. i gather that the idea is that you don't have to install a particular version of a language / database / whatever on your local machine, and you can run your programs against a known and standard environment. and for databases and stuff like that, you would include a set of standard data either in the docker container or outside it that loads each time you start it up. then you write your code outside and run it in the docker environment. is this close? basically yeah
|
# ? Oct 8, 2016 18:03 |
|
pram posted:30 cents per GB for a multi-az replicated netapp is 'bad' lol stfu the iops are really bad and when you get throttled you'll be paying 10x s3 prices for s3 performance. it's ok if you need a durable store for infrequently updated/accessed data but it's not great if you're trying to use it instead of hdfs or similar
|
# ? Oct 8, 2016 18:52 |
|
actually go is pretty great
|
# ? Oct 8, 2016 18:55 |
|
the talent deficit posted:the iops are really bad and when you get throttled you'll be paying 10x s3 prices for s3 performance. it's ok if you need a durable store for infrequently updated/accessed data but it's not great if you're trying to use it instead of hdfs or similar no youre retarded but let me clarify what this guy is saying for the uninitiated. instead of paying $150 for 500gb storage in a managed replicated share i should use hdfs instead. because 'iops are really bad' so i should of course use an instance (the cheapest) with 10g networking and st1 ebs drives. and i need 3 for resiliency in three different vpc subnets. thats only like $4120 a month, cheaper if i commit to paying almost $40k for a year youre a great consultant. should i buy hortonworks support too. go gently caress yourself buddy!
|
# ? Oct 8, 2016 19:27 |
|
MALE SHOEGAZE posted:actually go is pretty great but learn you some erlang and *random obscure ml derivative*
|
# ? Oct 8, 2016 19:29 |
|
|
# ? Oct 8, 2016 19:38 |
|
heh, actually, ECS is 'fine' if you need plebian performance for garbage apps. but its going to be bad. and also docker containers are too heavy. and it and kubernetes arent written in a good language. IMO the optimal deployment these days is a hand-rolled mesos framework across your 40g infiniband networked cluster nodes
|
# ? Oct 8, 2016 19:49 |
|
jre posted:Hey pram is back
|
# ? Oct 8, 2016 19:49 |
|
pram posted:heh, actually, ECS is 'fine' if you need plebian performance for garbage apps. but its going to be bad. and also docker containers are too heavy. and it and kubernetes arent written in a good language. IMO the optimal deployment these days is a hand-rolled mesos framework across your 40g infiniband networked cluster nodes this is how i roll
|
# ? Oct 8, 2016 20:17 |
|
go is a fine language
|
# ? Oct 8, 2016 20:21 |
|
uncurable mlady posted:go is a fine language For clowns at the devops circus Actually it's ok
|
# ? Oct 8, 2016 20:25 |
|
pram posted:no youre retarded but let me clarify what this guy is saying for the uninitiated. instead of paying $150 for 500gb storage in a managed replicated share i should use hdfs instead. because 'iops are really bad' so i should of course use an instance (the cheapest) with 10g networking and st1 ebs drives. and i need 3 for resiliency in three different vpc subnets. thats only like $4120 a month, cheaper if i commit to paying almost $40k for a year or you could pay $15 a month for s3 and get basically equivalent performance to efs
|
# ? Oct 8, 2016 22:44 |
|
the talent deficit posted:or you could pay $15 a month for s3 and get basically equivalent performance to efs no
|
# ? Oct 8, 2016 22:54 |
|
even if you used a fuse s3 mount it wouldnt even be close to similar performance. the hdfs s3/s3a connector is slow as poo poo too. you literally dont know what youre talking about
|
# ? Oct 8, 2016 22:56 |
|
the talent deficit posted:or you could pay $15 a month for s3 and get basically equivalent performance to efs * s3 is eventually consistent after updates * you can never modify part of or append to a file; all modifications are rewrites * listing directories can be really awful, because directories don't actually exist not all file systems are created equal
|
# ? Oct 8, 2016 23:04 |
|
and even if you were using hdfs w/ s3 youd still need to pay for a namenode server which makes it far more expensive than $15
|
# ? Oct 8, 2016 23:08 |
|
unironic chat: this thread is cool and good
|
# ? Oct 8, 2016 23:43 |
|
what is mesos?
|
# ? Oct 8, 2016 23:49 |
|
the website is the usual incomprehensible apache word salad
|
# ? Oct 8, 2016 23:50 |
|
build a docker machine that does a little job, maybe processes some batch items then you have mesos jobs that control multiple instances of those go into the mesos interface, change instances from 4 to 8 and it spins four up and they start eating work change it for 8 to 4 and four of them shut down You need a bit of state and tracking to keep your data straight, make sure you don't leave stuff hanging, but it's a neat frontend
|
# ? Oct 8, 2016 23:53 |
|
mesos is a distributed scheduler. really you'd likely be using marathon, which is a mesos framework for running docker containers dc/os is a mesos distribution that has the promise of being an 'app store' for distributed deployments of cassandra/kafka/spark etc. thats what 'mesophere' makes. in general its all neat but managing container config through environment variables feels like an anti-pattern and no sir i dont like it
|
# ? Oct 9, 2016 01:51 |
|
Smythe posted:Somebody post prams old av
|
# ? Oct 9, 2016 01:52 |
|
pram posted:mesos is a distributed scheduler. really you'd likely be using marathon, which is a mesos framework for running docker containers I would disagree with calling mesos a scheduler; it's really more of a resource broker. frameworks request resources from mesos and it provides offers back. if an offer is accepted, then mesos will start the frameworks bit of code (executor) on that resource slice. this is what makes mesos so much more flexible compared to container orchestrators. my favorite bit about mesos architecturally is that all state in the masters is derived from the slaves; this makes it much simpler to setup compared to systems that store state in a centralized data stores. the downside of this, though, is that you can only have a single active master.
|
# ? Oct 9, 2016 02:41 |
|
fwiw there seems to be more people using mesos (really, marathon like you said) in 'production' but i wonder how much of that is simply because it was more mature than k8s at the time. lot of big money going into k8s ecosystem stuff now though, so i think the gap is gonna close soon. kubeadm in 1.4 really helps streamline the init process for a cluster as well so you've got options at the commercial and OSS level
|
# ? Oct 9, 2016 04:56 |
|
i like to say mesos is the more general YARN though i kind of like running spark on yarn more than mesos but that's probably mostly due to the fact im using cloudera on most of my clusters already so it's one less piece of software to janitor
|
# ? Oct 9, 2016 06:04 |
|
docker is ok docker-compose up is great But it seems like fixing docker-swarm/kubernites/storm/etc errors is really bad and our prod system goes down all the time and no one can fix them because can't find master or partition error or smth is this the true computer janitoring world?
|
# ? Oct 9, 2016 22:00 |
|
|
# ? May 4, 2024 13:35 |
|
MeruFM posted:docker is ok Busywork expands to fill a vacuum. They weren't just going to let us start programagically deploying stuff and kick back.
|
# ? Oct 10, 2016 00:14 |