01:20:00  * ed209quit (Remote host closed the connection)
01:20:09  * ed209joined
05:04:08  * drakanjoined
05:14:26  <drakan>is anyone using manta outside of joyent ? i'm a little disappointed the concept hasn't been more widely adopted...
05:21:02  <melloc>drakan Samsung does, which is why they acquired Joyent.
05:21:27  <drakan>i like to think of that particular acquisition as joyent merging into samsung and taking it over. since tech wins, right ?
05:21:52  <melloc>And there are other Joyent customers with on-premises deployment, plus JPC customers.
05:22:07  <drakan>i'm on week 2 at $JOB dealing with petabyte scale streaming systems and i keep thinking to myself ... man, i wish i could do arbitrary compute over this data
05:22:30  <drakan>everything's so expensive and clunky because it's network bound...
05:24:30  <melloc>I don't know anything about what your stack looks like or the internal politics involved, but if you're currently moving data around a lot, could you justify a small several node Manta stand-up that you then transfer some data into when you need to do compute?
05:25:03  <drakan>well, i find myself trying to get a sense of what the latencies are like in production deployments of Manta
05:25:47  <drakan>but to flip it around -- why aren't startups built around a "Manta-first" kind of mindset? am i just too early?
05:26:11  <drakan>feels like we've been stuck in Kafka+Cassandra+Elasticsearch land for an unreasonably long time, heh
05:28:13  <melloc>The answer to the latency question is going to depend a lot on the hardware used, and on how you configure things, such as how many shards you use for the database.
05:28:55  <melloc>The good news is that we've spent a lot of effort on trying to make it easier to measure everything over the past two years, and understand where the bottle necks are n
05:30:14  <melloc>And once you know, you can scale out where needed (e.g., you can reshard if Postgres is the problem)
05:33:41  <drakan>i think i'm mainly concerned with the scaling for the compute side
05:33:57  <drakan>is it reasonable to run persistent and stateful services on manta's container layer ?
05:39:26  <melloc>I think that there might be several things that would get in the way of doing that today, but it's a good question for the mailing list. I haven't worked much with the jobs tier so my recall is fuzzy here, but I think that when you spawn a Manta job, you can specify an expected duration for the job.
05:39:52  <melloc>And if it runs longer than that, it gets killed? That might be wrong though.
05:41:52  <drakan>do you uhh.. deploy 108TB dense nodes... too ? https://eng.joyent.com/manufacturing/matrix.html looks both amazing and terrifying
05:42:53  <melloc>Running something stateful within the container during that duration works perfectly fine though. For example, mlogin sessions are basically a special kind of job.
05:45:48  <melloc>I think those are what we use in JPC right now. The systems we've been using for Samsung I believe are denser.
05:53:25  <melloc>I work more on the Triton side of things than Manta, so I'm unfortunately not particularly helpful here. The mailing list might be able to help more, especially about compute jobs. I can't find anything on compute job duration right now, just requesting memory/disk space.
05:53:47  <melloc>It's possible that your job could just run as long as you want it to.
05:54:24  <melloc>If the cn it's on crashed though, you might have to take care of restarting the job.
05:54:29  <drakan>as i'm primarily an operator, i *love* the idea of fixed duration jobs for stateful services... completely eliminates "hygiene restarts" and the like :)
05:56:24  <drakan>you've been super helpful, really appreciate the conversation. gonna think on it some more and try out a local deploy and see how it feels
06:05:34  <melloc>Glad to help! If you haven't see it already, this document should be helpful: http://joyent.github.io/manta/
06:07:44  <drakan>yeah, i have that in a tab here someplace ... i think my best baby step is standing up triton, lol. sigh.
06:07:59  <drakan>kinda surprised triton isn't in http://autopilotpattern.io/ yet
08:13:57  <_Tenchi_>Cistoo Mantl, Docker Swarm, Joyent Triton...
08:14:23  <_Tenchi_>Mesos on Joyent Triton
08:14:34  <_Tenchi_>triton is referenced in some capacity on that page
08:23:57  * neuroservejoined
10:03:13  * neuroservequit (Ping timeout: 248 seconds)
10:20:00  * ed209quit (Remote host closed the connection)
10:20:08  * ed209joined
12:32:47  * goberlejoined
12:36:56  * goberlequit (Remote host closed the connection)
12:37:18  * goberlejoined
13:58:31  * neuroservejoined
15:41:09  * neuroservequit (Ping timeout: 260 seconds)
18:17:12  * chrisBFjoined
18:45:47  * chrisBFquit (Remote host closed the connection)
20:20:00  * ed209quit (Remote host closed the connection)
20:20:08  * ed209joined