April 14, 2017 
< | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | >
April 14, 2017 [04:44:35] *** shyu <shyu!~shyu@106.38.122.5> has joined #project-FiFo[05:28:02] *** t4ryn_ <t4ryn_!foobar@unaffiliated/t4ryn> has quit IRC (Ping timeout: 255 seconds)[05:31:57] *** t4ryn <t4ryn!foobar@unaffiliated/t4ryn> has joined #project-FiFo[05:40:16] *** t4ryn <t4ryn!foobar@unaffiliated/t4ryn> has quit IRC (Ping timeout: 258 seconds)[05:46:45] *** t4ryn <t4ryn!foobar@unaffiliated/t4ryn> has joined #project-FiFo[08:00:47] *** shyu <shyu!~shyu@106.38.122.5> has quit IRC (Max SendQ exceeded)[08:27:34] *** KermitTheFragger <KermitTheFragger!~KermitThe@gateway.serviceplanet.nl> has joined #project-FiFo[13:04:49] *** dexta <dexta!~D3XTA@host86-177-117-121.range86-177.btcentralplus.com> has quit IRC (Ping timeout: 260 seconds)[13:05:52] *** dexta <dexta!~D3XTA@host86-177-117-121.range86-177.btcentralplus.com> has joined #project-FiFo[17:32:23] *** KermitTheFragger <KermitTheFragger!~KermitThe@gateway.serviceplanet.nl> has quit IRC (Remote host closed the connection)[19:00:37] *** tomww <tomww!~tom@unaffiliated/tomww> has quit IRC (Ping timeout: 245 seconds)[20:55:28] *** bsmithx10 <bsmithx10!~root@pool-108-52-236-160.phlapa.fios.verizon.net> has joined #project-FiFo[20:56:11] <bsmithx10> Can I scale up the storage nodes in Dalmatiner?[20:58:20] <Schaeferhund> [Heinz (heinz)] yes[20:58:20] <bsmithx10> I was thinking of making a Production grade deployment using https://github.com/joyent/manatee to management PG[20:59:06] <bsmithx10> and add storage nodes when rewuired.... which wont be for a long time because you absolutely nailed the implimentation[20:59:07] <Licenser> I've not used mantee but there are a few good soultions out there for PG clustering I think[20:59:22] <Licenser> :)[20:59:24] <bsmithx10> Licenser: any recommendations?[20:59:41] <bsmithx10> Im a SDC user, but I've been following fifo for a long time[21:00:00] <bsmithx10> Any plans to get prometheus out of the way?[21:00:05] <Licenser> Not an expert on that to be honest, since it's 90% passive (data ingestion is independant of postgres) I just use a cold standby with replication[21:00:23] <bsmithx10> id like to be able to just run dalmatiner and still do service discovery etc[21:00:36] <bsmithx10> i saw you had the scraping built into the proxy already[21:01:27] <Licenser> if you want to use pormetheus you can use the promethos remote storage writer[21:01:30] <Licenser> the proxy supports that[21:01:44] <bsmithx10> I saw, but the proxy supports scraping[21:02:00] <Licenser> yea it does no disocvery so, need to manually configure the targets[21:02:15] <bsmithx10> But I guess your correct, you wont have to keep track if the writer api is stable[21:02:37] <bsmithx10> no point in putting SD into the proxy[21:02:41] <Licenser> yea sadly teh writer api isn't stable today but it didnt' chagne in a while[21:03:07] <bsmithx10> Also huge props to brancing out of illumos and supporting freebsd![21:03:22] <Licenser> I try to avoid implementhing things that are already tehre. The whole metric collecting part is rather mature compared to the storage side :)[21:03:26] <Licenser> thanks mate[21:05:05] <bsmithx10> Quick question, do you think the prometheus server running on ZFS could perform the same write perf?[21:05:12] <bsmithx10> or did you do some magic?[21:05:36] <Licenser> to start with prometheus does not really do persistant storage[21:06:49] <Licenser> then part of the perofmrance is data layout based to allow efficient compression / cashing for ZFS[21:07:27] <bsmithx10> gotcha, so there is some massaging to allow for zfs to shine[21:07:55] <Licenser> well the file layout is totally costom, each series is a non spares block of data in a file[21:08:17] <Licenser> if you want to do somethinf really fancy allign the points per file to the zfs page size so data dopesn't span pages[21:08:38] <Licenser> that allow a metric to be fully encoded in a page giving it even more +[21:08:50] <bsmithx10> cool, I havent tried syslog yet[21:09:00] <bsmithx10> but i imagine it works ?[21:09:03] <bsmithx10> any gotchas?[21:09:25] <Licenser> be aware there are more then 1 syslog protocl :P[21:09:39] <bsmithx10> hmmm[21:09:53] <Licenser> I think most use the BSD one, or the other one[21:10:01] <Licenser> I dind't know that when Is tarted impolemenitng it[21:10:18] <bsmithx10> https://www.joyent.com/blog/docker-log-drivers[21:10:19] <Licenser> that all said the whole syslog/event store thing isn't too well tested havn't found many poeple using it[21:10:26] <bsmithx10> was going to be for those docker-log-drivers[21:10:33] <Licenser> give it a shot ;)[21:11:05] <bsmithx10> yea, i plan to.... Ill give it a go and see how it goes[21:11:33] <bsmithx10> cant wait.... thanks![21:11:43] <Licenser> :D let me know how it goes I'm excited[21:11:44] <bsmithx10> Awesome work, great talks![21:11:53] <Licenser> FWIW tachyon should work with SDC too[21:12:07] <bsmithx10> really?[21:12:14] <Licenser> yap it's not dependant on fifo[21:12:21] <Licenser> should actually work with everything that can do kstat[21:12:32] <bsmithx10> awesome[21:12:40] <bsmithx10> Because, they wrote this thing called CMON[21:12:44] <Licenser> not all features so, there is some metadat lookup happening but it's just some dat a you'd not get[21:12:46] <bsmithx10> did you see it?[21:12:50] <Licenser> did not[21:13:08] <bsmithx10> https://github.com/joyent/triton-cmon[21:13:16] <bsmithx10> https://github.com/joyent/triton-cmon-agent[21:13:28] <Licenser> will give it. alook[21:14:01] <bsmithx10> https://github.com/joyent/rfd/tree/master/rfd/0027[21:14:09] <bsmithx10> the rfd explains the architecture a little better[21:14:25] <bsmithx10> maybe you and Rkeine could share metric collectors etc.[21:15:26] <bsmithx10> i was hoping to expand it.... right now it doesnt have many metrics.[21:15:27] <Licenser> a quick glance seems that big difference is push vs pull[21:15:44] <Licenser> tachyon just lets you get everything you can get from kstat ^.^[21:15:58] <bsmithx10> think his idea is to impliment that[21:16:05] <bsmithx10> Thought it was a pull model[21:16:26] <bsmithx10> it provides service dscovery throught eh CMON proxy[21:16:41] <bsmithx10> Im not 100% sure if the cmon-agents are pushing or being polled[21:17:12] <bsmithx10> or if the porxy just allows for Prom to run the scrape against the SD endpoint[21:17:22] <Licenser> ah[21:17:32] <bsmithx10> i didnt really dig into the internals, but ic onfigured it in my on prem[21:18:33] <Licenser> yea tachyon uses a own wire format, which is a lot more efficient then the prometheus one[21:18:51] <bsmithx10> is tachyon running the gz[21:19:08] <bsmithx10> which language, erlang?[21:19:23] <Licenser> the collector yes, that is C it's quite litterally kstat + curl[21:19:40] <Licenser> it use nsq for transport (go) which can run in the GZ or Z, and the aggregator is in erlang[21:20:48] <bsmithx10> hmm, because i would be totally open to running it on my nodes[21:20:58] <Licenser> ^.&[21:20:58] <bsmithx10> as long as i dont have to install anythng crazy in the gz[21:21:13] <Licenser> the reason to have the meter in C is that it has as little impact as possible[21:21:18] <bsmithx10> gotcha[21:21:59] <Licenser> you can just grab the fifo binary for it[21:22:02] <bsmithx10> strange that you guys dont work together more closley[21:22:17] <bsmithx10> whats the repo? I gotta read :)[21:22:38] <Licenser> https://github.com/project-fifo/tachyon-meter[21:22:47] <bsmithx10> thx[21:24:37] <Licenser> joyent has no love for erlang and here is no love for node.js ;)[21:28:20] <bsmithx10> ahhh, but your metric stuff is in C![21:28:24] <bsmithx10> and GO![21:29:22] <Licenser> yea C is from the illumos people and go is from the nsq people :P[21:29:28] <Licenser> probably joyent never bothered to look at it[21:34:47] <bsmithx10> ahhh I feel some tensions :P[21:37:07] <Licenser> nah