05:00 <Tulah268> ARE YOU LOOKING FOR A HELP CHANNEL WHERE THE HELP DON'T KNOW SHIT BUT INSTEAD OF SAYING THEY DON'T SHIT THEY WILL SPEND 10 MINS JERKING YOUR CHAIN?? LOOK NO FURTHER THAN #UBUNTU A CHANNEL FULL OF DUMB NIGGERS THAT DON'T KNOW SHIT!!

05:00 <Tulah268> ARE YOU LOOKING FOR A HELP CHANNEL WHERE THE HELP DON'T KNOW SHIT BUT INSTEAD OF SAYING THEY DON'T SHIT THEY WILL SPEND 10 MINS JERKING YOUR CHAIN?? LOOK NO FURTHER THAN #UBUNTU A CHANNEL FULL OF DUMB NIGGERS THAT DON'T KNOW SHIT!!

05:00 <Tulah268> a c a r r i c o S a r g u n m 6 w 6 a t u r l e y a l e x a s h k a 2 a t k R o b d o r j m i v e n i n a r a j o n b r w n _ k r i g [ m ] _ w h i t e l o g g e r j o n r h k e a t h l e y j t f m u m m g o r n i k m s t r m p n k d a r a c h o m a r k j s l f r i t c h i e L i c e n s e r c a t e r n A m u n _ R a S c h w a r z b a e r D D R d o m 9 6 B o m b e C c x W r k m i s t e r _ z

05:00 <Tulah268> acarrico Sargun m6w6 aturley alexashka2 atk Robdor jmiven inara jonbrwn_ krig[m] _whitelogger jonrh keathley jtfmumm gornikm strmpnk darach omarkj slfritchie Licenser catern Amun_Ra Schwarzbaer DDR dom96 Bombe CcxWrk mister_zombie tokenrove voldyman betawaffle emilbayes cquinn magicbit doublec saolsen dlowe kempe bougyman larshesel bbhoss j14159 wyvern ericbmerritt_ jeremyheiler kjsnavely bibi dch SeanTAllen Viknet krig jbk oldsk00l malthe Miikka

15:04 <hmans> Seeing as HTTPServer is being removed from the standard library (right?), what would be the recommended place to head to if I were interested in building a HTTP server with Pony?

15:05 <SeanTAllen> well, if you are interested in performance, i would suggest writing a high-performance http parser first.

15:05 <SeanTAllen> that is a project unto itself and then many different http servers can be built around that

15:06 <SeanTAllen> lots of good start performance notes here: https://www.ponylang.org/reference/pony-performance-cheatsheet/

15:06 <SeanTAllen> or you could find a c http parser that is "high performance" and start by wrapping that with a pony shim and then using that to build the http server you want

15:07 <hmans> I don't know if performance is terribly relevant to me at this point.

15:07 <SeanTAllen> do you like the API for the http server that exists?

15:09 <hmans> Taking https://github.com/ponylang/ponyc/blob/master/examples/httpserver/httpserver.pony as a reference example of what an actual program consuming the API would look like, I would say it feels sane enough, yes. I haven't written anything against it so far.

15:10 <SeanTAllen> well then, i would say, it the performance and API for that are to your liking, fork the http server and start a project to do what you want with it and try to get other folks involved as well

15:10 <SeanTAllen> maybe roll in theodus' jennet if you want to provide a more "sinatra" like api

15:11 <SeanTAllen> https://github.com/theodus/jennet

15:11 <hmans> *bookmarks*

15:12 <hmans> I'm happy jennet's there, but I'm personally not very keen on Sinatralikes. Part of my ongoing pet project (the one I'm currently using Crystal for) is a small routing-tree-style framework that I like very much, so I'm generally interested in finding out how it would work in the context of Pony actors.

15:12 <hmans> But note taken

15:14 <SeanTAllen> well, that works as well

15:15 <SeanTAllen> i think in general things like jennet or what you describe that are at a higher level are probably better for getting people involved.

15:15 <hmans> Yeah, I will definitely check it out, at the very least for learning from its code.

15:16 <_rck> SeanTAllen: is there a way to check if an actor is getting overloaded?

15:16 <hmans> Right now I'm mostly trying to get a feel for what are the current Pony projects within the domain of web servers/apps/frameworks

15:17 <_rck> TCPConnectionNotify has a throttled/unthrottled to let it know that is experiencing backpressure, but I'm curious to know if there's something general for any actor

15:20 <SeanTAllen> _rck nothing general has been done yet. throttled/unthrottled are related to events external to the runtime

15:20 <SeanTAllen> its your choice of whether to apply/release runtime backpressure

15:20 <SeanTAllen> there's nothing that is surfaced to the programmer at this time about being muted/unmuted

15:20 <SeanTAllen> given that would be a message send,

15:20 <SeanTAllen> you wouldnt be told you were muted til after you were running again

15:21 <SeanTAllen> hmans: basically there are none. its a pretty greenfield for you to have fun in.

15:24 <_rck> ok, I guess one way is to consume messages as fast as possible to an internal queue and check for overload there

15:25 <SeanTAllen> well, not really.

15:25 <hmans> SeanTAllen: \o/

15:25 <SeanTAllen> _rck: what is your worry/concern?

15:30 <_rck> I was experimenting with the partitioned global key-value store we discussed the other day, but over time my latencies were increasing linearly

15:30 <_rck> I'm assuming the partition actors were getting overloaded, but I saw no way to check

15:39 <_rck> I might go with the persistent map impl that vaninwagen suggested, but I wanted to experiment with other solutions first

15:40 <SeanTAllen> if an actor gets overloaded, then the senders will be paused.

15:40 <SeanTAllen> i cant comment on latencies as i dont know how they are being measured.

15:41 <SeanTAllen> if you are getting backed up writing over tcp, you do need to apply/release backpressure yourself to account for that.

15:41 <_rck> this was round-trip latency from a client performing reading over tcp until they got a value back

15:42 <SeanTAllen> so quick question: how do you know the latency is coming from your pony app?

15:42 <SeanTAllen> can you do some per request timing inside the pony app to verify that you see latencies increasing over time?

15:43 <SeanTAllen> you could steal the metrics histograms from wallaroo perhaps...

15:43 <_rck> that's interesting, any pointers?

15:46 <_rck> oh, I see you wrote some posts on the blog about how you do it

15:47 <_rck> I'll take a look there!

15:47 <SeanTAllen> i can point you to the specific bit of code

15:48 <_rck> but to answer the question about latency, the clients are not really doing anything with the data, this was showing up during simple load testing

15:48 <SeanTAllen> here's the histogram class: https://github.com/WallarooLabs/wallaroo/tree/master/lib/wallaroo/core/metrics

15:49 <SeanTAllen> you'd need a way to record incoming and outgoing timestamp then could add the simple histogram.

15:49 <SeanTAllen> i understand the client isnt really doing anything but

15:49 <SeanTAllen> depending on how you are timing, then unrelated issues in a client could be being measured

15:49 <SeanTAllen> for example gc pauses in the client

15:49 <SeanTAllen> memory pressure in the client

15:50 <SeanTAllen> it could be getting scheduled oddly on the cpus

15:50 <SeanTAllen> (cpu pinning is a good idea there, set aside cpus only for client and only for server)

15:50 <_rck> ah, yeah, I see

15:51 <SeanTAllen> benchmarking is... fun?

15:51 <SeanTAllen> so, your test

15:51 <_rck> haha, yes it is

15:51 <SeanTAllen> are you adding new keys to k/v or just storing?

15:51 <SeanTAllen> errr

15:51 <SeanTAllen> just reading

15:51 <_rck> just reading, database prepopulated with random data

15:53 <SeanTAllen> so the data set is consistent across the run yes?

15:53 <SeanTAllen> and your access to it should be consistent (as in, your arent suddenly only reading a couple keys where earlier you were reading many or vice versa or anyhting like that)?

15:56 <_rck> yeah, the key distribution is pretty uniform

15:57 <SeanTAllen> are you monitoring memory usage of the server process? could there be a memory leak? from the sounds of it, the memory should grow for a but then settle into something stable.

15:57 <_rck> not right now, no

15:57 <_rck> right, that's something to keep in mind

15:58 <_rck> when I'm doing erlang, observer is a very valuable tool

15:58 <SeanTAllen> are you using master or the last release?

15:58 <SeanTAllen> ie master or 0.21.3 ?

15:59 <_rck> last release, yes

15:59 <_rck> is master stable enough for day to day usage?

16:04 <_rck> thanks for taking the time to go through this, btw

16:11 <SeanTAllen> well, there's a "memory" leak in 0.21.3 that might impact you.

16:11 <SeanTAllen> there's a number of breaking changes on master that may or may not impact on what you are doing

16:11 <SeanTAllen> but if you see memory growth, it might be related to the change that was reverted on master

16:12 <SeanTAllen> and will be released this weekedn

16:12 <_rck> ah, that's great to hear

16:13 <_rck> I guess I'll be lazy and wait to check if the new release fixes the issue then haha

16:14 <_rck> thanks for the effort in putting a new release together

16:29 <alexashka2> quick question: is there a map literal, like the one for array? Something of the form of {a: b}?

16:33 <SeanTAllen> alexashka2: no

16:33 <alexashka2> perhaps the obvious follow-up: why not?

16:35 <alexashka2> also, why is inserting into a map, a failable operation?

16:36 <SeanTAllen> the answer to the first would be "because there isnt"

16:36 <SeanTAllen> you'll have to be more specific re the 2nd

16:37 <alexashka2> fun ref insert(key: K, value: V): V ? => - this is a method in map.pony in collections. Why is there a question mark for an operation that can't fail as far as I can tell?

16:39 <SeanTAllen> because of

16:39 <SeanTAllen> https://www.irccloud.com/pastebin/JmOqrVUz/

16:40 <SeanTAllen> that comment should be removed as its misleading

16:41 <SeanTAllen> if any error is encountered, it will throw an error

16:41 <SeanTAllen> that error would have to be handled.

16:41 <alexashka2> I understand, I am asking how an insert operation can ever fail

16:41 <SeanTAllen> there's no way at this time for the compiler to prove there will never be an error there

16:41 <SeanTAllen> you arent

16:41 <SeanTAllen> you are asking with its partial

16:41 <SeanTAllen> those arent the same question

16:42 <SeanTAllen> im not trying to be pedantic for the sake of it

16:42 <SeanTAllen> its just an important distinction here

16:42 <alexashka2> SeanTAllen: as far as i can tell, the implementation is partial, whereas it could easily be total

16:42 <SeanTAllen> how?

16:43 <alexashka2> SeanTAllen: the same way it is in Idris as one example?

16:43 <alexashka2> a langauge that makes far stronger totality guarantees than pony as far as I can tell?

16:44 <SeanTAllen> right

16:44 <SeanTAllen> ok

16:44 <SeanTAllen> im going to call that trolling

16:44 <SeanTAllen> "easily"

16:45 <SeanTAllen> just add dependent typing

16:45 <alexashka2> SeanTAllen: well you're asking me how one would implement a core data structure in pony, a language I have no familiarity with. That's a rather strange ask

16:46 <SeanTAllen> you told me it could easily be total

16:46 <SeanTAllen> i asked you how

16:46 <SeanTAllen> " the implementation is partial, whereas it could easily be total"

16:46 <alexashka2> let's remove the easily part :)

16:46 <SeanTAllen> sure anything can be total in any language if you add enough features

16:46 <SeanTAllen> im going to step away from this conversation now

19:53 <vaninwagen> alexashka2: from the looks of it HashMap.insert could be total by just returning the given value in the else clause of the try, to make the compiler happy. As this case should never happen, but the compiler cannot know about it, we need to trick him a bit here.

19:53 <SeanTAllen> vaninwagen: that would be a bad idea

19:54 <SeanTAllen> it "cant happen" right now

19:54 <SeanTAllen> that could end up being a legit error at some point

19:54 <SeanTAllen> there's nothing to do for that until eventually the value dependent types come along

20:01 <vaninwagen> SeanTAllen: couldn't we ensure insert does insert in a unit-test, which fails if there is a swallowed error and thus the value is not properly inserted?

20:03 <SeanTAllen> sorry, i dont understand vaninwagen

20:03 <SeanTAllen> how does that get rid of insert being partial?

20:05 <vaninwagen> SeanTAllen: let me come up with a playground...

20:06 <vaninwagen> ah no

20:06 <vaninwagen> it is not possible

20:06 <vaninwagen> my bad, value is consumed

20:06 <vaninwagen> it needs to be partial, you are right

20:07 <vaninwagen> you're the real pro-Sean

20:07 <SeanTAllen> i look forward to the day we get value dependent types

20:07 <vaninwagen> me too

20:07 <SeanTAllen> see how hopeful i sound?

20:07 <SeanTAllen> one day, one day luke will open that pr

20:11 <vaninwagen> we need to send someone to get him off this irish island https://img.rasset.ie/000e1a91-800.jpg

20:13 <vaninwagen> _rck you have a partitioned key value store in pony?

22:37 <SeanTAllen> Pony-stable release coming tonight.