Fork me on GitHub
#onyx
<
2016-03-15
>
greywolve08:03:33

lucasbradstreet: you have a PR for onyx-metrics simple_smile

lucasbradstreet09:03:35

Thank you @greywolve. I'll merge and cut a new release asap

lucasbradstreet11:03:00

On fire šŸ˜„

lucasbradstreet11:03:34

Sorry man, can you branch off master and create the PR against master?

lucasbradstreet11:03:36

Maybe I can just cherry pick the commits

lucasbradstreet11:03:44

The versioned branch 0.8.x releases are really supposed to be just for cutting releases. GitHub makes it really hard to make this the default viewing branch, but then PR off master

greywolve11:03:11

is this the case in general for all onyx repos? should i generally branch off master?

lucasbradstreet11:03:25

I havenā€™t found a good solution for this yet šŸ˜•

greywolve11:03:10

so create a new branch off master in my fork, apply changes there, then PR back to onyx master?

lucasbradstreet11:03:56

Yup, I would probably just branch off master in your fork, cherry-pick the couple of commits from your PR branch, then PR back to master

lucasbradstreet11:03:58

Just as an initial comment, it looks good. I'm going going to drop the mapvs/and into [] and just use regular maps with doalls, since it's more performant if you're not using it as a vector. It's probably a drop in the bucket but it's our default approach

greywolve11:03:18

go for it, could also just use a map instead of a mapv, and that would be the same thing

greywolve11:03:35

into in that context is a transducer, and will eagerly consume the collection

greywolve11:03:56

"In my opinion, the presence of doall, dorun, or even ā€œunchunkā€ is almost always a sign that something never should have been a lazy sequence in the first place."

greywolve11:03:01

worth considering, i guess simple_smile

greywolve11:03:01

but i guess consistency is good , if that's your default

gardnervickers13:03:24

@lucasbradstreet and @greywolve is it ok if I cherry pick that onto my taskbundles branch? Thereā€™s going to be a bit of conflicts that need sorting.

lucasbradstreet13:03:13

Please do. I was going to add a task bundle / new style test for it but if you want to handle it that'd be great

gardnervickers13:03:18

Schweeet šŸ˜„

lucasbradstreet16:03:21

Jepsen post is up! Can anyone please upvote it here if youā€™re around?

nha16:03:43

"We were in, however, need" the formulations feels a bit weird, no ? Shouldn't it be something like "We, however, were in need" ?

richiardiandrea16:03:57

@lucasbradstreet: shared it through LambdaX! Very nice one indeed!

lucasbradstreet16:03:30

@richiardiandrea: Thank you, I saw it and thought that was super quick!

lucasbradstreet16:03:40

@nha: good point. I might make that edit

lucasbradstreet16:03:24

Itā€™s on the second page of newest if anyone comes along later

lucasbradstreet16:03:35

Supposedly if you upvote from the link it wonā€™t be worth as much

lucasbradstreet16:03:44

Maybe just a myth tho

nha16:03:48

Oh I did not know that

ymilky17:03:36

For those here dealing with Kafka and not in the onyx gitter channel, if you want to try a new kafka client and be a guinea pig for the alpha version, I've written a 0.9+ set of libraries for Kafka consumer, producer, admin, embedded server, serializers, partitioners, callbacks, etc. and growing. It needs some hands-on testing before an official version is posted, but it's working in my projects and codebases, no issues. If you have any feedback, problems, etc., I'd be happy to chat with you. Hope to make an Onyx plug-in using it when it is better tested. https://github.com/ymilky/franzy

lvh17:03:13

Hey simple_smile Iā€™m trying to understand how messaging works. reading https://onyx.readthedocs.org/en/latest/user-guide/messaging/

lvh17:03:49

Iā€™m wondering how the connection scaling (m-squared vs n-squared) works? I thought Aeron used UDP in prod, so I was expecting that to be connectionless

michaeldrogalis17:03:25

Oh whoops, I need to delete our readthedocs account. We moved official docs to the website: http://www.onyxplatform.org/docs/user-guide/latest/messaging.html

michaeldrogalis17:03:47

I dont think there's any updates on the particular page, just FYI

lucasbradstreet17:03:11

Hi @lvh, Aeron is essentially building connections analogous to TCP on top of UDP, with the Aeron media driver being sort of like a userland TCP stack. The guarantees and performance will differ, but you get the point. The idea is that we only create one Aeron connection between two Onyx nodes, which is shared between v-peers, even though there exists m v-peers on the node.

lvh17:03:07

lucasbradstreet: gotcha; thanks simple_smile

michaeldrogalis17:03:48

@lvh If you're interested in Aeron in general, I'd recommend their Gitter. The devs who build it are really amazing folks.

lucasbradstreet17:03:06

Their talks are also great

lvh17:03:30

cool, will do simple_smile Aeron does seem really interesting

ymilky17:03:48

I can second Aeron, I've liked what I have seen so far.

lucasbradstreet18:03:00

It now shows the number of peers on each hosts, and itā€™ll show you which peers are allocated to jobs

lucasbradstreet18:03:21

Plus the spiffy visualisation with task maps showing on hover

otfrom18:03:47

michaeldrogalis: and I know that at least Richard Warburton is a good hugger. šŸ˜„

otfrom18:03:06

I've not hugged Martin Thompson yet. He's on my list though. I only really met him at a job interview (I was the candidate) and then very quickly at a conference.

lucasbradstreet18:03:14

@greywolve: back to your post about doall. He's totally right that it's usually an antipattern. We've found that map + doall is faster than mapv so it's mostly what we do

lucasbradstreet18:03:41

If you don't return the transact results in write batch we'd use run! instead

lucasbradstreet19:03:22

It's nice to have the results in the event map for availability via lifecycles tho

lucasbradstreet19:03:34

I think it's really mostly a problem when you have the doall far away from the map

lucasbradstreet19:03:59

Anyway, this will be totally dominated by the transact so it's mostly academic

lucasbradstreet19:03:53

No longer true. It was depending on a dated snapshot that got auto upgraded

greywolve19:03:21

lucasbradstreet: makes sense simple_smile