03:57:01 GMT <programo> How can a new programmer in C could contribute to Redis? How could one start contributing to Redis?
04:48:43 GMT <markuman> programo: go through the issues
08:16:22 GMT <colo-work> hi
08:17:28 GMT <colo-work> we're maxing out a single redis instance (being CPU-bound) on a single machine that hosts our entire workload. we're considering our options: sharding they keyspace across multiple redis instances (by doing the sharding in the clients), or sharding via redis cluster. what would you recommend?
08:58:56 GMT <antirez> colo-work: I'm for Redis Cluster, it's not perfect but far better than home made solutions or other alternatives IMHO
08:59:55 GMT <colo-work> antirez, I see, thanks :) afaict, Redis cluster needs explicit client support, right?
09:00:20 GMT <ahfeel> antirez: is there any plans to make redis cluster auto discover and auto balance etc?
09:00:28 GMT <ahfeel> colo-work: yes
09:11:06 GMT <ahfeel> antirez: by te way what was the design decision not having any redis instance forwarding the request to the right node and returning the answer instead of having to let the client do this ? Simplicity ? Number of waiting connections ?
09:11:08 GMT <antirez> colo-work: yep that's a problem if you use a language with poor support, best thing is to spend some day testing for your use case
09:11:47 GMT <antirez> ahfeel: auto discovery not sure what means, since the cluster pretty much auto discovers everything already. Auto balance, there is the "redis-trib rebalance" command introduced a few months ago, but there are plans to do more than that.
09:12:10 GMT <colo-work> antirez, that's what we're planning to do. thanks (to you and all contributors) for creating Redis, btw :) it truly is an awesome piece of software.
09:12:12 GMT <antirez> ahfeel: decision is that it's better to make the client a bit more complex and save an hop (latency) in the case of a latency sensitive service like Redis
09:12:24 GMT <antirez> colo-work: thanks for using it
09:13:25 GMT <ahfeel> antirez: well by auto discovery I mean having the system quite fully automatic like an elasticsearch would do, you just launch instances in cluster mode with a given cluster name, and they link to each other automatically
09:13:57 GMT <ahfeel> antirez: usecase would be adding / removing redis nodes on the fly
09:14:31 GMT <ahfeel> antirez: because from what I understood the current cluster is not really designed to change the cluster topology easily no ?
09:14:31 GMT <antirez> ahfeel: that's supported (but of course you need to move data to the node after adding it to the cluster). You can run a new node, link it to the cluster, and move some data.
09:14:45 GMT <antirez> ahfeel: no, the contrary exactly, it's everywhere in the doc
09:14:54 GMT <antirez> ahfeel: you may want to read the cluster tutorial
09:15:41 GMT <ahfeel> antirez: oh ok, sorry about that then :) I'll rtfm and come back later :P
09:17:50 GMT <ahfeel> antirez: also, just saw your tweets, what are main reasons for ppl moving away from docker w/ redis ?
09:19:24 GMT <Schwarzbaer_> Hi; one sanity check, please. I want to create a (rather cinematic, "Person of Interest" style) frontend to monitor activities in a distributed system. So I thought that every process should create data that's needed for visualization and write it into the values of a hashmap, and also the timestamps of updates into a second hashmap. Then, on the client, I'd have a Redis instance that's slaved to the master that everybody is writing to. Thus, if the
09:19:24 GMT <Schwarzbaer_> bandwidth gets too tight for live updates, I'll still always get the latest updates. Is that correct?
09:31:07 GMT <Schwarzbaer_> ...or is that simply not how slaving works?
09:39:13 GMT <antirez> ahfeel: no idea... never touched Docker with a stick so far :-) Not sure about the real merits.
09:39:42 GMT <*> antirez  is here all day but unfortunately can't reply to help questions too much, sorry, back to modules interface hacking. LAter
09:49:09 GMT <antirez> yoav_s: so Jemalloc's response was that actually we should see less fragmentation with the new version.
10:55:53 GMT <badboy_> heho
11:06:05 GMT <Inge-> re
12:05:09 GMT <forkfork> has anyone gotten anywhere on a redis module that supports blocking commands
12:07:27 GMT <Schwarzbaer_> "Blocking commands"?
12:10:36 GMT <forkfork> similar to the subscribe command
12:12:24 GMT <Inge-> ?
12:12:39 GMT <Inge-> there are blocking pop commands
12:24:16 GMT <forkfork> I'm referring to the new module system though
12:25:08 GMT <forkfork> so when the module cmdfunc is called, the function some something similar to brpop
12:25:24 GMT <forkfork> or to put it another way, how would one implement brpop using the new module system
12:46:41 GMT <antirez> forkfork: planned but working currently on supporting true module data types with RDB/AOF serialization and so forth.
12:46:54 GMT <antirez> forkfork: will arrive soon or later...
12:47:22 GMT <antirez> the problem with blocking operations in modules is, that it's kinda pointless if you can't also trap events
12:47:30 GMT <antirez> because you usually want to unblock when something interesting happens
12:47:41 GMT <antirez> so the work is more extensive than one may think at a first glance.
12:47:52 GMT <forkfork> right, understand
12:49:33 GMT <forkfork> the module I'm planning on doing is to add a transaction log style data type - meaning that a very large number of subscribers can do something like pub sub without a lot of redis memory growth
12:52:03 GMT <forkfork> subscribers keeping their position in a message queue rather than having redis maintain where each consumer is up to and duplicating messages per subscriber
12:52:52 GMT <forkfork> anyway, module system looking really good - I'll be very excited to see where you get to with it
13:18:57 GMT <ws2k3> is there anythng i can do to improve redis performence? i can only get to 40 k ops/sec(no pipelining)?
13:25:15 GMT <Inge-> just started the windows versoni of redis on my machine and redis-benchmark is giving me 127K get ops/sec
13:32:59 GMT <ws2k3> Inge- did you use pipelining?
13:35:48 GMT <Inge-> nope
13:35:56 GMT <Inge-> this is pure redis-benchmark
13:36:54 GMT <ws2k3> Inge- i tested with redis-bench aswel
13:46:42 GMT <Inge-> what kind of cpu?
13:47:28 GMT <ws2k3> Inge- E5410  @ 2.33GHz
13:48:37 GMT <Inge-> hm. mine is an i7-4790K and it is giving almost 2x the performance, adjusted for frquency
13:50:03 GMT <ws2k3> Inge- but i did benchmarkt from another machine
13:50:07 GMT <ws2k3> Inge- not on localhost
13:53:46 GMT <ws2k3> Inge- when i run it on localhost it maxes on 35 k
14:50:47 GMT <Inge-> single core?
14:51:28 GMT <Vide> hello, I was playing with a redis cluster and with redis-trib and I cannot make rebalance to work properly, it always ends with a error
14:52:12 GMT <Vide> and leaves the cluster in a bad state, so I have to run redis-trib fix a couple of times to get everything in order again
14:54:20 GMT <Vide> here it is what I'm doing http://pastebin.com/623m2ass
14:54:30 GMT <Vide> any idea?
15:23:55 GMT <Vide> mmh actually it fails even with a manual reshard :/
15:24:07 GMT <Vide> any idea on how to debug it?
16:02:55 GMT <programo> Is disque repository similar to amazon sqs?
17:55:20 GMT <codepython7771> is there a way to expose a redis interface to an android phone directly?
17:55:29 GMT <codepython7771> pub / sub is what i am looking for
17:56:22 GMT <minus> you can expose redis publically, of course
17:56:42 GMT <minus> whether it's a good idea or not is another question
17:57:35 GMT <codepython7771> minus: I have a local network
17:57:48 GMT <codepython7771> minus: so my phone needs to access redis on the ubuntu server
17:57:57 GMT <codepython7771> minus: how do i access it from android's java interface?
17:58:55 GMT <minus> edit the redis config to bind to all interfaces, allow it in the firewall, connect to it via the machine's ip
17:59:31 GMT <codepython7771> so use java's interface for the client ? Is it tested on android?
18:00:22 GMT <minus> no clue. i'd assume it to work if the app has internet permissions
18:00:42 GMT <codepython7771> hope someone here who uses the java client knows?
18:01:45 GMT <programo> Is disque repository similar to amazon sqs?
18:03:06 GMT <codepython7771> minus: most redis clients dont work on android
18:03:52 GMT <minus> they should work just fine if the app has the right permission
18:04:38 GMT <codepython7771> Someone reported none of them work on android - stack overflow
20:42:40 GMT <ws2k3> Inge- how you mean single core?
20:43:04 GMT <ws2k3> Inge- the machine has 4 cores. but as far as i know redis is single threated so it cannot handle more cores
20:45:48 GMT <Inge-> true. but if it had a single core,  that could explain slower perf when running redis-benchmark on the same machine
20:46:23 GMT <Inge-> otherwise I would expect better performance when running locally
21:21:07 GMT <antirez> forkfork: I think I may use your idea as an example module! To show new in module data types I'm implementing.
21:21:26 GMT <antirez> oops... she/he is no longer here.
21:21:44 GMT <antirez> anyway going to bed, good night
21:52:59 GMT <mustmodify> I work with a lot of relational databases.... want to do something with Redis. And I'm pretty sure I have the basic outline; just need to have someone verify that I'm on the right path. End goal is to be able to generate data like this: https://gist.github.com/mustmodify/d22cbb30d4766b738439af6f8d36b85d ... but we don't store (in this case) BMI. We store height and weight. And weight may be in kg or lbs. And sometimes we don't have that data. So there's l
21:53:29 GMT <mustmodify> Things I don't know: Do I want to update the document the next night? Or dump it and redo everything?
21:55:14 GMT <mustmodify> What data type do I want to use? Do I want to store the JSON? Or generate the json based on native Redis data types?
22:03:26 GMT <mustmodify> Looking at the documentation, one viable option is to use a list. So for BMI, I might have a list bmi_values with JSON string values. Pros: can grab the entire list quickly. Can add to the list easily. Cons: can't easily find a single value (which I would only want if I was going to update the list). So this would mean I would regenerate the list every night rather than updating visits for whom values have changed... unless I keep a separate list of list 
23:03:40 GMT <minus> mustmodify: another option is to use one key per data point and make it a hash. pro: you can easily access the data points individually and modify parts. con: overhead. and you need to know which keys you want to use when accessing. if you wanted to aggregate a bunch of them you could do that in lua in redis
23:04:04 GMT <minus> it all boils down to the access pattern, as usual