09:26:16 GMT <timothy> halo, sometimes I see keys called something:somethingelse, is it just a convention?
09:28:09 GMT <minus> yes
10:48:45 GMT <ahfeel_> Hello :) I have a question related to the Redis docker image
10:49:23 GMT <ahfeel_> It doesn't tweak any sysctl core.net values by default.. I'm not 100% sure either this should be tweaked on the host, on the container or on both.. Anybody has experience about this ?
10:58:04 GMT <_ikke_> ahfeel_: a dockter container does not run it's own kernel
10:58:15 GMT <_ikke_> it's just an isolated process
10:58:35 GMT <_ikke_> So you need to tweak it on the host kernel
11:03:22 GMT <ahfeel_> _ikke_: well, it's already tweaked on the host kernel, but when checking the sysctl values in the container, they are still the low ones
11:03:51 GMT <ahfeel_> and I've seen a few discussions about passing somaxconn settings to containers etc, for which an option has been added to the docker cli
11:03:57 GMT <ahfeel_> so I'm a bit doubtful about that
11:19:48 GMT <ahfeel_> _ikke_: any idea ?
11:23:07 GMT <timothy> ahfeel: most of the core.net values are not in cgroups so you can't tune them in a single cointainer
11:23:28 GMT <timothy> but if they are you can tune them inside the cointainer
11:55:12 GMT <kobzar> hello
11:55:20 GMT <kobzar> 2all
12:11:23 GMT <kobzar> i have one interest question. For example i have one database with 10 millions rows (string). If i will use EXPIRED for all this keys...does server will work without crazy stuck?
12:51:20 GMT <Arkentias> Hello guys! I've been facebashing this issue for days, I've entered every possible combination of words related to this issue on google but haven't found an answer ( probably because I lack the knowledge to really understand what could be the cause )
12:51:23 GMT <Arkentias> So here goes
12:53:12 GMT <Arkentias> I have a Kubernetes setup with 3 redis sentinels, when I contact them from any pod inside my kubernetes with redis-cli I have no trouble at all communicating, but from an ExpressJs application using IO redis, it always tells me that all sentinels are unreachable..
12:53:26 GMT <Arkentias> oh god it got splinched
12:54:04 GMT <Arkentias> First part was : I have Kubernetes running with 3 redis-sentinel pods, and I can access them from any pod in my kubernetes with redis-cli
12:57:50 GMT <ebalsumgo> Arkentias: have you exposed a service?
12:58:13 GMT <Arkentias> Yes I have, that's how I access it via redis-cli without problem
12:58:25 GMT <Arkentias> I just put in the name of the service
13:29:10 GMT <dupondje> Hi. I got some small question about redis replication. What if I set a slave to read-write, and I write data to it. Will the data get overwritten somehow by the master? Or will it stay out of sync? The main thing I need it is to have a redis rw when the master is down.
13:42:15 GMT <minus> dupondje: if the master writes to the same key it will overwrite it, otherwise nothing will happen to it
14:02:00 GMT <Arkentias> Is there a way to have like 2 redis servers running in parallel all the time with the same data?
14:03:35 GMT <Arkentias> Without sentinel ideally
14:11:18 GMT <dupondje> minus: ok! Would be nice if it could reconnect to the master for example, that it did a resync
14:11:21 GMT <dupondje> but ok
14:11:24 GMT <dupondje> :)
14:13:23 GMT <minus> writing to a slave will never write something to the master, dupondje. it's master-slave replication, not master-master after all
14:17:18 GMT <dupondje> minus: indeed. But the thing I was thinking about: master-slave (slave with slave-read-only no). And just use the slave when master is down. If master comes back online. It would be nice if the slave did a resync of the master again. So master & slave contain the same data again.
14:19:08 GMT <minus> the slave does pull new data from the master of course, but it won't wipe anything. if you have different kinds of access/data you could just run multiple redis instances
14:20:29 GMT <dupondje> well just need r-w slave when master is down :)
17:50:42 GMT <ahfeel> I'm considering using Sentinel to have redis HA, but I'm a bit concerned by latency. Our app is in PHP, and I wonder if reaching to one of the sentinel node to get the master address and then connect to the master on every single request isn't a big overhead !?
17:50:51 GMT <ahfeel> Anybody has a better solution / best practice for this ?
17:51:37 GMT <ahfeel> I might cache the master address locally in a file or whatever for a few seconds also to avoid this, but that sounds kinda creepy
17:52:03 GMT <minus> do you have a performance problem?
17:53:07 GMT <ahfeel> well, we have a high traffic website and adding 1/2ms on each request would be quite a pity
17:53:44 GMT <minus> considered using a connection pool?
17:53:47 GMT <ahfeel> (network round trip to connect to sentinel, then ask for master, then connect to master, this can possibly be 2ms)
17:54:01 GMT <ahfeel> Well in PHP it's not that easy unfortunately -_-
17:54:21 GMT <minus> stop using PHP :P
17:54:29 GMT <ahfeel> yeah, well, not an option :D
17:54:34 GMT <ahfeel> That's the only drawback tbh :<
17:54:50 GMT <minus> is redis on a different (set of) machines?
17:55:02 GMT <ahfeel> yeah
17:55:42 GMT <minus> you set up a local HAProxy instance on the app server(s) and connect to that. HAProxy could take care of connecting to the master and keep a pool of connections to it open
17:55:42 GMT <ahfeel> Well, it's currently on another node, but it's just a master and a slave, no sentinel, no reading from the slave
17:55:59 GMT <ahfeel> Well, the problem then is the failover
17:56:03 GMT <minus> however adding HAProxy does add a delay on failover i think
17:56:20 GMT <ahfeel> I would have to "specialize" HAProxy to make it "sentinel master" aware
17:56:31 GMT <minus> otoh a fail will manifest in the loss of connections anyway
17:56:33 GMT <ahfeel> it would be really awesome if sentinel nodes were just acting as proxies
17:56:57 GMT <minus> oh, you just let HAProxy connect to the redis instances directly and pick the one that's master
17:57:13 GMT <minus> maybe not as reliable as asking sentinels though
17:57:16 GMT <ahfeel> mhhh...
17:57:29 GMT <ahfeel> I just stumbled upon https://github.com/enriclluelles/redis-sentinel-proxy
17:57:52 GMT <ahfeel> That's the kind of stuff that makes sense, this should be integrated right into sentinel or redis itself
17:57:57 GMT <ahfeel> :P
17:58:12 GMT <minus> https://discuss.pivotal.io/hc/en-us/articles/205309388-How-to-setup-HAProxy-and-Redis-Sentinel-for-automatic-failover-between-Redis-Master-and-Slave-servers
17:58:31 GMT <minus> we have it running like that
17:58:47 GMT <ahfeel> :o
17:58:53 GMT <ahfeel> You have that in production ?
17:59:01 GMT <minus> can't tell you how well that works though, since there's no IT
17:59:03 GMT <minus> yes
17:59:12 GMT <ahfeel> xD no IT
17:59:13 GMT <ahfeel> lol
17:59:16 GMT <ahfeel> big website ?
18:00:07 GMT <minus> not a website; we deal in apps
18:00:18 GMT <minus> but not much traffic
18:01:57 GMT <ahfeel> ok
18:03:29 GMT <ahfeel> seems there's a twemproxy plugin that can work with sentinel
18:03:43 GMT <ahfeel> Well, I'll have a look over all these solutions
18:03:46 GMT <ahfeel> Thanks for the hint !
20:47:29 GMT <dupondje> mmm, would be cool if redis supported SLAVEOF <socket>
20:47:59 GMT <minus> yes
20:51:00 GMT <dupondje> strange nobody added that yet :)