04:31:35 GMT <siddv29> Hi guys, i am new to redis DB. Could someone provide any tutorial link with deep dive in redis. I went through the redis.io tutorial. would love to understand redis' architecture in much more detail
10:01:09 GMT <mikecmpbll> can anyone help me figure out what might be going on here? https://gist.github.com/mikecmpbll/e081f92b862e794ca0332ee90a8b32a9
10:01:24 GMT <mikecmpbll> looks like they're all (the sentinels) trying to reconfigure eachother in a loop
10:02:34 GMT <mikecmpbll> one thing i notice is the two "known-slave" in the sentinel config are actually two IPs for the same server, which could be the cause. not sure how these are discovered though so don't know how to address the issue.
10:15:43 GMT <minus> Are you doing anything with proxies or so?
10:16:30 GMT <minus> Don't attach slaves through haproxy or so
10:16:45 GMT <badboy_> the sentinels seem to see different IPs for a single node
10:16:57 GMT <badboy_> ensure sentinels are all using the right IP to talk to other sentinels
10:17:05 GMT <badboy_> either bind to a fixed IP or change the advertised IP
10:18:43 GMT <mikecmpbll> nothing with proxies. there are two networks involved but all the servers are on both networks
10:19:57 GMT <mikecmpbll> how do the sentinels "see" an IP for a node
10:20:37 GMT <mikecmpbll> if nodes are resolving IPs differently, i'm not sure how to sort that
10:22:47 GMT <badboy_> Sentinel nodes try to detect the IP from the interface they are bound to and than advertise that
10:23:08 GMT <badboy_> and the other side sees a node connect with another IP
10:23:15 GMT <badboy_> and thus lead to both sides seeing different things
10:23:25 GMT <badboy_> so just pick one and stick with it ;)
10:51:45 GMT <mikecmpbll> "<@badboy_>	Sentinel nodes try to detect the IP from the interface they are bound to" — how does a sentinel get bound to interface? sorry if i'm being a moron just trying to figure this out.
10:53:32 GMT <mikecmpbll> oh, literally the bind directive in the config? i'm not using that at all at present.
11:07:31 GMT <mikecmpbll> so if i explicitly bind everything, sentinels and servers, to their interface on the same vlan ..
11:13:26 GMT <minus> how many sentinels do you have?
11:13:40 GMT <mikecmpbll> 3
11:13:49 GMT <minus> 2 in one DC, 1 in the other?
11:13:52 GMT <mikecmpbll> correct
11:13:59 GMT <minus> with a quorum of 2?
11:14:23 GMT <mikecmpbll> indeed. (yes i know, it doesn't make sense right now, i'm just getting it set up and i'll add more :D)
11:15:40 GMT <minus> i'm not sure it can ever make sense with sentinels in the DCs. the DC with fewer sentinels can never promote to master on netsplit
11:16:06 GMT <minus> well, maybe if you put at least one sentinel in a third location
11:17:31 GMT <mikecmpbll> hmmm.
11:18:23 GMT <mikecmpbll> can i have 4 sentinels and a quorum of 2? :D
11:18:36 GMT <minus> of course
11:18:44 GMT <minus> if you want 2 masters
11:18:47 GMT <minus> :D
11:20:23 GMT <mikecmpbll> why 2 masters? 😖
11:20:44 GMT <mikecmpbll> oh yeah nvm.
11:29:23 GMT <mikecmpbll>  what happens if >=quorum of sentinels become unavailable
11:29:47 GMT <mikecmpbll> no reconfiguration will ever happen?
11:30:26 GMT <minus> i'd assume so
11:31:02 GMT <badboy_> you can have 4 sentinels, but that won't buy you anything
11:31:11 GMT <badboy_> you will need 3 sentinels in contact to do a failover
11:31:46 GMT <badboy_> it's always (n/2)+1 for the actual failover, no matter what quorum you configured for down detection
11:32:23 GMT <mikecmpbll> i see.
11:32:23 GMT <mikecmpbll> so this isn't the right tool for the job, i'm assuming? i want datacentre independence like i have for all my other services.
11:35:47 GMT <badboy_> I don't know what you want to achieve^^
11:36:43 GMT <mikecmpbll> if one DC goes down my clients can failover to the redis slave at the other DC :D
11:38:42 GMT <badboy_> in that case you wil always need an external node to help with decisions
11:39:08 GMT <badboy_> also it's not clients doing a failover, it's the actual data nodes switching ;)
11:39:14 GMT <badboy_> (but I guess I know what you mean)
11:40:38 GMT <mikecmpbll> hmmm yeah. so i could have 2 at each DC and 1 at another place. therefore if either production DC goes down we got a majority of sentinels external to that to detect it
11:40:54 GMT <badboy_> right
11:41:03 GMT <badboy_> or just one per datacenter, but definitely one externally
11:41:46 GMT <mikecmpbll> ah yeah. that's good. i can understand that 👍🏼. just this IP issue to figure out now.
11:42:10 GMT <badboy_> yes, use "bind" to explicitely bind to one IP
11:42:33 GMT <mikecmpbll> righto, i'll stick binds in everywhere and see how i get on.
11:42:44 GMT <mikecmpbll> thanks for your help @badboy_ & minus :')
13:34:54 GMT <mikecmpbll> is there a way I can see the announce-ip from redis-cli? https://github.com/antirez/redis/blob/06263485d46696ba76a653d2b594f3493103c001/src/sentinel.c#L2478-L2484
13:44:49 GMT <badboy_> config get announce-ip
13:45:07 GMT <badboy_> oh, well, that will show the configured one
13:51:20 GMT <mikecmpbll> yeah, i'm interested to see how it is configured with bind address as 0.0.0.0
14:03:02 GMT <badboy_> SUBSCRIBE __sentinel__:hello on one of your Redis nodes
14:03:20 GMT <badboy_> then you will see what they announce
14:05:14 GMT <mikecmpbll> <3 thanks
14:13:01 GMT <mikecmpbll> this is what i'm trying to achieve .. https://goo.gl/obwLe0
14:15:12 GMT <badboy_> well, can sentinel 3 reach 10.1.24.* IPs?
14:15:22 GMT <mikecmpbll> no :(
14:18:06 GMT <badboy_> then this will never work
14:18:15 GMT <badboy_> Sentinels need to be able to talk to each other
14:21:11 GMT <mikecmpbll> well, they could talk to eachother, but on different interfaces.
14:25:26 GMT <mikecmpbll> maybe i can make them all announce public ips and do some NAT for ones on the same local network :/
14:38:39 GMT <minus> is there a library for sorted sets like they exist for python?
14:38:54 GMT <minus> is there a library for sorted sets like they exist in redis for python?
15:03:38 GMT <fanfan> whois soveran
18:20:19 GMT <stickperson> is there a way to blpop multiple items?
18:21:53 GMT <minus> no
18:37:48 GMT <shodan45> anyone using the predis php lib?
18:38:13 GMT <shodan45> trying to figure out transactions
18:48:42 GMT <shodan45> WATCH is like a lock, right?
18:50:47 GMT <minus> no, it doesn't lock anything
18:50:59 GMT <minus> well, it's passive
18:51:14 GMT <shodan45> I need to protect against a race condition - I need to DEL 2 things, but only if HGET firstthing foo is a certain value
18:51:16 GMT <minus> if any of the watched keys are touched the transaction aborts
18:51:37 GMT <minus> you can do that in a lua script
18:51:42 GMT <minus> those are atomic
18:51:47 GMT <shodan45> yeah... I was hoping to avoid that
18:51:51 GMT <minus> but you'll have to do the check in lua
18:52:32 GMT <minus> watch/multi/exec should work too though, as long as you just have one DEL command
18:52:37 GMT <minus> and no other modifications
18:52:43 GMT <shodan45> already have too many different languages :/
18:52:46 GMT <minus> but it might fail and you have to retry
18:53:14 GMT <shodan45> just 1 DEL? why is that?
18:54:06 GMT <minus> ah nvm
18:55:22 GMT <minus> should probably work
18:55:42 GMT <minus> the docs should tell you everything you need to know
18:56:10 GMT <shodan45> except that I have to "translate" the redis docs into php/predis :)
18:56:59 GMT <shodan45> hah. I just realized that the race condition can't happen.
18:57:02 GMT <shodan45> wheeeee
18:58:10 GMT <shodan45> it's impossible to go from "ok to delete" to "not ok to delete"