12:28:09 GMT <styles> Hey everyone. I'm confused how to use redis for metrics. I want to do a RPM stat.
12:28:23 GMT <styles> I am trying to use a key and set an expiration, but it doesn't seem to ever expire
12:28:33 GMT <styles> rpush key, value
12:28:42 GMT <styles> expire key 1 minute
12:29:09 GMT <minus> do you keep adding new stuff?
12:29:47 GMT <minus> the expiry applies to the key itself, styles, not to the elements you push into it
12:30:15 GMT <styles> Ok, so a key being a list of items
12:30:19 GMT <styles> So I don't want al ist then, right?
12:30:25 GMT <styles> a list*
12:31:03 GMT <minus> redis is bad for time series, fyi
12:31:31 GMT <styles> I just want to use it for a few simple ones
12:31:41 GMT <styles> Nothing crazy. I've looked at OpenTDB and other things
12:31:52 GMT <styles> I don't want the data long term either, just for the last like 15 minutes type
12:31:52 GMT <minus> you could create a key per minute
12:32:07 GMT <styles> What if I wanted to do an average of the RPM?
12:32:28 GMT <minus> over the last 15 minutes?
12:32:54 GMT <styles> Yeah
12:33:56 GMT <minus> you could discreetize your data into keys, one per minute, and to get the avg over 15 minutes query the 'last' 15 keys and average their value
12:34:59 GMT <styles> minus, could I do like drone:rpm:hour:minute:second 387
12:35:02 GMT <minus> the averaging could be done client-side or with a lua script
12:35:16 GMT <styles> then do like range drone:rpm:hour 15 ?
12:35:24 GMT <minus> no
12:35:33 GMT <minus> well, as long as you know the key names you can
12:35:50 GMT <minus> wildcard-style matching doesn't work well with redis
12:35:55 GMT <styles> ah ok good to know
12:36:13 GMT <styles> so I should just make each one of those keys and query them though?
12:36:21 GMT <minus> KEYS/SCAN can do that, but they're O(n) with n being the number of all keys in the DB
12:36:54 GMT <minus> yes, you could, but it's not really optimal
12:37:41 GMT <minus> depending on the amount you throw at it the one second resolution could be a problem
12:37:53 GMT <styles> humm
12:37:58 GMT <minus> with one second resolution you need to do 900 queries to redis to get the avg
12:38:00 GMT <styles> This is actually a pain in the butt lol
12:38:59 GMT <minus> maybe you can pull a stunt with sorted sets and using timestamps as scores. but that requires one entry per "hit" pretty much
12:39:36 GMT <styles> What about doing a incrby on a key that's the time like rpm:hour:minute:second
12:39:43 GMT <styles> then set that whole key to expire after the total duration
12:40:02 GMT <styles> and if I want to look it up, like you said, just range over the keys and add them up
12:40:09 GMT <minus> yeah, that's what we were talking about before. works but 900 queries for an avg
12:40:29 GMT <styles> kk
12:40:32 GMT <styles> I'll test that now
12:40:36 GMT <styles> thanks minus
12:40:42 GMT <minus> i mean it's totally not a problem if you only have a handful of those stats and query them every few seconds
12:41:40 GMT <styles> Can you "search" for keys?
12:43:06 GMT <styles> ah scan ok
12:44:49 GMT <styles> it's working, thanks
16:27:30 GMT <rob_> hi
16:27:35 GMT <rob_> how do people monitor redis?
16:43:45 GMT <minus> not at all here. the application does implicitly
16:44:30 GMT <rob_> how do you tell if there's a problem?
16:45:05 GMT <minus> when the application fails of course
16:45:15 GMT <rob_> minus: :o
16:45:28 GMT <rob_> i feel you arent taking this very seriously
16:45:30 GMT <minus> 100% dependent on redis here
16:46:29 GMT <minus> well, HAProxy executes PINGs and sentinel does failovering
16:46:34 GMT <eugecm> question. According to the documentation the SCAN command may return any given element multiple times. Does this happen under some concrete conditions or is there always a chance of getting a duplicate?
16:47:06 GMT <rob_> minus: how do you know if there has been a problem?
16:47:38 GMT <minus> i guess i don't, good point
17:33:42 GMT <eugecm> any ideas on my previous question with the SCAN command?
17:34:32 GMT <rob_> minus: sounds bad
17:34:48 GMT <minus> eugecm: assumption: when the key contents changed.
17:35:07 GMT <minus> rob_: kinda, yeah, but it's a managed service, so everything is fine
17:35:20 GMT <rob_> managed by who?
17:35:30 GMT <minus> our hoster
17:35:33 GMT <eugecm> minus: fair enough, I guess the function will have to be idempotent then
17:35:35 GMT <rob_> ok
17:35:42 GMT <rob_> do they monitor it for you then?
17:35:44 GMT <eugecm> minus: thanks
17:36:04 GMT <minus> rob_: i hope so, lol
17:36:24 GMT <rob_> :|
23:40:21 GMT <alex88> hello everyone, I'm new with lua scripts, if timestamps is a null key, what does redis.call("LPOP", "timestamps") returns?
23:40:36 GMT <alex88> nil? so I can check that return value with nil? like `ret == nil` in lua?
23:41:31 GMT <alex88> doing `local next_queue = redis.call("LPOP", "timestamps")` seems it never enters `if next_queue == nil then` even when it's empty
23:42:24 GMT <alex88> oh nvm, it returns false