[07:18:30] <elukey>	 good morning :)
[09:05:09] <klausman>	 \o
[09:05:23] <klausman>	 Currently typing up my interview notes for last week
[09:24:28] <aiko>	 o/
[09:24:39] <aiko>	 me too!
[09:31:38] <elukey>	 :)
[11:53:12] <klausman>	 Oh and btw, Thursday is a public holiday here, and I'll take Friday off as well
[12:02:14] <chrisalbon>	 Morning
[12:03:24] <klausman>	 \o
[12:19:55] <chrisalbon>	 Back to early morning meetings for me
[13:47:56] <elukey>	 hello hello
[13:51:30] <elukey>	 klausman: o/ thanks for the review!
[13:51:35] <elukey>	 (https://gerrit.wikimedia.org/r/c/operations/puppet/+/793714)
[13:51:45] <elukey>	 I was unsure about the role's name etc.., but I guess we can change it as we go
[14:09:27] <klausman>	 re-+1'd
[14:09:52] <elukey>	 obrigado
[14:10:02] <elukey>	 I'll wait for Eric's suggestions etc.. before proceeding
[14:10:08] <klausman>	 ack
[14:10:11] <elukey>	 but the cluster should be ready to be bootstrapped in theory
[14:10:37] <elukey>	 and we could start with a simple keyspace for scores
[14:11:04] <elukey>	 one thing that may be nice is to save the score alongside with the model's version
[14:11:41] <elukey>	 it has pros/cons, but it is simple and reliable in my opinion
[14:11:46] <klausman>	 And expire old versions along a LRU scheme?
[14:12:13] <elukey>	 this may be difficult, I am not sure if there is any LRU algorithm in cassandra
[14:12:48] <elukey>	 maybe we could have a clean up job that periodically drop old things if not needed
[14:16:34] <chrisalbon>	 I think we'd want the version number. People will ask for it.
[14:18:20] <elukey>	 I think it may also be great to have previous score/version tuples 
[14:19:02] <elukey>	 so basically not considering it as a score cache, but a score datastore (maybe with a limited amount of history)
[14:19:56] <elukey>	 the amount of storage per node is not a lot (1.4T)
[14:20:10] <elukey>	 we have 4.2T / cluster basically
[14:20:18] <elukey>	 and things gets replicated three times
[14:20:29] <elukey>	 but it is easy to expand it
[14:20:42] <chrisalbon>	 you mean as an output?
[14:21:37] <elukey>	 mmm sorry what do you mean with output?
[14:21:59] <elukey>	 (for the replication I meant internally, Cassandra replicates by default)
[14:22:07] <chrisalbon>	 Ah got it. Now I understand
[14:22:32] <elukey>	 I mentioned the size of the disks because it may not be a lot for all our use cases, like online feature store and score datastore etc..
[14:23:07] <elukey>	 but for the moment I'd use the cluster only for the score cache/datastore, and I'd see how it goes
[14:23:26] <elukey>	 so we'll get some familiarity with cassandra 
[14:23:33] <chrisalbon>	 Yeah that sounds right
[14:34:54] <elukey>	 klausman: I have just pushed a file called 'machine-learning' to pwstore, when you have a moment can you tell me if you can read the secret?
[14:35:42] <elukey>	 it is for https://wikitech.wikimedia.org/wiki/ORES/Deployment#Update_revscoring_in_PyPI
[14:41:37] <klausman>	 sec
[14:43:12] <klausman>	 Yes, I can read (and decrypt) it
[14:44:17] <elukey>	 \o/
[14:44:19] <elukey>	 super
[14:44:34] <elukey>	 ok so documentation updated, now we should have all the info to publish revscoring versions etc..
[14:44:51] <elukey>	 (I know that everybody is super happy about it)
[14:45:03] <klausman>	 Nice!
[16:00:54] <elukey>	 going afk folks, have a nice evening :)