[09:31:58] <dcausse>	 so you can alter a hive table but you need a ALTER TABLE tbl CHANGE col col new_schema CASCADE; the CASCADE part seems to only change the partitions metadata so it's hopefully "cheap"
[09:33:16] <dcausse>	 the thing I don't know is how to coordinate this ALTER with a running refinery and the schema update, is refine smart enough to insert to the new hive schema but with an old schema events?
[10:11:31] <dcausse>	 lunch
[13:17:22] <inflatador>	 <o/
[13:25:17] <dcausse>	 o/
[13:53:17] <ebernhardson>	 \o
[13:57:27] <inflatador>	 dcausse do you have anything for pairing? I don't have much, just trying to get rid of those icinga checks ATM
[13:58:07] <dcausse>	 inflatador: nothing on my side, let's skip
[13:58:09] <dcausse>	 o/
[13:58:29] <inflatador>	 ACK, works for me
[15:06:51] <Trey314159>	 inflatador: FYI, in case you hadn't noticed, next Monday is a US holiday
[15:09:11] <inflatador>	 Trey314159 thanks. Never take anything for granted with me ;P
[15:10:22] <Trey314159>	 I was surprised when I saw it so I'm getting the word out. (I mentioned it to Erik in a meeting)
[18:40:16] <ebernhardson>	 curious result,  taking all more_like logs for a day and groupby(qiprofile, source, ua group) we get no qiprofile, from api, with an `Other` ua issuing ~50% of more_like requests, and they have a hit rate of 3%. Second most popular is classic_noboostlinks via api and Chrome Mobile with a 65% hit rate
[18:41:58] <ebernhardson>	 oh wait, no sigh my aggregations missed a condition :P recalculating...
[18:53:59] <ebernhardson>	 ok better, the top groups by (qiprofile, source, ua) all have good hit rates, varying from 50-80%. The (none, api, other) group is 120k reqs and 8% hit rate. Something called ajabat website bot did 35k that day with 3% hit rate
[18:55:20] <ebernhardson>	 comes out as 5k-8k reqs/hour (~= 2qps) with no profile
[18:55:52] <ebernhardson>	 2qps seems like it shouldn't be enough to matter. classic_noboostlinks is seeing 1-1.2M req's for same hours