[07:28:20] <zpapierski>	 ebernhardson: thanks for the suggestion, but that would require to set up flink along with python repl, right?
[07:28:31] <zpapierski>	 (that's probably not complicated, though)
[08:03:29] <gehel>	 dcausse: 1-on-1?
[08:03:36] <dcausse>	 oops
[09:01:50] <zpapierski>	 meal break
[09:48:37] <gehel>	 Lunch
[10:16:08] <dcausse>	 lunch
[12:40:26] <zpapierski>	 errand
[15:05:40] <ebernhardson>	 \o
[15:06:00] <ebernhardson>	 zpapierski: not really, i mean from the repl you can poke around the system
[15:06:18] <ebernhardson>	 zpapierski: you can look in directories, run 'subprocess.check_output(['du', '-sh', '/path/to/data'])
[15:06:20] <ebernhardson>	 etc.
[15:06:37] <dcausse>	 o/
[15:06:40] <ebernhardson>	 zpapierski: think of it like a really bad bash shell :)
[15:07:09] <zpapierski>	 or a really powerful one... with little to no regards for user comfort
[15:07:36] <ebernhardson>	 the right solution is ssh into the machines, but we can't do that. I guess i figured out some time ago if i can run code on the machine i can look around it, it's just more work :)
[15:09:36] <dcausse>	 you can ask yarn to run on a particular worker node?
[15:10:03] <ebernhardson>	 dcausse: you can check the hostname and try again :)
[15:10:07] <dcausse>	 :)
[15:10:46] <dcausse>	 ebernhardson: any objection if I deploy https://gerrit.wikimedia.org/r/c/wikimedia/discovery/analytics/+/708211 ?
[15:11:07] <ebernhardson>	 i have a thing somewhere that uses skein instead for same purpose, because you can directly control spinning up/down executors instead of roundabout with spark. But i forgot where...
[15:11:23] <ebernhardson>	 dcausse: yea should be good
[15:11:35] <dcausse>	 shipping, thanks
[15:25:36] <dcausse>	 ebernhardson: can't remember if a change in the "spark" folder requires an explicit "refresh" in airflow or is it automatic like the dag folder?
[15:29:47] <ebernhardson>	 dcausse: only the plugins folder requires a restart, the deciding factor is that airflow imports the plugin code and never triggers the code loading again
[15:30:07] <ebernhardson>	 (dags may kinda-sorta feel like they are imported, but they instead go through a custom load process outside python imports)
[15:32:09] <dcausse>	 ok thanks!
[16:50:43] <dcausse>	 I wonder what python "requests" does, it takes 30secs to initiate some connection while curl is immediate on some urls 
[16:51:40] <ebernhardson>	 hmm, indeed that doesn't sound right.  I would probably make a super short python script, then run `strace -e trace=network python scripy.py`
[16:52:03] <ebernhardson>	 it might only really tell you that it opened a socket and waited a bunch
[16:53:21] <dcausse>	 what strange is that it's only on some urls, for instance search.svc is ok but https://kubernetes1003.eqiad.wmnet:4007 is not
[16:55:24] <dcausse>	 it's waiting in connect
[16:55:25] <ebernhardson>	 curl is defaulting to ipv4, 
[16:55:43] <ebernhardson>	 requests is querying something with ipv6, not sure if its trying to use it yet
[16:56:03] <dcausse>	 ah maybe
[16:56:56] <ebernhardson>	 connect(3, {sa_family=AF_INET6, sin6_port=htons(4007), inet_pton(AF_INET6, "2620:0:861:103:10:64:32:23", &sin6_addr), sin6_flowinfo=htonl(0), sin6_scope_id=0}, 28^C) = ? ERESTARTSYS (To be restarted if SA_RESTART is set)
[16:57:28] <ebernhardson>	 yea, requests is failing to do ipv6. `curl -6 -k --head`  gives the same delay
[16:57:39] <dcausse>	 yeah seeing this as well
[16:57:48] <dcausse>	 while curl is doing connect(6, {sa_family=AF_INET, sin_port=htons(4007), sin_addr=inet_addr("10.64.0.247")}, 16) = -1 EINPROGRESS (Operation now in progress)
[16:57:50] <dcausse>	 ok
[17:06:54] <dcausse>	 used https://pythonadventures.wordpress.com/2019/06/28/force-requests-to-use-ipv4/ for now (somewhat hackish tho)
[17:07:19] <dcausse>	 will check with service-ops tomorrow might be known issue
[17:18:46] <ebernhardson>	 i suppose randomly useful info, almost every function strace reports has a man page in section 2.  `man 2 setsockopt` etc. I don't know what most of the strace output means, but can guess and read :) 
[17:24:32] <dcausse>	 personnaly I just post my question here and I wait for Erik to answer :P
[17:33:33] <ebernhardson>	 lol
[19:53:51] <ebernhardson>	 random english question re: "Rewrite not found top-level assets into the per-site docroot to support per-site favicon/etc.". On reading my patch before submitting i noticed lots of hypens and that i don't actually know where they go. Light searching suggests appropriate compound adjectives, maybe? Do those even go there?
[19:54:06] <ebernhardson>	 (doesn't actually matter, just curious :)
[20:28:15] <mpham>	 one step closer to the euthanasia roller coaster we talked about a while ago: https://mainichi.jp/english/articles/20210821/p2a/00m/0na/004000c
[21:48:18] <Trey314159>	 ebernhardson: random English answer—your intuitions are good! Hyphenating phrases that act as adjectives (i.e., modify other nouns) is good practice.
[21:49:04] <Trey314159>	 Let me know when you are ready to branch out into en dashes!! :)