14:00 PM
bitmap
killed the rest of the daily script, we can restart the subscriptions process once we're done
14:01 PM
alastairp
jmp_music: essentia shouldn't be needed.
14:01 PM
the current way that process works is that users download a binary which generates .json files
14:01 PM
bitmap
ruaok: yvanzo: ready to proceed?
14:01 PM
ruaok
ready. shall we do some tweeting?
14:01 PM
(ready to do that)
14:01 PM
bitmap
sure
14:02 PM
alastairp
the training process and doing data->model->prediction act only on these .json files
14:02 PM
jmp_music
So I 'll handle only the .json files (low level) and start the processing
14:02 PM
Ok that's clear too
14:02 PM
alastairp
we only have essentia in the webserver to build the `streaming_extractor_music_svm` tool, which does predictions
14:02 PM
with your tool, we will be able to remove essentia from the acousticbrainz-server repository too
14:03 PM
because we can replace it with sklearn inference
14:03 PM
ruaok
tweeted. let's give it a minute.
14:04 PM
are we going to take the sites down via the gateway?
14:04 PM
Mr_Monkey
Same. ready to take BB down
14:05 PM
jmp_music
Ok! Now I understand. I was a little bit confused because i was checking the gaia's code and it uses and other kinds of files (e.g. .sig)
14:05 PM
bitmap fixed an issue with musicbrainz_ro permissions on the genre table making MB ISE
14:05 PM
bitmap
ruaok: we could do that, git2consul seems to work now
14:05 PM
ruaok
up to you.
14:05 PM
I am ready in any case.
14:06 PM
Lotheric joined the channel
14:06 PM
Mr_Monkey
Seemed to work without it yesterday no?
14:06 PM
alastairp
right. gaia has existed since before json was a standard. .sig is just a yaml file
14:06 PM
ruaok
we took the sites down on monday.
14:06 PM
alastairp
internally, gaia converts json to yaml before loading it
14:06 PM
ruaok
bitmap: let us know when you want to start.
14:06 PM
Mr_Monkey
Right, the other yesterday.
14:06 PM
Lotheric__ has quit
14:06 PM
alastairp
(we won't have to do this)
14:06 PM
jmp_music
alastairp: OK. That's all for now! Thanks a lot for clearing up the mess in my head :) .
14:07 PM
bitmap
ruaok: ok, I've starting putting MB in read-only already
14:07 PM
alastairp
jmp_music: great, no problem. Feel free to ask these kinds of questions at any time
14:07 PM
ruaok
ok, I'll take CB, LB and MeB down.
14:07 PM
bitmap
seems to be live already
14:07 PM
Mr_Monkey
BB is down
14:07 PM
ruaok
LB down
14:08 PM
CB down
14:08 PM
jmp_music
alastairp: So you think it is a good idea to handle directly the json file and not transform it to yaml.
14:08 PM
bitmap
cool
14:08 PM
ruaok
MeB down.
14:08 PM
bitmap
does MSB need to come down?
14:08 PM
ruaok
is down
14:09 PM
sorry
14:09 PM
alastairp
jmp_music: yes, absolutely
14:09 PM
jmp_music
alastairp: just loading it to a Pandas dataframe and continuing the post-process there.
14:09 PM
bitmap
aight, putting pg into maintenance mode in consul
14:10 PM
I'll wait a min for that to propagate
14:11 PM
queries are dropping already
14:12 PM
ok let's see if it stops
14:12 PM
pg stopped
14:12 PM
restarting container
14:13 PM
bitmap checks ShmSize on container
14:13 PM
looks to be set
14:13 PM
yvanzo
docker container inspect foo | jq '.[0].HostConfig.ShmSize'
14:13 PM
ruaok
restart?
14:14 PM
bitmap
pg back up, pg bouncer too
14:14 PM
consul service out of maint mode
14:15 PM
ruaok
MeB up.
14:15 PM
CB up
14:15 PM
LB
14:15 PM
up
14:15 PM
bitmap
pushed DB_READ_ONLY -> 0 for MB, waiting for that to propagate
14:15 PM
Mr_Monkey
BB back up
14:16 PM
ruaok
MSB up.
14:16 PM
jmp_music has quit
14:17 PM
bitmap
looks good
14:17 PM
bitmap removes banner
14:19 PM
thanks ruaok yvanzo Mr_Monkey :)
14:19 PM
ruaok
np
14:19 PM
Mr_Monkey
š
14:20 PM
yvanzo
thank you!
14:20 PM
bitmap
I'll watch the logs for pgbouncer errors
14:24 PM
running ProcessSubscriptions again too to get the rest of the emails out
14:26 PM
I'll do the stats and reports too
14:27 PM
this might send out new emails for people who already got them though. should've stopped this before it started :\
14:27 PM
if they're subscriptions changed again in the past 30 mins
14:27 PM
their*
14:28 PM
yvanzo
better sending it two times than zero
14:28 PM
bitmap nods
14:31 PM
v6lur has quit
14:35 PM
bitmap
14:35 PM
Mr_Monkey
Nice.
14:51 PM
bitmap
14:51 PM
so 1GB is not enough to run some of these reports
14:52 PM
BrainzGit
14:52 PM
BrainzBot
14:55 PM
reosarevok
bitmap: how much did we give it before? I'm kind of amazed all these are failing now but not earlier :/
14:56 PM
bitmap
I think this is related to the new parallel query support in 9.6+, which use more shared memory
14:57 PM
it was just using the docker default before, 64M, but I guess it didn't need much in 9.5
14:58 PM
reosarevok
That's kind of insane
14:58 PM
So is the idea that it's running a lot more stuff in parallel and as such it should be done faster if it manages to run them at all?
15:00 PM
BrainzGit
15:00 PM
15:00 PM
BrainzBot
15:00 PM
bitmap
right
15:00 PM
it will try to execute separate parts of the plan for a single query in parallel
15:01 PM
reosarevok
But there's no way to tell it "hey, maybe don't run *this* in parallel" if we know it won't work?
15:02 PM
bitmap
not that I know of
15:02 PM
it's probably better if we just give it the memory it needs
15:02 PM
shivam-kapila
ruaok: Please take a look at the JSON structure I defined in this PR. The API will work on this standard. I have added a `submission_type` to validate the submission as type feedback. Please let me know if its not required.
15:03 PM
bitmap
these reports scan / join entire tables though so they are kinda extreme cases
15:03 PM
reosarevok
I mean, as long as we can give it enough, sure I guess :)
15:04 PM
Just worried at some point we'll get a query that would have worked fine (if slowly) earlier and doesn't work at all now
15:04 PM
BrainzGit
15:04 PM
BrainzBot
15:06 PM
BrainzGit
15:07 PM
yvanzo
15:09 PM
ruaok
shivam-kapila: the plural of `feedback` is `feedback`. stupid english.
15:09 PM
`feedbacks` is not a valid english word.
15:09 PM
shivam-kapila
Sorry š
15:10 PM
ruaok
#stopapologizing
15:10 PM
english is stupid.
15:10 PM
while of course
15:10 PM
#stopapologising is also valid
15:10 PM
#englishisreallystupid
15:11 PM
more seriously... given that we have a unique endpoint, I don't think we need to have the `payload` indirection.
15:11 PM
submission_type is also not needed, since it uses its own endpoint.
15:12 PM
iliekcomputers: do you have thoughts on the JSON for the feedback? in particular, I am not certain that POSTing a JSON document that could easily be done with two arguments is wise.
15:13 PM
POST to /api/feedback/user?msid=[]&score=-1 with an empty body would be easier to do for a client.
15:14 PM
well /api/feedback/user/dumbass?msid=[]&score=-1
15:16 PM
shivam-kapila
I felt so. Such a simple API can be done with POST request params
15:16 PM
iliekcomputers
ruaok: I think it'd probably be easier to give a json for post requests (specially with python `requests`), but i don't have any specific preferences either way
15:17 PM
ruaok
I'm not convinced either way, TBH.
15:17 PM
I do, however, feel that we don't need payload and subission type in fetching the feedback
15:17 PM
15:24 PM
shivam-kapila
Added payload just to be flexible in case we extend the API for bulk imports. submission_type will go for sure
15:24 PM
iliekcomputers
is the score variable?
15:24 PM
if not can't we just have two endpoints /like /dislike or something
15:24 PM
ruaok
score can be +1 or -1
15:26 PM
iliekcomputers
should we have a constraint on that in the db as well?
15:26 PM
ruaok
seems overkill, but sure.
15:27 PM
shivam-kapila
I was thinking to check it API level to prevent unnecessary DB access
15:27 PM
ruaok
I think that is sufficient.
15:29 PM
shivam-kapila
> if not can't we just have two endpoints /like /dislike or something
15:29 PM
score an go with +1,-1,0. +1 for love, -1 for hate, 0 in cases when user unmarks the track as loved/hated.
15:29 PM
+1/-1 ==> store in DB
15:29 PM
0 ==> Remove entry from DB
15:29 PM
iliekcomputers
i would probably have built 3 endpoints for them, unless we're planning on extending them.
15:30 PM
ruaok
also seems overkill to me.
15:31 PM
do you feel strongly about having 3 endpoints?
15:32 PM
iliekcomputers
nah, not really
15:33 PM
i'd say keep the json though if it's a post request.
15:33 PM
ruaok
ok, fair.
15:33 PM
iliekcomputers
most request libraries support it well out of the gate.
15:33 PM
ruaok
one endpoint, POST json, ditch the payload/type.
15:34 PM
that ok?
15:34 PM
iliekcomputers
sounds good to me
15:34 PM
ruaok
shivam-kapila: you?
15:34 PM
shivam-kapila
Absolutely