Hm. Okay okay. I was able to access the url so I asked. Thanks
alastairp
sure, I guess our webserver accepts requests to all domains
shivam-kapila
Got it
ruaok
yvanzo: I'm reading through the updated docs for musicbrainz-docker and I've got two questions:
1. how do I expose the port for PG only? Can I set an alternate port number other than 5432?
2. How do I tune PG? I need to set more shared_buffers.
yvanzo
ruaok: for 1. create a file local/compose/publish-db-port.yml (see compose/publishing-all-ports.yml for example) and run admin/configure add local/compose/publish-db-port.yml; docker-compose up -d
2. cannot be done without building your own image for db service, should be added.
ruaok
thanks for #1.
#2 could a drastic issue we need to address. stock PG is dog slow. :(
shared_buffers = 512MB
if we're recommending a machine of 16GB, should we set that to 4096MB as default?
yvanzo
Maybe but 16GB was mainly required for live indexing.
ruaok
well, we can argue over defaults and what-not, but we need a way to tweak the settings.
because a noob will not know that they need to tune DB and will get the impression that our shit is dog slow.
yvanzo
Working on a patch, should be quick.
ruaok
<3
no rush. I've fixed my setup.
(but I know how)
the admin/configure stuff is really nice too.
yvanzo
the idea is have a file like default/indexer.ini that could be appended to db configuration.
oh man. Mr_Monkey userscript that can pre-fill language twice on https://bookbrainz.org/work/create?author=23f21... pls (I need to add 28+ works for small short-stries and it means I have to type "norw" and select from a drop down 56 times )
ishaanshah[m]
If we remove artist_msid from group_by I think we can fix a part of LB-547
i'm not completely sure we want to do that calculation on names
different artists can have same names
ishaanshah[m]
Another part is duplicates because of different names
iliekcomputers
ideally the bug would be fixed by messybrainz
i'd not worry about it tbh.
ishaanshah[m]
Ok
For range queries we should compare the timestamp right
iliekcomputers
so if you look at hdfs
ishaanshah[m]
Like, WHERE listen.timestamp > min_ts AND listen.timestamp < max_ts
pristine__
ishaanshah[m]: that bug will be better handled once we have the mapping right. Artist with two different msids can have same mbid and it leads to weird results. The ideal way is to use mbid everywhere once we have the mapping :)
iliekcomputers
the strcuture of the data is data/year/month.parquet
where month.parquet contains that month's listens
for a month, you'd just load the month's data and query on that.
won't need a where
but for week, yes, the where would make sense.
ruaok
pristine__: I'm working on the aa relations right now. once that is built up, I'm going to do the same for the msb mapping
ishaanshah[m]
But that won't work for week
iliekcomputers
even for week
pristine__
ruaok: no hurry :)
iliekcomputers
you should load just the month's data and put the where on that.
ishaanshah[m]
Ohk
Now suppose I import a months data
and create a temporary view
Then I should first filter out the the weeks listens
pristine__
ruaok: I see that we want to send recordings to lemmy that means we don't need a tar.
ruaok
iliekcomputers: thoughts on where this code should live? it uses the MB database to calculate relationship info that will be used in LB recommendation stuff. I'm inclined to make a new top-level dir in listenbrainz-server called `relations` and stuff put the code there.