5s cache on /api/ only, i'll wait few minutes to see if it has any impact (unlikely), then i'll raise duration a bit
alastairp
ok
I'd be really interested to see what the histogram of inter-interval requests is
how frequently the most popular mbids are requested
zas
alastairp: what are /<uuid>/(low|high)-level queries vs /api/v1/<uuid> , website vs web service ?
alastairp
old-style API vs new API
zas
hmmm; and no redirect ?
alastairp
not yet, we just serve the same from both endpoints
we could start to deprecate it. add a redirect as a first point
zas
ok, first ones are not as easy to match (lack of a prefix)
alastairp
the thing is, as it's accessed by api clients and not browsers, they're almost certainly not going to cache redirects, even permanent ones
mm, right. so if we add a redirect then it's little work for the python server, but then makes caching easier
I'll open a ticket
zas
ok, for now, i'll cache more widely and for a longer time, if we want to measure any effect
Mr_Monkey: where are you atm ? Rob & I are heading out to find something to eat, are you in ?
aidanlw17
alastairp: hey, for the bulk high level get, I've been writing a similar single query to the low level one
But I have a question about raising exceptions for no data found
The load_high_level function https://github.com/metabrainz/acousticbrainz-se... raises no data found if the row is none or if there isn't any metadata, just a placeholder when the hl calculation failed
I was just wondering whether you think I should raise an exception if this is the case in the bulk get case, or if we should just skip the row altogether so that we can still return the high level data for the remaining (mbid, offset) pairs
ruaok
iliekcomputers: did you get near finishing the review for the ratelimiter?
Mr_Monkey
zas: having lunch at home in the end. Had some stuff to do in the house
alastairp
hi aidanlw17
for bulk get, we should raise no errors if there is no highlevel data for a given mbid
regardless of the error (doesn't exist, or only a hl row and no data rows)
like bulk get lowlevel, we just skip mbids if they don't exist (otherwise a user would have to make a request to see if each mbid exists before doing a bulk get... that removes all benefit from doing a bulk get in the first place
aidanlw17
Thanks alastairp. After thinking about it a bit more I changed it to skip like that. One other question, when we skip a row I was thinking that it might be beneficial to have a way of alerting the user that one of their recordings was not returned
alastairp
we decided to not do that - the idea is that they know what mbids they requested. If the id isn't in the response, it means it doesn't exist
you're right, we could have added a "missing" field, but we decided not to
aidanlw17
Okay I see, thanks for clarifying that
ferbncode
spellew: ping (how did your finals go? :) o/ )
Mr_Monkey
zas, ruaok: Is either one of you at Rob's? Can I come over?
zas
we are right now
and yes, i think you can come over, we just came back from lunch
which means that perhaps it should be fixed sooner rather than later
Mr_Monkey
reosarevok: ¯\_(ツ)_/¯
reosarevok
alastairp: do you remember how you decided for just /collection without a specified username on pymbngs?
alastairp
oh, I see that it's mentioned that the bindings use it. Interesting, why did we use the wrong endpoint?
Mr_Monkey
I suppose so
alastairp
reosarevok: yeah, good question
perhaps "I tried it and it worked, so I left it as-is"
reosarevok
I mean, maybe that was originally going to be the right endpoint, it was later changed, and that's why it's not documented?
Since Picard does the same
alastairp
You can also perform a get request on the collection list, which will present you an of collections for the currently authenticated user, with the amount of releases in each collection:
ZoeB: well, the genre one being missing I did wrong, so eh :D
ZoeB
So we're even? :D
Implementing genres was an impressive and wonderful change, thank you all! I've been looking forward to this for ages.
ruaok
you'd to have to screw up a lot more to be even with reosarevok, ZoeB.
yvanzo
zas: can we temporarily disable it for /ws/2/collection only?
zas: and /ws/2/collection/
zas
yes, i'll do that
yvanzo
Thanks, I did not find any code change possibly causing it either.
ZoeB
Yeah, there are the tag files, thank you!
yvanzo
np :)
zas
yvanzo: done
yvanzo
zas: thank you, asking for feedback on community
ZoeB has left the channel
Lotheric has quit
zas: just checked bug history, it’s been first reported on 12th May on #musicbrainz, that is the day before schema change release.
zas
first caching changes started on 11th, so it might be it. Since i changed minor things, that may have affect the behavior. I think I'll disable this stuff completly before re-introducing it more cleanly. It's rather interesting, but it needs to be tuned, per service.
