#metabrainz

ishaanshah[m]: you can add the build option back in, copying how it is in acousticbrainz-server's test.sh

alastairp: Sorry my client got disconnected

The test.sh and integration-test.sh work without having to be built if volume has been mounted

So it would be better to not have build step right?

As it slows down the process considerably

no - it's better to build. because at the moment, if I add a new dependency to requirements.txt there is no way to rebuild the image

my initial idea when I wrote the script was to have 2 modes: 1 which would build+setup+bring up db+test+take down everything, so that you could just run one command have have it do everything

and a second set of commands that let you build and bring up manually, so that as you say you can specify a specific test to run and not wait for the other steps

but this requires you to remember what steps you need to run based on what changes you make

I'm open to suggestions about a better way of structuring this command to make it better

alastairp: We can maybe add a -b which builds the containers only, Similar to frontend-tester.sh

Sorry for the late reply, matrix is acting wierd

ishaanshah[m]: yes, exactly. however, it should also build when running with no arguments

if there are 3 test scripts in listenbrainz I strongly suggest that you think about unifying them

ishaanshah[m]: I am doing the -b one. I will push in a while

alastairp: You suggest that test containers should be rebuilt every time we run test.sh?

Ok, maybe we could do -nb which will skip building

shivam-kapila: yes

So that we will be conscious about what we are doing

Yeah -nb makes sense

I don't think -nb is useful

do you understand what I said about the two modes of running tests?

Yes, I understood but was thinking about how to skip that part

Yeah the build and up and run for all tests

And the 2nd one for specific test run

Because waiting each time after a small change is a bit frustrating

Yes. For that I agree with -nb.

Mostly there is no change in requirements.txt

but if you follow the behaviour currently in test.sh you can already do this

$ ./test.sh -u

$ ./test.sh path/to/testfile.py # as many times as you need to

this will not build before running tests

Oh yes -u makes sense. By default if the db is up it means we can skip build

👍 exactly

the idea is to make a standalone call to test.sh do the right thing. and I believe that the right thing here is to build the image

Ok, makes sense

> $ ./test.sh path/to/testfile.py

Does AB have this currently?

yes

If will be good to be in sync with other projects IMO

I wondered how we could share a single test.sh file over all projects, but I don't know the best way to do this

I think that functionality has not been implemented in integration-test.sh yet

The build one?

ishaanshah[m]: ^

the -u one

Hmm... integration-test.sh doesnt have any options AFAIK

Its a fixed set of commands

yes, are you adding it?

Not currently. I am not sure about its need for integration tests.

alastairp: Can you offer an opinion for the same?

To skip the build part?

it depends on how quickly you need feedback on those test runs

if integration tests are _only_ run in CI, it probably doesn't matter

if you need to run a single integration test yourself many times, perhaps it's a good idea

Oh, I was adding a integration_test

ishaanshah[m]: Its not a skip actually. You will need to run -u always so that the build is done once

Yes, but doing it once is fine

keep in mind that -u in acousticbrainz and listenbrainz does _not_ build

it only starts the database server

but perhaps that's a good idea

But its good to rebuild to be consisent with everytime build thing

I didn't know that integration tests are run in CI only

I guess its not needed then

shivam-kapila: yeah, perhaps that's a good idea

ishaanshah[m]: Actually you should run them on local setup too

ishaanshah[m]: I didn't say that

I don't know how integration tests are used

ishaanshah[m]: To run tests in a file you can add the path in integration-test.sh

Yes, I did that, the only thing that was bugging me was the time to get feedback

Yes it buulds everytime

We can structure integration-tesh.sh to be similar to test.sh. I think we should wait for iliekcomputers' opinion too

Yes, I will just comment the build part in my local setup for now

Lol. Workarounds

Thanks for the -u tip alastairp. I didn't know that :D

I will add back the build command in test.sh then

ishaanshah[m]: Can I continue with that. I was already adding a build part to it.

ya sure

ishaanshah[m]: perhaps this should be added to some documentation, to make it clearer that that this functionality exists

thanks

-u is in docs alastairp

Here the flags are listed

iliekcomputers, alastairp: https://github.com/mayhem/timescale-testing/blo...

this is output from my timescale import program here: https://github.com/mayhem/timescale-testing/blo...

this is a subset of all listens from a feb dump that includes iliekcomputers, rob and zastai.

zastai's duplicates are not in there yet, but the last.fm fuzzy type dupes (off by ~1 sec) and straight up duplicates are handled

the output shows which listens where identified as duplicate, which one was chosen/rejected and the diff between the two listens.

the duplicate detection logic is here:

ah right - this is cases where there are 2 sources, where the only thing that differs is the timestamp is off by a second, so you choose only one?

s/2 sources/2 listens/

if could use a second set of eyes (or six) to sanity check this duplicate handling code.

alastairp: that is the fuzzy case, yes.

that's neat

but there are also other duplicates that we bodged into influx that we should clean up.

I am waiting for a new dump to take care of the latest causes for duplicate data.

but in the meantime, I'd appreciate looking at this to see if it makes sense.

there are some miniscule stats at the bottom of the HTML file.

and those bodges were the ones that someone was talking about a few days ago - exactly the same, but some additional flag in influx to be able to add them to stop a conflict?

Yeah Zastai was talking about it

alastairp: yes.

shivam-kapila: no, those are different.

those are the ones I am still waiting data on

Oh sorry then

np.

zastai's data had no duplicates as of feb 2020.

Freso: you also had a ton of dups in your stream, yes

?

anyone who has had dups and would like them cleaned up, should provide me and example ASAP.

ok, I'll have a look through the code. to confirm, the lookahead is you going forward in the listens a certain number of seconds in order to see if there are any dups?

did ollie have some?

ruaok: I think I might, but those also exist in Last.fm. Not sure if I have new/unique ones in LB.

or was his case that he was sending prod listens to alpha and wanted to keep them?

Not sure when or how old they are though.

alastairp: correct.

correct to the first question, re lookahead.

ruaok: cool, will look through it later this afternoon. goti t

alastairp: great thanks.

Freso: I'll include you in the next test run and then we'll see.

ollie as in acid2?

Alright. 👍

yes, acid2

k

I remember us doing something special for him, but I think it was just that we copied listens from alpha to beta, even when we said that we wouldn't

This might be reason for influx having mixed shards. For some users that doesnt allow to delete listens

the dump died for some reason

there's an error in sentry

:/

doh

btw, sentry 10 is out too. I have to upgrade it at the uni too

alastairp: shall we chat about the recommendation stuff for a minute?

sure, let me open it

what kind of chat do you want to have?

overall approach.

I read and disgested large chunks of the music recommendation chapter. a really nice summary, I must say.

I didn't pay a whole lot of attention to the context focused recommendations, since we have very little of that data.

I have two goals with all of this:

they were talking about rewriting it for a new edition of the book, covering some new tools, that will be good

1. Create a project that will open up people to come in an play with recommendations

2. Draw people in to improve the data sets that feed into the tools that are being created in order to create the recommendations. General outreach.

from the high level perspective if our goal was to get the best recommendations, we should be focusing on the collaborative filtering stuff.

so initial key deliverables are going to be 1) get data for us and others to use, 2) make some demos using that data, 3) enhance the data as needed as we see that we need more stuff

however, that is in a lot of ways, not low hanging fruit -- there is a lot of stuff that still needs to be done and pristine__'s moment has stalled and her contract is up soon.

right, I think getting the initial use-cases clear is an important first step

yes, more or less those deliverables are about right.

yes, and that might be a good goal for this convo:

that is, a general "you might want to listen to this" is quite different imo from "here's a playlist tailored to you based on some criteria that you specified"