Moving AcousticBrainz to Hetzner

Hi, all. I worked on the recent migration of AcousticBrainz to the central Hetzner infrastructure that hosts all our other projects. It was a fun experience that I would like to share on this blog.

This was the first time I worked with a production database of this scale, and it was a real learning experience. It really felt like I had jumped off the deep end, but it was really fun!

For those that don’t know, AcousticBrainz is a music technology project which crowd sources acoustic information for music recordings and is a collaboration between the Music Technology Group at Universitat Pompeu Fabra and MetaBrainz. AcousticBrainz has already collected information about 3.7 million unique recordings and has individual submissions from users for over 11 million recordings.

All the data is stored in a single PostgreSQL database for now. The server that AcousticBrainz used to run on (we called it spike, after the Tom and Jerry character) had gotten old and started spitting out hard disk failure warnings, so we decided to move it to the central Hetzner infrastructure where other MetaBrainz projects are hosted.

We use Docker for all services running in Hetzner and it has worked pretty well for us so far. So the first task was creating a production Docker environment for AcousticBrainz. Consul is used to provide configuration values for the AcousticBrainz server which needed some new code and consul template files to be written. This is relatively simple stuff that did not take too long. We also have a repository to store all configuration values and scripts that need to be run on each of our servers. So I also wrote code to run the three different services that AcousticBrainz needs in different Docker containers.

After that, I started work on creating data dumps of the AcousticBrainz data. There was already some code that dumped the entire database into an lzma compressed file. However, it was old code that hadn’t been run in a long time and the database had gotten biiig since then. The way the code worked was that it dumped each table as a file into a directory and then added the entire directory at once into a tar file. However, this approach doesn’t work now, because the table that stores the low‐level JSON data that users submit to us has become too big to be stored in a single text file uncompressed. The lowlevel_json file has 11 million rows right now with each row containing a relatively large JSON document stored in a column of Postgres’ cool JSONB type. The table takes around 357 GB when stored inside Postgres and this balloons to much over the space we had on spike. So, I wrote some code that dumped 500,000 rows into a file and compressed it before dumping the next 500,000 rows.

The compressed AcousticBrainz data dump was around 169 GB in size which seemed reasonable. Then, I realized that the server we were planning to run the webserver on (called boingo, after Oingo Boingo) did not have enough storage space or computational power to hold and work with the database. This led to us getting a new shiny server called frank (after Frank Ocean!) which has a pretty big 7200 RPM hard disk and over a 100 GB of RAM. We also decided to upgrade to PostgreSQL 10 during the migration, which led me to creating a Docker image for PostgreSQL 10 that we could use in production.

After this, I imported the data into the empty Postgres server which worked pretty well. Everything seemed set for a small downtime for migration where we’d just create a small incremental data dump, move it to frank and import, bring spike down, bring the webserver up on frank and be done with it. The The steps were written up and we were ready to go.

Things started, I brought the site down on spike, created an incremental dump, imported it to frank. Everything worked. We decided to do a integrity check of the new database once before bringing the new site up. This is where the trouble started. The number of rows in one of the the tables was 10 million when it should have been around a 100M, yikes. We realized that there had been a bug in the original data dump code that
we’d written. It was a pretty small bug, the key we were using to dump the data was incorrect. One line fix. I thought that we need more tests for our data dumps code.

Well, at that point we decided to just go ahead and dump and import the table individually instead of stopping the whole process. The downtime was much longer than expected because of this, the table was pretty big, 100 million rows is no joke, it took pg_dump hours to dump it. Then, I dropped the table on frank and began an import of the dumped file. We had decided to not drop constraints before importing for sanity reasons, but that turned out to not be that good of an idea. It took the import 5–6 hours before it was even halfway done and the time to import new rows was increasing. We gave up, stopped the import and dropped all constraints before starting a new clean import. This worked much much faster and was done in around an hour. At that point, we did another sanity check of the database, before bringing the site back up.

Some static files like binaries and old dumps we linked to were still hosted on spike (another thing I missed!), so I had to whip up a quick pull request changing links temporarily. I was doing this at 3 in the morning and I had started working on this the previous day 11 in the morning. It was the longest, most intense production deployment I have ever done. Pretty fun—now that I think about it—but I was tired then.

Later, I set up an FTP server on frank and moved the static files we were hosting there.

There were a lot of things that I learned in this entire process. First thing was that we should really sanity check literally everything before bringing any production service down. Second thing was that importing data with constraints in a database (especially for large amounts of data) is not very feasible. Third is that this level of control is not something that I would ever get as a new grad in any big company. Being thrown off the deep end here at MetaBrainz was really awesome. Another thing that I forgot to mention was that the entire migration process was done remotely over IRC with me sitting in college in Hamirpur, India and my teammates in Barcelona. This really teaches efficient communication and teamwork.

In hindsight, there are a few things that I’d do differently given the chance again. I’d definitely have sanity checked the imported database before actually going through with the downtime. It would have saved a lot of pain and the downtime would have been much lower. This is the biggest thing I learned from the migration process. Sanity check as often as possible.

All in all, working with production grade big data projects has been pretty awesome, and I hope I continue to learn as much as possible as early as possible.

AcousticBrainz at the 2018 MetaBrainz Summit

We had an in-person meeting at the MTG during the MetaBrainz summit to discuss the status and future of AcousticBrainz. We came up with a rough outline of things that we want to work on over the next year or so. This is a small list of tasks that we think will have a good impact on the image of AcousticBrainz and encourage people to use our data more.

State of AcousticBrainz

AcousticBrainz has a huge database of submissions (over 10 million now, thanks everyone!), but we are currently not using the wealth of data to our advantage. For the last year we’ve not had a core developer from MetaBrainz or MTG working on existing or new features in AcousticBrainz. However, we now have:

  • Param, who is including AcousticBrainz in his role with MetaBrainz
  • Rashi, who worked on AcousticBrainz for GSoC and is going to continue working with us
  • Philip, who is starting a PhD at MTG, focused on some of the algorithms/data going into AcousticBrainz
  • Alastair, who now has more time to put towards management of the project

Because of this, we’re glad to present an outline of our next tasks for AcousticBrainz:

Short-term

Some small tasks that are quick to finish and we can use to show off uses of the data in AcousticBrainz

Merge Philip’s similarity, including an API endpoint

Philip’s masters thesis project from last year uses PostgreSQL search to find acoustically similar recordings to a target recording. This uses the features in AcousticBrainz. We need to ensure that PostgreSQL can handle the scale of data that we have.

An extension of this work is to use the similarity to allow us to remove bad duplicate submissions (we can take all recordings with the same MBID and see if they are similar to each other, if one is not similar we can assume that it’s not actually the same as the other duplicates, and mark it as bad). We want to make these results available via an API too, so that others can check this information as well.

Merge Existing PRs

We have many great PRs from various people which Alastair didn’t merge over the last year. We’re going to spend some time getting these patches merged to show that we’re open to contributions!

Publish our Existing models

In research at MTG we’ve come up with a few more detailed genre models based on tag/genre data that we’ve collected from a number of sources. We believe that these models can be more useful that the current genre models that we have. The AcousticBrainz infrastructure supports adding new models easily, so we should spend some time integrating these. There are a few tasks that need to be done to make sure that these work

  • Ensure that high-level dumps will dump this new data (If we have an existing high-level dump we need to make a new one including the new data)
  • Ensure that we compute high-level data for all old submissions (we currently don’t have a system to go back and compute high-level data for old submissions with a new model, the high-level extractor has to be improved to support this)

Update/fix some pages

We have a number of issues reported about unclear text on some pages and grammar that we can improve. Especially important are

  • API description (we should remove the documentation from the main website and just have a link to the ReadTheDocs page)
  • Front page (Show off what we have in the project in more detail, instead of just a wall of text)
  • Data page (instead of just showing tables of data, try and work out a better way of presenting the information that we have)

Fix Picard plugin

When AB was down during our migration we were serving HTML from our API pages, which caused Picard to crash if the AB plugin was enabled while trying to get AB data. This should be an easy fix in the Picard plugin.

High Impact

These are tasks that we want to complete first, that we know will have a high impact on the quality of the data that we produce.

Frame-level data

We want to extract and store more detailed information about our recordings. This relies on working being done in MTG to develop a new extractor to allow us to get more detailed information. It will also give us other improvements to data that we have in AB that we know is bad. This data is much bigger than our current data when stored in JSON (hundreds of times larger), so we need to develop a more efficient way of storing submissions. This could involve storing the data in a well-known binary data exchange format. A bunch of subtasks for this project:

  • Finish the essentia extractor software
  • Decide on how to store items on the server (file format, store on disk instead of database)
  • Work out a way to deal with features from two versions of the extractor (do we keep accepting old data? What happens if someone requests data for a recording for which we have the old extractor data but not the new one?)
  • Upgrade clients to support this (Change to HTTPS, change to the new API URL structure, ensure that clients check before submission if they’re the latest version, work out how to compress data or perform a duplicate check before submission)
  • Deduplication (If we have much larger data files, don’t bother storing 200 copies for a single Beatles song if we find that we already have 5-10 submissions that are all the same)

MusicBrainz Metadata

Rashi’s GSoC project in 2018 helped us to replicate parts of the MusicBrainz database into AcousticBrainz. This allows us to do amazing things like keep up-to-date information about MBID redirects, and do search/browse/filtering of data based on relationships such as Artists just by making a simple database query. We want to merge this work and start using it.

Dumps

When we changed the database architecture of AcousticBrainz in 2015 we stopped making data dumps, making people rely on using the API to retrieve data. This is not scalable, and many people have asked for this data. We want to fix all of the outstanding issues that we’ve found in the current dumps system and start producing periodic dumps for people to download.

Build more models

In addition to the existing models that we’ve already built (see above, “Publish our Existing models”), we have been collecting a lot of metadata that we could use to make even more high-level models which we think will have a value in the community. Build these models and publicly release them, using our current machine learning framework.

Wishlist

These are tasks that we want to complete that will show off the data that we have in AcousticBrainz and allow us to do more things with the data, but should come after the high-impact tasks.

Expose AB data on MusicBrainz

As part of the process to cross-pollinate the brainz’s, we want to be able to show a small subset of AB data that we trust on the MB website. This could include information such as BPM, Key, and results from some of our high-level models.

Improve music playback

On the detail page for recordings we currently have a simple YouTube player which tries to find a recording by doing text search. We want to improve the reliability and functionality of this player to include other playback services and take advantage of metadata that we already have in the MusicBrainz database.

Scikit-learn models

The future of machine learning is moving towards deep learning, and our current high-level infrastructure written in the custom Gaia project by MTG is preventing us from integrating improved machine learning algorithms to the data that we have. We would like to rewrite the training/evaluation process using scikit-learn, which is a well known Python library for general machine learning tasks. This will make it easier for us to take advantage of improvements in machine learning, and also make our environment more approachable to people outside the MusicBrainz community.

Dataset editor improvements

Part of the high-level/machine learning process involves making datasets that can be used to train models. We have a basic tool for building datasets, however it is difficult to use for making large datasets. We should look into ways of making this tool more useful for people who want to contribute datasets to AcousticBrainz.

Search

With the integration of the MusicBrainz database into AcousticBrainz, we will be able to let people search for metadata related to items which we know only exist in AcousticBrainz. We think that this is a good way for people to explore the data, and also for people to make new datasets (see above). We also want to provide a way that lets people search for feature data in the database (e.g. “all recordings in the key of Am, between 100 and 110BPM”).

API updates

As part of the 2018 MetaBrainz summit we decided to unify the structure of the APIs, including root path and versioning. We should make AcousticBrainz follow this common plan, while also supporting clients who still access the current API.

We should become more in-line with the MetaBrainz policy of API access, including user-agent reporting, rate limiting, and API key use.

Request specific data

Many services who use the API only need a very small bit of information from a specific recording, and so it’s often not efficient to return the entire low-level or high-level JSON document. It would be nice for clients to be able to request a specific field(s) for a recording. This ties in with the “Expose AcousticBrainz data on MusicBrainz” task above.

Everything else

Fix all our bugs and make AcousticBrainz an amazing open tool for MIR research.


Thanks for reading! If you have any ideas or requests for us to work on next please leave a comment here or on the forums.

AcousticBrainz migration: We’re on!

We had to postpone the migration of AcousticBrainz last week since we ran out of time (our database is getting to be sizable!). We’ve migrated the bulk of our data and are now ready to move the last bits and call the move complete.

Downtime will start very soon — follow us on twitter for more detailed updates.

AcousticBrainz downtime: Migrating hosting to our other servers

Today we’re going to migrate the AcousticBrainz service from its standalone server that we’ve rented in the past few years to our shared infrastructure at Hetzner. We’ve been prepping for this move for a few weeks now and the actual process to follow has been used before, so we don’t expect the downtime to be more than 1 hour.

We’re sorry for the downtime that will be coming — to keep up with what we’re doing, please follow our progress on Twitter. We hope to start the migration in the next hour or two from when this blog post goes up.

 

GSoC 2018: More detailed integration of AcousticBrainz with MusicBrainz

Here comes an end to a fantastic summer for this year and time to wrap up my GSoC project which I have been working in for the last 3 months (the official GSoC coding period).

Hello people!!

I am Rashi Sah, an undergraduate student at the National Institute of Technology, Hamirpur, India. I have been working on a really cool AcousticBrainz project for MetaBrainz Foundation Inc. as a participant in Google Summer of Code ‘18. It has been an amazing experience and I’ve learned a lot over the summer, spending countless days and nights to successfully take the project to the stage of completion. I decided to contribute to MetaBrainz in late December, then spent some time understanding the codebase of the project and then began creating pull requests and pushing commits for many features, tasks and fixing bugs since January 2018. This blog post consists of my GSoC experience as a student and the work I’ve done for the program so far.

Before starting the GSoC program, I started looking for some good-first-bugs initially and found some tickets to work on. Then I talked to the AcousticBrainz community members and started contributing. I created some big PRs mostly for adding new features to AcousticBrainz. I also worked on many bug fixes which are already merged into the AcousticBrainz codebase. New feature additions PRs include AB-21, AB-98 and AB-298. In mid‐February, I started looking for a suitable idea to work on for GSoC program and to create a proposal for the same. As the month of March was approaching, I did a lot of proposal discussion with MetaBrainz community members especially with Alastair, AcousticBrainz project lead who has helped me a lot in reviewing and guiding me to improve my proposal to a better extent. Later April, my proposal for a more detailed integration of AcousticBrainz with MusicBrainz got accepted. In the community bonding period, I mostly tried to continue my work which I was already doing for the past 3–4 months.

Getting entity information from the MusicBrainz database

The first thing I worked on when the official GSoC coding period began was adding a way to directly access MusicBrainz database for different entities to the MusicBrainz database module in BrainzUtils (a Python utility for all of our MetaBrainz projects). I worked on getting artist and release entity information from the MusicBrainz database via a direct connection. (See PRs BU-13 and BU-14.) Later, I worked on setting up the MusicBrainz server by adding a service in AcousticBrainz’s docker-compose files allowing us to easily read data directly from the MusicBrainz database in AcousticBrainz (PR AB-334). Our major aim of the project was to implement both the methods of MusicBrainz database access in AcousticBrainz especially importing the MusicBrainz database in AcousticBrainz from scratch and then to decide which methods works better while implementing a particular functionality in AcousticBrainz using MusicBrainz data.

Import the MusicBrainz data in AcousticBrainz database

MusicBrainz’s database contains a huge number of tables, but I analysed the use case of MB data in AB and made a list of those tables that we would actually require in our AcousticBrainz integrations. Then I made a PR (AB-338) for creating new tables in the AB database under the MusicBrainz schema. Later, I worked on a big PR (AB-340) which imports MB data corresponding to each and every recording present in AcousticBrainz’s database and writes the data into the tables of the MusicBrainz schema in AB. This PR was really huge and I had to take care of a lot of integrity constraints and foreign key dependencies.

Update MB data in AB for every new recording added to AB

Another feature I worked on after importing the MB data was updating the MB data present in AB whenever any new recording is added to the AcousticBrainz database (see PR AB-346) by importing the data from MB’s database via the direct connection. While working on a few bug fixes, I and my mentor, Param realized that the MB data import is taking a lot more time than expected when I applied the MusicBrainz importer script for full MB data dumps (of around 2.8 GB). So, I then worked on making the MusicBrainz importer more efficient and was able to import the data for few recordings within seconds (see PR AB-348). I had to figure out a lot for each table import and to detect the parts of the code which were making things slower.

To reduce the load on the processor, I included a sleep schedule of 5 seconds in the MusicBrainz importer module to wait before importing data for any new recording (see PR AB-354). During my GSoC period, I learned how important it is to write tests and make them run fast. I wrote tests for almost every script inside the db module. Later, I worked on writing tests for the MusicBrainz importer script (AB-352).

Apply replication packets to keep MB data in AB updated with the actual MusicBrainz database

Then came another tricky part of this project which was to update the MusicBrainz schema data in AB whenever there is any change in the actual MusicBrainz database whether it is an update or a deletion taking place. MusicBrainz provides hourly replication packets which describe the changes to the database in a specific period. Replication packets are .tar.bz2 archives with a collection of files in them which can be downloaded via the MetaBrainz API. Lukas Lalinsky, a long-time contributor to MetaBrainz projects, the founder of AcoustID and maintainer of the mbdata Python module, had worked on implementing replication packets on MB data. I did a lot of modifications in his script to apply replication packets to the MusicBrainz schema data till it’s recent update for the recordings data present in AcousticBrainz (see AB-350).

Integration with MB database: Use MBID redirect information to get original entity

After working on the direct connection and importing the MusicBrainz data, keeping it updated by all means, it was time to start working on writing evaluation scripts to decide the better method for any integration we apply in AcousticBrainz. I wrote a script to implement an integration in AB with MB database to use the redirect information of an entity and then returns the original entity corresponding to the MBID provided (see PR AB-356).

Evaluate both methods of MusicBrainz database access in AcousticBrainz

Now moving towards the last work of my GSoC period and the most important as well. After working on both the methods, we really needed to evaluate both in order to test which one is more efficient for any specific integration with the MB database. I first wrote an evaluation script which fetches the data from the recording and low-level tables. For this case, the difference between the time taken by both methods comes out to be really large (approx. 70 seconds for around 250+ recordings). So whenever we would have to get the data from local AB tables and MB tables as well, we would go for the import database method as this method turns out to be faster than the other one. Next I tested with the MBID redirect integration part in which I didn’t find much difference between both the methods (PR AB-357). But I ran these tests locally, the tests in production may yield different results.

All in all, it has been an exciting summer. By this time I am familiar with a very good part of the AcousticBrainz codebase. I really look forward to work on adding a lot more integrations with MB data in AcousticBrainz and plan to completely remove AB’s dependency over the web service to use the MusicBrainz database which would be very useful for the users.

Details of contributions made

By the end of the GSoC coding period, I have opened a total of 39 PRs of which 35 are pull requests to the AcousticBrainz server, 3 are pull requests to BrainzUtils and 1 pull request to the AcousticBrainz client and have made a total of 135 commits (109 in AB, 9 in BU, 3 in AC and 14 in AB master) and out of them, pull requests created and merged during the official GSoC coding period are PRs to AcousticBrainz server and PRs to Brainzutils.

These last three months were full of thrill, excitement and much frustration as well. And this doesn’t end here, I’d love to contribute in the future and act as a maintainer for the AcousticBrainz project. I believe people must try to contribute to open source organizations as it helps you learn and gain much experience in a short period of time especially when working for a great platform like Google Summer of Code.

I am really happy working with the awesome MetaBrainz community and the people here are fantastic. I’d love to stay being a part of MetaBrainz in future as well. So in the end a big thanks to my mentor Param Singh, without his help & support throughout the program, wouldn’t have been possible for me to reach the end phase of GSoC, and my organization admin Robert Kaye, AcousticBrainz project lead Alastair Porter and all of the MetaBrainz Foundation community members for choosing me as a GSoC student and thus providing me such a great opportunity and also for being very kind and helpful throughout the program. And I want to thank Google for making this all possible. Hope I get a chance to work with you all again!!

Another AcousticBrainz update and a survey

Last year we started working on features to improve data produced from information about recordings that you submit to AcousticBrainz. First part of it was a way to create datasets that are used to train high-level models. The next were dataset creation challenges.

We already have a significant number of datasets created by AcousticBrainz community. The list of public datasets is available at https://beta.acousticbrainz.org/datasets/list. A couple of days ago our experimental challenge has concluded. It was related to classifying music with and without vocals. You can see final results at https://beta.acousticbrainz.org/challenges/14095b3b-4469-4e4d-984e-ef5f1a55962c.

Your feedback on high-level data

The latest addition to AcousticBrainz is a way to provide feedback about high-level output that you can see on summary pages for recordings. After a model is applied to all of the AcousticBrainz data we can understand how well it performs on a larger scale. This should help us make further improvements to models and underlying datasets. Keep in mind that you need to be logged in with your MusicBrainz account to see this.

Survey about new features

To help us understand how well new features work for you, we created a survey for you to participate in. If you have used AcousticBrainz, please fill out the survey here: https://goo.gl/forms/Oh3a9INBCCsW2I1i1. It shouldn’t take more than 5 minutes. We’ll keep it open for about a week.

Your feedback is very much appreciated. Especially considering that we don’t have a lot of ways to collect it from people. Some come to IRC and tell us about issues they are having, some comment on blog posts or create tickets in JIRA. But at this point we need a better overview of the current state of the project.

Thank you! 🎶

We’re actually really going to take the HTTPS plunge!

Closing in on three years after stating that “We’re going to take the HTTPS plunge!”, we’re actually really going to do it now. 🙂

Most of our sites have forced HTTPS for some time (metabrainz.org, critiquebrainz.org, bookbrainz.org, listenbrainz.org), but there are still a couple of stragglers, notably musicbrainz.org and acousticbrainz.org.

For MusicBrainz, our beta site is now all HTTPS, web service and all. The main, non-beta musicbrainz.org will be going HTTPS-only except for what’s under /ws/ (ie., the web service) to allow taggers and other programs not currently using HTTPS some transition time. We do not currently have an ETA for when we will make the final jump to HTTPS-only on the MusicBrainz web service, as that partly depends on feedback from our web service users, which leads me to:

If you’re currently using the MusicBrainz web service, please try and switch your program to using beta.musicbrainz.org and see whether your program breaks or not and let us know the status of it. We are aware that some Python versions and MusicBrainz libraries do not support our setup, so while your program may fail now, it might simply be because of dependencies of your program not being updated yet and you might not need to do anything specifically on your end – however, some programs/libraries might need some updates, so the more people test and report back, the better we’ll be able to judge when we can go all-HTTPS-only on musicbrainz.org.

For AcousticBrainz, we now have a shiny new Let’s Encrypt certificate on https://acousticbrainz.org thanks to our systems administrator Zas! As a result, we are going to start redirecting all HTTP traffic to HTTPS on the AcousticBrainz website, including API queries.

In order to give everyone time to verify that their scripts correctly recognise and validate our Let’s Encrypt certificate, we are going to delay the redirect until July 1, 2016. On this date, any HTTP query will automatically be redirected to HTTPS. We will also enable HSTS, so that compliant browsers will redirect to HTTPS on the client-side.

If you have any questions about either the MusicBrainz or the AcousticBrainz transition, please ask.